Category: From BI to AI

Posted on by

8 Keys to Managing the Linguistic Copycats that are Large Language Models

Over the past year, Generative AI has taken the world by storm as a variety of large language models (LLMs) appeared to solve a wide variety of challenges based on basic language prompts and questions.

A partial list of market-leading LLMs currently available include:

Amazon Titan
Anthropic Claude
Cohere
Databricks Dolly
Google Bard, based on PaLM2
IBM Watsonx
Meta Llama
OpenAI’s GPT

The biggest question regarding all of these models is simple: how to get the most value out of them. And most users fail because they are unused to the most basic concept of a large language model: they are designed to be linguistic copycats.

As Andrej Karpathy of OpenAI stated earlier this year,

"The hottest new programming language is English."

And we all laughed at the concept for being clever as we started using tools like ChatGPT, but most of us did not take this seriously. If English really is being used as a programming language, what does this mean for the prompts that we use to request content and formatting?

I think we haven’t fully thought out what it means for English to be a programming language either in terms of how to “prompt” or ask the model how to do things correctly or how to think about the assumptions that an LLM has as a massive block of text that is otherwise disconnected from the real world and lacks the sensory input or broad-based access to new data that can allow it to “know” current language trends.

Here are 8 core language-based concepts to keep in mind when using LLMs or considering the use of LLMs to support business processes, automation, and relevant insights.

1) Language and linguistics tools are the relationships that define the quality of output: grammar, semantics, semiotics, taxonomies, and rhetorical flourishes. There is a big difference between asking for “write 200 words on Shakespeare” vs. “elucidate 200 words on the value of Shakespeare as a playwright, as a poet, and as a philosopher based on the perspective on Edmund Malone and the English traditions associated with blank verse and iambic pentameter as a preamble to introducing the Shakespeare Theatre Association.”

I have been a critic of the quality that LLMs provide from an output perspective, most recently in my perspective “Instant Mediocrity: A Business Guide to ChatGPT in the Enterprise.” https://amalgaminsights.com/2023/06/06/instant-mediocrity-a-business-guide-to-chatgpt-in-the-enterprise/. But I readily acknowledge that the outputs one can get from LLMs will improve. Expert context will provide better results than prompts that lack subject matter knowledge

2) Linguistic copycats are limited by the rules of language that are defined within their model. Asking linguistic copycats to provide language formats or usage that are not commonly used online or in formal writing will be a challenge. Poetic structures or textual formats referenced must reside within the knowledge of the texts that the model has seen. However, since Wikipedia is a source for most of these LLMs, a contextual foundation exists to reference many frequently used frameworks.

3) Linguistic copycats are limited by the frequency of vocabulary usage that they are trained on. It is challenging to get an LLM to use expert-level vocabulary or jargon to answer prompts because the LLM will typically settle for the most commonly used language associated with a topic rather than elevated or specific terms.

This propensity to choose the most common language associated with a topic makes it difficult for LLM-based content to sound unique or have specific rhetorical flourishes without significant work from the prompt writer.

4) Take a deep breath and work on this. Linguistic copycats respond to the scope, tone, and role mentioned in a prompt. A recent study found that, across a variety of LLM’s, the prompt that provided the best answer for solving a math problem and providing instructions was not a straightforward request such as “Let’s think step by step,” but “Take a deep breath and work on this problem step-by-step.”

Using a language-based perspective, this makes sense. The explanations of mathematical problems that include some language about relaxing or not stressing would likely be designed to be more thorough and make sure the reader was not being left behind at any step. The language used in a prompt should represent the type of response that the user is seeking.

5) Linguistic copycats only respond to the prompt and the associated prompt engineering, custom instructions, and retrieval data that they can access. It is easy to get carried away with the rapid creation of text that LLM’s provide and mistake this for something resembling consciousness, but the response being created is a combination of grammatical logic and the computational ability to take billions of parameters into account across possibly a million or more different documents. This ability to access relationships across 500 or more gigabytes of information is where LLMs do truly have an advantage over human beings.

6) Linguistic robots can only respond based on their underlying attention mechanisms that define their autocompletion and content creation responses. In other words, linguistic robots make judgment calls on which words are more important to focus on in a sentence or question and use that as the base of the reply.

For instance, in the sentence “The cat, who happens to be blue, sits in my shoe,” linguistic robots will focus on the subject “cat” as the most important part of this sentence. The cat “happens to be,” implies that this isn’t the most important trait. The cat is blue. The cat sits. The cat is in my shoe. The words include an internal rhyme and are fairly nonsensical. And then the next stage of this process is to autocomplete a response based on the context provided in the prompt.

7) Linguistic robots are limited by a token limit for inputs and outputs. Typically, a token is about four characters while the average English content word is about 6.5 characters (https://core.ac.uk/download/pdf/82753461.pdf). So, when an LLM talks about supporting 2048 tokens, that can be seen as about 1260 words, or about four pages of text, for concepts that require a lot of content. In general, think of a page of content as being about 500 tokens and a minute of discussion typically being around 200 tokens when one is trying to judge how much content is either being created or entered into an LLM.

8) Every language is dynamic and evolves over time. LLMs that provide good results today may provide significantly better or worse results tomorrow simply because language usage has changed or because there are significant changes in the sentiment of a word. For instance, the English language word “trump” in 2015 has a variety of political relationships and emotional associations that are now standard to language usage in 2023. Be aware of these changes across languages and time periods in making requests, as seemingly innocuous and commonly used words can quickly gain new meanings that may not be obvious, especially to non-native speakers.

Conclusion

The most important takeaway of the now-famous Karpathy quote is to take it seriously not only in terms of using English as a programming language to access structures and conceptual frameworks, but also to understand that there are many varied nuances built into the usage of the English language. LLM’s often incorporate these nuances even if those nuances haven’t been directly built into models, simply based on the repetition of linguistic, rhetorical, and symbolic language usage associated with specific topics.

From a practical perspective, this means that the more context and expertise provided in asking an LLM for information and expected outputs, the better the answer that will typically be provided. As one writes prompts for LLMs and seek the best possible response, Amalgam Insights recommends providing the following details in any prompt:

Tone, role, and format: This should include a sentence that shows, by example, the type of tone you want. It should explain who you are or who you are writing for. And it should provide a form or structure for the output (essay, poem, set of instructions, etc…). For example, “OK, let’s go slow and figure this out. I’m a data analyst with a lot of experience in SQL, but very little understanding of Python. Walk me through this so that I can explain this to a third grader.”

Topic, output, and length: Most prompts start with the topic or only include the topic. But it is important to also include perspective on the size of the output. Example, “I would like a step by step description of how to extract specific sections from a text file into a separate file. Each instruction should be relatively short and comprehensible to someone without formal coding experience.”

Frameworks and concepts to incorporate: This can include any commonly known process or structure that is documented, such as an Eisenhower Diagram, Porter’s Five Forces, or the Overton Window. As a simpe example, one could ask, “In describing each step, compare each step to the creation of a pizza, wherever possible.”

Combining these three sections together into a prompt should provide a response that is encouraging, relatively easy to understand, and compares the code to creating a pizza.

In adapting business processes based on LLMs to make information more readily available for employees and other stakeholders, be aware of these biases, foibles, and characteristics associated with prompts as your company explores this novel user interface and user experience.

Posted on by — 1 Comment

Zoom Faces Challenges in Navigating the Age of Generative AI

Note: This piece was accurate as of the time it was written, but on August 11th, Zoom edited its Service Agreement to remove the most egregious claims around content ownership. Its current language is more focused on the limited license needed to deliver content and establishes that user content is owned by the user. Amalgam Insights considers the changes made as of August 11th to be more in-line both with industry standards and with enterprise compliance concerns.

On August 7, 2023, Zoom announced a change to its terms and conditions in response to language discovered in Zoom’s service agreement that gave Zoom nearly unlimited capability to collect data and an unlimited license to use this information going forward for any commercial use. In doing so, Zoom has brought up a variety of intellectual property and AI issues that are important for every software vendor, IT department, and software sourcing group to consider over the next 12-18 months.

Analyzing Zoom’s Service Agreement Language

This discovery seems to have been a few months in the making as these changes seem to have initially been made back in March 2023 as it was launching some AI capabilities. Looking at each section, we can see that 10.2 and 10.3 focus on the usage of data.

Although this data usage may seem aggressive at first, one has to understand that Zoom‘s primary function is video conferencing, which requires moving both video and audio data across multiple servers to get from one point to another. This requires Zoom to have broad permission to transfer all data involved in a standard video, conference, or webinar, which includes all the data being used and all of the service data created. So, in this case, Amalgam Insights believes this access to data is not such a big deal as Zoom probably needed to update this language simply to support even basic augments, such as cleaning up audio or improving visual quality with any sort of artificial or machine learning capabilities.

However, in Amalgam, insights perspective, 10.4 is of much more aggressive set of terms. This change provides Zoom with a broad-ranging commercial license to any data used on Zoom‘s platform. This means that your face, your voice, and any trade, secrets, patents, or trademarks used on Zoom now become commercially usable by Zoom. Whether this was the intention or not, this section both sounds aggressive and crosses the line on the treatment that companies expect for their own data.

This is an extremely aggressive stance by most intellectual property standards. And it stands out as conflicting in comparison, to how data is positioned by Microsoft and Salesforce, enterprise application platform companies that aren’t exactly considered innocent or naïve in terms of running a business.

What went wrong here? Zoom is traditionally known as a company that is for the most part end user-centric. Zoom’s mission includes the goal, to “improve the quality and effectiveness of communications. We deliver happiness.” And Eric Yuan’s early stories about wanting to speak with loved ones remotely and refusing to do on-site meetings in promoting the power of remote meetings are part of the Zoom legend.

However, Zoom is also facing the challenge of meeting institutional shareholder demands to increase stock value. When Zoom’s stock rose in the pandemic, it reached such amazing heights that it led to extreme pressure for Zoom to figure out how to 5X or 10X their company revenue quickly. Knowing that the stock was in a bit of a bubble, Zoom initially tried to purchase Five9, a top-notch cloud contact center solution, but ran into problems during the acquisition process as the stock prices of each company ended up being too volatile to come to an agreement on both the value and price of the stock involved.

And I speculate that at this point Zoom is focused on bringing its stock back up to pandemic heights, a bubble that may honestly never be reached again. For Zoom, 2020 was a dot-com-like event, where its valuation wildly exceeded its revenue. And as other video conferencing, and event software solutions ended up quickly improving their products, Zoom’s core conferencing capabilities started to be seen as a somewhat commoditized capability.

Following the mission of the company would have meant looking more deeply at communications-based processes, collaboration, transcription, and perhaps even emoji and social media enhancement: all of the ways that we communicate with each other. But, the problem is that there is really only one play right now that can quickly leads to a doubling or tripling of stock price and that is AI. There’s no doubt that the amount of video and audio that Zoom processes on a daily basis can train a massive language model, as well as other machine learning models focused on re-creating and enhancing video and audio.

Positioned in a way where it was understood that Zoom would enhance current communicative capabilities, it could’ve been a very positive announcement for Zoom to talk about new AI capabilities. Zoom has taken initial steps to integrate AI into Zoom with Meeting Summary and Team Chat Compose products. But given the limited capabilities of these products, the licensing language used in the service agreement seems excessive.

The language used in section 10 of Zoom’s service agreement is very clear about maintaining the right to license and commercialized all aspects of any data collected by Zoom. And that statement has not been modified. Whether this is because of an overactive lawyer or Zoom’s future ambitions, or promises made to a board or institutional investors is beyond my pay grade and visibility. But I do know that that phrase is obviously not user-friendly, and Zoom is not providing visibility to those changes at the administrative level. The language and buttons used to support zooms, a model and commercialization efforts are very different on the administrative page compared to the language used in the service agreement.

Image from Zoom’s August 7th blog post

Understanding that legal language can take time to change, it makes sense to wait a few days to see if Zoom reverts to prior language or further modifies section 10 to represent a more user-friendly and client-friendly promise. And I think this language reflects a couple of issues that go far beyond Zoom.

First, service agreements for software companies in general, are often treated as an exercise in providing companies with maximum flexibility, while taking away basic rights from end users. This is not just a product management issue; this is an industry issue where this language and behavior is considered status quo both in the technology industry and in the legal profession. When companies like Alphabet and Meta, previously Facebook, were able to get away with the level of data collection associated with supporting each free user without facing governance or compliance consequences in most of the world, that set a standard for tech companies’ corporate counsel. Honestly, the language used in Zoom‘s current service agreement as of August 7, 2023 is not out of scope for many companies in the consumer world that provide social technologies.

The second issue is the overwhelming pressure that exists to be first or early to market in AI. The remarkable success of ChatGPT and other open AI-related models has shown that there is demand for AI that is either interesting or useful and can be easily used and accessed by the typical user or customer. This demand is especially high for any company that has a significant amount of text, data, audio, or video. The recent March 2023 announcement of Bloomberg GPT is only the starting point of what will be a wide variety of custom language, models and machine learning models that come to market over the next 12 to 18 months. Zoom obviously wants to be part of that discussion, and there are other companies, such as Microsoft, Adobe and Alphabet as well as noted start-ups like OpenAI that have done amazing AI work with audio and video already. Part of the reason that this stands out is that Zoom is one of the first companies to change its policies and aggressively seek a permanent commercial license associated with all user content and forcing and opt-out process that lacks auditability or documentation regarding how users can trust that their data is no longer being used to train models or support any other commercial activities Zoom may wish to pursue. But Amalgam Insights is absolutely sure that Zoom will not be the last company to do this by any means. This language and the response should also serve as both a warning and a lesson to all other companies, seeking to significantly change their service agreements to support AI projects.

What is next for Zoom?

From Amalgam Insights’ perspective, there are three potential directions that Zoom can pursue going forward.

One, do nothing or make minimal changes to the current policy. Consumer and social media-based technology policies have set a precedent for the level of data and licensing access in Zoom’s service agreement, but this level of customer data usage is considered extreme in most business software agreements. Will Zoom end up being a test case for pushing the boundaries for business data use? This seems unlikely given that Zoom has not traditionally been considered an aggressive company in pushing customer norms. Zoom does try to move fast and scale fast, but Zoom’s mistakes have typically been more due to incomplete processes rather than acts of commission and intentionally trying to push boundaries.

Two, rewrite parts of Section 10 that are intrusive from a licensing and commercial usage perspective. Amalgam Insights hopes that this is an opportunity for Zoom to lead from an end user licensing or service agreement perspective in making agreements more transparent and in using more exact legal language that feels cooperative instead of coercive. The legal approach of including all possible scenarios may be considered professionally competent, but the business optics are antagonistic.

Three, come out with an explicit enterprise version of technology that is not managed under these current rules set in section 10 so that data is not explicitly used for models and cannot easily be turned on through a simple toggle switch in the administration console. As my friend and data management analyst extraordinaire Tony Baer stated on LinkedIn (where you should be following him) “The solution for Zoom is to be more explicit: an enterprise version where data, no matter how anonymized, is not shared for Generative AI or any other Zoom commercial purpose whatsoever, and maybe a more general and/or freemium edition (which is how many consumers have already been roped in) where Zoom can do its Gen AI thing.”

Recommendations

The first recommendation is actually aimed towards the CIO office, procurement office, and other software purchasers. Be aware that your software provider is going to pursue AI and will likely need to change terms and conditions associated with your account to do so. This is a challenge, as multinational enterprises now face the possibility of approaching or exceeding 1,000 apps and data sources under management and even businesses of 250 employees or less average one app per employee. There is a massive race towards aggregating data, building custom AI models, and commercializing the outputs as benchmarks, workflows, automation, and guidance. But Zoom is not a one-off situation and your organization isn’t going to escape the issues brought up in Zoom’s service agreement language just by moving to another provider. This is an endemic and market-wide challenge, far beyond what Zoom is experiencing.

The second recommendation: One solution to this problem may be for vendors to split their product into public consumer-facing products and private products from a EULA and terms and conditions perspective. This wouldn’t be the worst approach, and would maintain the consumer expectation of free services that are subsidized by data and access while giving businesses, the confidence that they are working with a solution that will protect their intellectual property from being accessed or recreated by a machine learning model. This also potentially allows for more transparency in legal language as this product split is considered. Tech lawyer Cathy Gellis, stated “There can be the lawyerly temptation to phrase them (terms of service) as broadly as possible to give you the most flexibility as you continue to develop your service. But the problem with trying to proactively obtain as many permissions as you can is that users may start to think you will actually use them and react to that possibility.” In 2023, software vendors should assume that corporate clients will be wary of any language that puts trade secrets, patents, trademarks, or personally identifiable information at risk. Any changes to terms of service or service agreements should be reviewed both from a buy-side and sell-side perspective. This may include bringing in procurement or specialized software purchasing teams to reflect the customer’s perspective.

The third recommendation goes back to the ethical AI work that Amalgam Insights did several years ago. AI must be conducted in context of the same culture and goals that are considered pervasive within the company. Any AI policy that goes significantly outside the culture, norms, and expectations of the company will stand out. And this can be a challenge, because AI has been treated as an experiment in many cases, rather than as a formalized, technical capability. As AI development and policy is shaped, this is a time when new products, governance, and documentation need to be tightly aligned to core business and mission principles. AI is a test of every company’s culture and purpose and this is a time when the corporate ability to execute on lofty qualitative ideals will be actively challenged.

Zoom’s misstep in aggressively pursuing rights and access to client data should not just be seen as a specific organizational misstep, but as part of a set of trends that are important for enterprise, IT, purchasing, and legal departments as well as all software and data source vendors seeking to pursue AI and further monetize deep digital assets. The next 12 to 18 months are going to be a wild time in the technology market as every software vendor pursues some sort of AI strategy, and there will be mountains of new legal language, technical capabilities, and compliance aspects to review.

Posted on by

Instant Mediocrity: a Business Guide to ChatGPT in the Enterprise

In June of 2023, we are firmly in the midst of the highest of hype levels for Generative AI as ChatGPT has taken over the combined hype of the Metaverse, cryptocurrency, and cloud computing. We now face a deluge of experts who all claim to be “prompt engineering” experts and can provide guidance on the AI tools that will make your life easier to live. At the same time, we are running into a cohort of technologists who warn that AI is only one step away from achieving full sentience and taking over the world as an apocalyptic force.

In light of this extreme set of hype drivers, the rest of us do face some genuine business concerns associated with generative AI. But our issues are not in worshipping our new robot overloads or in the next generation of “digital transformation” focused on the AI-driven evolution of our businesses that lay off half the staff. Rather, we face more prosaic concerns regarding how to actually use Generative AI in a business environment and take advantage of the productivity gains that are possible with ChatGPT and other AI tools.

Anybody who has used ChatGPT in their areas of expertise has quickly learned that ChatGPT has a lot of holes in its “knowledge” of a subject that prevent it from providing complete answers, timely answers, or productive outputs that can truly replace expert advice. Although Generative AI provides rapid answers with a response rate, cadence, and confidence that mimics human speech, it often is missing either the context or the facts to provide the level of feedback that a colleague would. Rather, what we get is “instant mediocrity,” an answer that matches what a mediocre colleague would provide if given a half-day, full-day, or week to reply. If you’re a writer, you will quickly notice that the essays and poems that ChatGPT writes are often structurally accurate, but lack the insight and skill needed to write a university-level assignment.

And the truth is that instant mediocrity is often a useful level of skill. If one is trying to answer a question that has one of three or four answers, a technology that is mediocre at that skill will probably give you the right answer. If you want to provide a standard answer for structuring a project or setting up a spreadsheet to support a process, a mediocre response is good enough. If you want to remember all of the standard marketing tools used in a business, a mediocre answer is just fine. As long as you don’t need inspired answers, mediocrity can provide a lot of value.

A few things for you to consider as your organization starts using ChatGPT. Just like when the iPhone launched 16 years ago, you don’t really have a choice on whether your company is using ChatGPT or not. All you can do is figure out how to manage and govern the use. Our recommendations typically take one of three major categories: Strategy, Productivity, and Cost. Given the relatively low price of ChatGPT both as a consumer-grade tool and as an API where current pricing is typically a fraction of the cost of doing similar tasks without AI, the focus here will be on strategy and productivity

Strategy – Every software company now has a ChatGPT roadmap. And even mid-sized companies typically have hundreds of apps under management. So, now there will 200, 500, or however many potential ways for employees to use ChatGPT over the next 12 months. Figure out how GPT is being integrated into the software and whether GPT is being directly used to process data or indirectly to help query, index, or augment data.

Strategy – Identify the value of mediocrity. The average large enterprise getting mediocrity from a query-writing or indexing perspective is often a much higher standard than the mediocrity of text autocompletion. Seek mediocrity in tasks where the average online standard is already higher than the average skill within your organization.

Strategy – How will you keep proprietary data out of Gen AI? – Most famously, Samsung recently had a scare when it saw how AI tools were echoing and using proprietary information. How are companies both ensuring that they have not put new proprietary data into generative AI tools for potential public use and that their existing proprietary data was not used to train generative AI models? This governance will require greater visibility from AI providers to provide detail on the data sources that were used to build and train the models we are using today.

Strategy – On a related note, how will you keep commercially used intellectual property from being used by Gen AI? Most intellectual property covered by copyright or patent does not allow for commercial reuse without some form of license. Do models need to figure out some way of licensing data that is used to train commercial models? Or can models verify that they have not used any copyrighted data? Even if users have relinquished copyright for the specific social networks and websites that they initially wrote for this does not automatically give OpenAI and other AI providers the same license to use the same data for training. And can AIs own copyright? Once large content providers such as music publishers, book publishers, and entertainment studios realize the extent to which their intellectual property is at risk with AI and somebody starts making millions with AI-enabled content that strongly resembles any existing IP, massive lawsuits will ensue. If an original provider, be ready to defend IP. If using AI, be wary of actively commercializing or claiming ownership of AI-enabled work for anything other than parody or stock work that can be easily replaced.

Productivity – Is code enterprise-grade: secure, compliant, and free of private corporate metadata? One of the most interesting new use cases for generative AI is the ability to create working code without having prior knowledge of the programming language. Although generative AI currently cannot create entire applications without significant developer engagement, it can quickly provide specifically defined snippets, functions, and calls that may have been a challenge to explicitly search for or to find on a Stack Overflow or in git libraries. As this use case continues to proliferate, coders need to understand their auto-generated code well enough to check for security issues, performance challenges, appropriate metadata and documentation, and reusability based on corporate service and workload management policies. But this will increasingly allow developers to shift from directly coding every line to editing and proofing the quality of code. In doing so, we may see a renaissance of cleaner, more optimized, and more reusable code for internal applications as the standard for code now becomes “instantly mediocre.”

Productivity – Quality, not Quantity. There are hundreds of AI-enabled tools out there to provide chat, search-based outputs, generative text and graphics, and other AI capabilities. Measure twice and cut once in choosing the tools that you use to help you. It’s better to find the five tools that matter than the 150 tools that don’t maximize the mediocre output that you receive.

Productivity – Are employees trained on fact-checking and proofing their Gen AI outputs? Whether employees are creating text, getting sample code, or prompting new graphics and video, the outputs need to be verified against a fact-based source to ensure that the generative AI has not “hallucinated” or autocompleted details that are incorrect. Generative AI seeks to provide the next best word or the next best pixel that is most associated with the prompt that it has been given, but there are no guarantees that this autocompletion will be factual just because it is related to the prompt at hand. Although there is a lot of work being done to make general models more factual, this is an area where enterprises will likely have to build their own, more personalized models over time that are industry, language, and culturally specific. Ideally, ChatGPT and other Generative AI tools are a learning and teaching experience, not just a quick cheat.

Productivity – How will Generative AI help accelerate your process and workflow automation? Currently, automation tends to be a rules-driven set of processes that lead to the execution of a specific action. But generative AI can do a mediocre job of translating intention into a set of directions or a set of tasks that need to be completed. While generative AI may get the order of actions wrong or make other text-based errors that need to be fact-checked by a human, the AI can accelerate the initial discovery and staging of steps needed to complete business processes. Over time, this natural language generation-based approach to process mapping is going to become the standard starting point for process automation. Process automation engineers, workflow engineers, and process miners will all need to learn how prompts can be optimized to quickly define processes.

Cost – What will you need to do to build your own AI models? Although the majority of ChatGPT announcements focus on some level of integration between an existing platform or application and some form of GPT or other generative AI tool, there are exceptions. BloombergGPT provides a model based on all of the financial data that it has collected to help support financial research efforts. Both Stanford University Alpaca and Databricks Dolly have provided tools for building custom large language models. At some point, businesses are going to want to use their own proprietary documents, data, jargon, and processes to build their own custom AI assistants and models. When it comes time for businesses to build their own billion-parameter, billion-word models, will they be ready with the combination of metadata definitions, comprehensive data lake, role definitions, compute and storage resources, and data science operationalization capabilities to support these custom models? And how will companies justify the model creation cost compared to using existing models? Amalgam Insights has some ideas that we’ll share in a future blog. But for now, let’s just say that the real challenge here is not in defining better results, but in making the right data investments now that will empower the organization to move forward and take the steps that thought leaders like Bloomberg have already pursued in digitizing and algorithmically opening up their data with generative AI.

Although we somewhat jokingly call ChatGPT “instant mediocrity,” this phrase should be taken seriously both in acknowledging the cadence and response quality that is created. Mediocrity can actually be a high level of performance by some standards. Getting to a response that an average 1x’er employee can provide immediately is valuable as long as it is seen for what it is rather than unnecessarily glorified or exaggerated. Treat it as an intern or student-level output that requires professional review rather than an independent assistant and it will greatly improve your professional output. Treat it as an expert and you may end up needing legal assistance. (But maybe not from this lawyer. )

Posted on by

Hybrid Workforce Management: Navigating the Complexities of a Diverse Workforce in the Modern Era

Businesses face the challenge of managing a variety of workforce challenges across the wide variety of people at the company: freelancers, outsourcing firms, consultants, contingent labor, and full-time labor. The United States Government Accountability Office estimates that about 40% of workers are not full-time workers and fall within a variety of roles including contractors, part-time workers, and on-call workers that may track time through different systems and methods. With this challenge in mind, companies need a comprehensive management view that automates processes and helps companies focus on conducting work more quickly rather than be mired in a sea of paperwork and processes. Workforce management is no longer simply a matter of managing active full-time employees, but supporting a comprehensive practice that consolidates workforce management across contingent, part-time, full-time, and other categories of workers.

To effectively manage their hybrid workforce effectively across financial, operational, and management capacities, companies must consolidate workforce management tasks onto a single platform and a consistent set of data to avoid constant switching back and forth across inconsistent data. This platform should include contingent labor, internal labor, time and payroll, workforce scheduling, financial budgeting, employee engagement, and onboarding capabilities, including governance, risk, and compliance management across all areas. Data across all of these areas should ideally be within a single data store that provides a shared version of the truth for all stakeholders in workforce management across HR, finance, and line-of-business management roles.

Amalgam Insights believes the following capabilities should be considered in developing a comprehensive workforce management system.

Manage payroll, performance, and relevant benefits for employees, consultants, and freelancers.

Workforce management efforts must consider the combination of standard payroll systems, time and attendance systems, scheduling systems, contingent labor management, on-demand services, third-party temporary labor and consulting firms, and self-employed contractors. In doing so, companies must decide which benefits and services are consistent across various labor types and what resources are needed to maximize the productivity of each class of workers. Regardless of labor type, compensation must be timely, accurate, and provided based on contractual agreements based on relevant labor law. By managing all classes of workers across a shared and consistent set of characteristics, companies may be better positioned to see if there are part-time or contingent workers who should be made full-time employees or to see which tasks are better supported by specific workers, skillsets, geographies, shifts, and other identifying work characteristics.

Supporting differing compliance requirements based on geography, status, and corporate asset access.

Workers with privileged access to trade secrets or classified information must all be treated with relevant compliance and confidentiality standards regardless of their work categorization. At the same time, companies must manage differing standards across wages, benefits, and tax obligations that exist in each jurisdiction where a worker is located.

Standardizing Key Performance Indicators (KPIs) and Management by Objectives (MBOs) across different work categories by focusing on the quality and quantity of relevant outputs and deliverables.

Even within a single department, the combination of roles, geographies, seniority, and employee status can lead to widely disparate individual goals. As companies identify appropriate KPIs and MBOs on an individual level that maximizes the value that each person brings to the workplace, they must also ensure that teams are aligned to shared corporate success metrics rather than disparate and disconnected metrics that may inadvertently pit workers against each other to pursue personal success.

Using a feedback-based set of processes to create a consistent employee experience and corporate culture that provides all workers with a shared set of expectations, goals, worker preferences, and employee support.

Employee feedback is only as useful as the corpus of data created and the management response associated with the suggestions and criticisms provided. At the same time, feedback can also be part of a continuous learning and continuous improvement initiative if feedback is stored as analytic and decision-guiding data that is tracked and monitored over time. Feedback can also be analyzed to see if workers are engaged in processes that are designed to improve the worker or corporate experience.

Understand the top-line and bottom-line financial contribution of contract and contingent work.

Although revenue per full-time employee is an outward-facing metric used by public companies to show efficiency, the business reality is that contractors and part-time employees also represent investments that should be reflected in workforce costs in determining corporate productivity and profitability. If companies are effectively replacing skills with contingent labor, this should be noted and tracked. Conversely, if there are significant gaps between full-time and other employees, companies should figure out the cause of these gaps and whether they can be closed through training, onboarding, or technical augmentation.

Taking Steps to Create a Consolidated Workforce Management Environment

Ultimately, companies have a responsibility to support the relevant stakeholders and shareholders associated with the company. However, this responsibility cannot be met if the company lacks consistent visibility to every worker who is attached to corporate work output, regardless of employment status, geography, department, or role. As companies seek to improve productivity and to allow executives to be more strategic in their approach to support productive workers while maintaining all relevant compliance responsibilities and a shared version of all relevant data, Amalgam Insights provides the following recommendations for human resources, finance, and managerial roles tasked with creating a better work environment.

First, ensure that you have the data necessary to maintain consistency of work expectations. Workers should be able to expect some baseline of employee experience even as they differ in location, employment status, and compensation if for no other reason than to provide every worker with a standard set of expectations and professional responsibilities.

Second, measure the profitability and revenue across the entire workforce based on a holistic view of hours, skills, geographies, and business goals. This capability can potentially uncover if specific hiring or labor sourcing strategies may be more profitable, or at least aligned to higher revenue, rather than simply treating all hiring and contracting exercises as an exercise in managing costs.

Finally, manage contingent labor with metrics and standards similar to traditional employee labor. When 40% of labor consists of either part-time, contractors, or on-demand workers, a workforce management solution that only looks at full-time payroll, onboarding, time, attendance, and benefits is no longer sufficient to understand the finance and operational details of the holistic workforce. Frontline and hourly workers seeking to manage their scheduling and time need a consistent and mobile experience on par with full-time workers. Regardless of how these metrics are presented from a public perspective, companies must have an internal basis for tracking the skills and work of every person who conducts work for a company, regardless of formal employment status.

By taking these steps, companies can fully empower all workers to acknowledge their contributions, manage skill portfolios, and further invest in the success of the complete workforce.

Posted on by — 1 Comment

Why Are Spreadsheets Still A Common FP&A Tool?

“The status quo is not a neutral state, but a mindset to uphold the decisions of the past.”

In 2023, effective business planning, budgeting, and forecasting is a necessary capability to keep organizations running. Already this year, we have seen unexpected banking failures, unpredictable labor markets, and continued supply chain and logistics challenges based on geopolitical challenges. In light of these challenges, Amalgam Insights believes that businesses must have a shared version of the truth that they use as they look at their budget and finances.

And, in this case, we specifically talk about a “shared” version of the truth rather than the “single version of the truth” typically associated with data warehouses and enterprise applications. This is because data changes quickly and every stakeholder can potentially make different decisions to define and augment their data, even basic changes such as language and currency translation that can lead to different versions of the truth. In this analytically enhanced and globally complicated world, it makes more sense to have a shared version of the truth that is augmented with personalized or localized data and assumptions. However, this consistently shared version of the truth can be hard to accomplish in organizations where planning is handled in a distributed and personalized manner through spreadsheets. In the enterprise world, finance professionals are inured to the basic realities of auditable data, processes, and results. And they are often asked to provide reports and memos that are used at the executive level or by external investors and public markets to ascertain the health of the market. Given the assumed importance of this formality, why would experienced professionals use spreadsheets in the first place?

Let’s face it; spreadsheets are easy to use. They are the lingua franca of data; a format that every experienced data user has been trained on. And with plug-ins and Visual Basic, spreadsheets can now handle relatively complex analytic use cases. Even if they aren’t quite data science tools, spreadsheets can provide structured analytic outputs. Also, spreadsheets are accessible on every computer through Excel, Google Sheets, or other common spreadsheet software. And with the emergence of cloud-based spreadsheets, it is now possible for two or more people to collaborate within a single spreadsheet.

Spreadsheets also provide users with the ability to customize their own analytic views with their own personalized views of data and the ability to hypothesize by building their own models. Who hasn’t looked at data and wondered “what if the data looked a bit differently?” or “what if we have a drastic scenario that suddenly increases or decreases a fundamental aspect of the business?” In light of COVID, rapid interest rate hikes, global shortages in commodities production, trained labor shortages, and the increasingly unstable banking environment we are in, it is important to be able to test potential extreme assumptions and support a wide variety of scenarios. Between the ease of use, availability, and personalization aspects of spreadsheets, it is not hard to figure out why spreadsheets are still a leading tool for financial planning and analysis. Even so, Amalgam Insights has found that once organizations pass Dunbar’s number (approximately 150 employees), they start to struggle with collaborative tasks simply because it becomes difficult for any one employee to know all of the other employees who need to be involved in the business planning process and spreadsheets have been designed to maximize individual productivity, rather than collaborative work, for decades. From a practical perspective, people tend to work with the people they know best. This is fine for a small company with a dedicated office where everyone knows each other. According to US Census data, the typical 1,000-person company has 19 locations, making it highly unlikely that all of the key budget stakeholders will be in one office. In this regard, Amalgam Insights finds the following challenges in supporting spreadsheet-based planning at scale.

The distributed nature of work also makes spreadsheet governance a challenge, as it is easy for spreadsheets to suffer from version control issues, a structured workflow process, and for file owners to lose control of the inputs and outputs that they are responsible for supporting. The lack of version control, workflow, and activity tracking is especially challenging in industries and geographies that require tracking of any personal data either related to employees or customers.

Spreadsheets also struggle in large data environments, which are quickly becoming commonplace in the business planning world. Although a core enterprise database may only be a few gigabytes, accurate planning now often includes access to sales, operations, and potentially even IT transactional data sources that can quickly expand beyond the memory and data size constraints that spreadsheets are designed to use. From Amalgam Insights’ perspective, the size and variety of data are the biggest technical constraints that spreadsheets face as planning solutions.

Spreadsheets lack advanced analytic and machine learning capabilities. Although algorithmic, statistical, and machine learning tools are increasingly becoming part of the FP&A world, especially in forecasting, Amalgam Insights finds in practice that most organizations have not yet embraced complex analytics as a core part of their FP&A approach. Based on current job site metrics, Amalgam Insights estimates that less than 2% of FP&A professionals currently have a machine learning or data science certification or degree, making this an early innovator capability that has still not crossed the chasm to become a standard job requirement for FP&A.

But perhaps the most significant challenge with spreadsheet models is that they are often fragile: created based on the logic of a single person rather than on defined business logic and with little to no documentation associated with the plans, forecasting algorithms, and multi-tabular complexity that inevitably occurs when a spreadsheet is the primary planning tool for a business, which can also lead to costly data accuracy issues. The model is only as adaptable as the spreadsheet creator’s knowledge of the industry and is dependent on that employee staying employed. Considering that it is unrealistic to expect an FP&A senior analyst to remain in that role for more than five years before either getting promoted or getting a better offer, this human risk is a significant challenge for business planning solutions.

As organizations grow in size to support more than a handful of locations and a set of workers that exceeds Dunbar’s number of 150 colleagues, Amalgam Insights believes that it becomes necessary to adopt a formalized planning solution that supports collaboration, scale, advanced analytics, continuous planning across many scenarios, and advanced forecasting analytics. Otherwise, it is difficult for businesses to maintain a consistent and shared version of the truth across financial planning and analysis personnel that can drive both departmental and executive planning efforts.

Ultimately, the use of spreadsheets as a formal system of record for business planning is a risky one for any organization with a formal corporate structure, governed industry or geography, or any organization that has a significantly distributed business. The ubiquity of the spreadsheet makes it an easy place to start modeling a budget, and the value of the spreadsheet in helping users to structure small datasets will exist for the foreseeable future. But the fragility of the data structure, lack of user and version control governance, inability to scale, and the difficulty of verifying data with other sources while avoiding human error all lead to the need of supporting a more formalized planning solution over time. As organizations face a future of keeping distributed groups focused on a shared version of the truth and collectively consider a variety of scenarios at any given time, the risk of spreadsheet fragility needs to be matched up against the value of using a formalized FP&A solution designed to analyze, govern, and protect all relevant business data, formulas, and outcomes.

Posted on by

Navigating The Road to Retail Analytic Success

Analytics in the Retail and Consumer Packaged Goods (CPG) markets is more complex than the average corporate data ecosystem because of the variety of analytic approaches needed to support these organizations. Every business has operational management capabilities for core human resources and financial management, but retail adds the complexities of hybrid workforce management, scheduling, and operational analytics as well as the front-end data associated with consumer marketing, e-commerce, and transactional behavior across every channel.

In contrast, when retail organizations look at middle-office and front-office analytics, they are trying to support a variety of timeframes ranging from intraday decisions associated with staffing and customer foot traffic to the year-long cycles that may be necessary to fulfill large wholesale orders for highly coveted goods in the consumer market. Over the past three years, operational consistency has become especially challenging to achieve as COVID, labor skill gaps, logistical bottlenecks, commodity shortages, and geopolitical battles have all made supply chain a massive dynamic risk factor that must be consistently monitored across both macro and microeconomic business aspects.

The lack of alignment and connection between the front office, middle-office, and administrative analytic outputs can potentially lead to three separate silos of activity in the retail world—     connected only by some basic metrics, such as receipts and inventory turnover, that are interpreted in three different ways. Like the parable of the blind men and an elephant where each person feels one part of the elephant and imagines a different creature, the disparate parts of retail organizations must figure out how to come together, as the average net margin for general retail companies is about 2% and that margin only gets lower for groceries and for online stores.

Analytic opportunities to increase business value exist across the balance sheet and income statement. Even though consumer sentiment, foot traffic, and online behavior are still key drivers for retail success, analytic and data-driven guidance can provide value across infrastructure, risk, and real-time operations. Amalgam Insights suggests that each of these areas requires a core analytic focus that is different and reflects the nature of the data, the decisions being made, and the stakeholders involved.

Facing Core Retail Business Challenges

First, retail and CPG organizations face core infrastructure, logistics, and data management challenges that typically require building out historic analysis and quantitative visibility capabilities often associated with what is called descriptive or historical analytics. When looking at infrastructure factors such as real estate, warehousing, and order fulfillment issues, organizations must have access to past trends, costs, transactions, and the breadth of relevant variables that go into real estate costs or complex order fulfillment associated with tracking perfect order index.

This pool of data ideally combines public data, industry data, and operational business data that includes, but is not limited to, sales trends, receipts, purchase orders, employee data, loyalty information, customer information, coupon redemption, and other relevant transactional data. This set of data needs to be available as analytic and queryable data that is accessible to all relevant stakeholders to provide business value. In practice, this accessibility typically requires some infrastructure investment either by a company or a technology vendor willing to bear the cost of maintaining a governed and industry-compliant analytic data store. By doing so, retail organizations have the opportunity to improve personalization and promotional optimization.

A second challenge that retail analytics can help with is associated with the risk and compliance issues that retail and CPG organizations face, including organized theft, supplier risk, and balancing risk and reward tradeoffs. A 2022  National Retail Federation (NRF) survey showed that organized retail crime had increased over 26% year over year, driving the need to identify and counter organized theft efforts and tactics more quickly. Retailer risk for specific goods and brands also needs to be quantified to identify potential delays and challenges or to determine whether direct store delivery and other direct-to-market tactics may end up being a profitable approach for key SKUs. Risk also matters from a profitability analysis perspective as retail organizations seek to make tradeoffs between the low-margin nature of retail business and the consumer demand for availability, personalization, automation, brand expansion, and alternative channel delivery that may provide exponential benefits to profits. From a practical perspective, this risk analysis requires investment in a combination of predictive analytics and the ability to translate the variance and marginal cost associated with new investments with projected returns.

A third challenge for retail analytics is to support real-time operational decisions. This use case requires access to streaming and log data associated with massive volumes of rapid transactions, frequently updated time-series data, and contextualized scenarios based on multi-data-sourced outcomes. From a retail outcome perspective, the practical challenge is to make near-real-time decisions, such as same-day or in-shift decisions to support stocking, scheduling, product orders, pricing and discounting decisions, placement decisions, and promotion. In addition, these decisions must be made in the context of broader strategic and operational concerns, such as brand promise, environmental concerns, social issues, and regulatory governance and compliance associated with environmental, social, and governance (ESG) concerns.

As an example, supply chain shortages often come from unexpected sources. An unexpected geopolitical example occurred in the United States when the government’s use of containers as a temporary barrier to block illegal immigration checkpoints on the US-Mexico border led to shortages at US ports for delivery. This delay in accessing containers was not predictable based solely on standard retail metrics and behavior and demonstrates one example of how unexpected political issues can affect a hyperconnected logistical supply chain.

Recommendations for Upgrading Retail Analytics in the 2020s

To solve these analytic problems, retail and CPG organizations need to allow line-of-business, logistics, and sourcing managers to act quickly with self-service and on-demand insights based on all relevant data. This ultimately means that to take an analytic approach to retail,     Amalgam Insights recommends the following three best practices in creating a more data-driven business environment.

  • Create and implement an integrated finance, operational, and workforce management environment. Finance, inventory, and labor must be managed together in an integrated business data store and business planning environment or the retail organization falls apart. Whether companies choose to do this by knitting together multiple applications with data management and integration tools or by choosing a single best-in-breed suite, retail businesses have too many moving parts to split up core operational data across a variety of functional silos and business roles that do not work together. In the 2020s, this is a massive operational disadvantage.
  • Adopt prescriptive analytics, decision intelligence, and machine learning capabilities above and beyond basic dashboards. When retail organizations look at analytics and data outputs, it is not enough to gain historical visibility. In today’s AI-enabled world, companies must have predictive analytics, statistical analysis, detect anomalies quickly, and have the ability to translate business data into machine learning and language models for the next generation of analytics and decision intelligence. Retail can be more proactive and prescriptive with AI and ML models trained to their enterprise data to support more personalized and contextualized purchasing experiences.
  • Implement real-time alerts with relevant and actionable retail information. Finally, timely and contextual alerts are also now part of the analytic process. As retail organizations have moved from seasonal purchases and monthly budgeting to daily or even hourly decisions, regional and branch managers need to be able to move quickly if there are signs of business danger coordinated revenue leakage, brand damage across any of the products held within the store, unexpected weather phenomena, labor issues, or other incipient macro or microeconomic threats.
Posted on by

Workday AI and ML Innovation Summit: Chasing the Eye of the AI Storm

We are in a time of transformational change as the awareness of artificial intelligence (AI) grows during a time of global uncertainty. The labor supply chain is fluctuating quickly and the economy is on rocky ground as interest rates and geopolitical strife create currency challenges. Meanwhile, the commodity supply chain is in turmoil, leading to chaos and confusion. Rising interest rates and a higher cost of money are only adding to the challenges faced by those in the global business arena. In this world where technology is dominant in the business world, the global economic foundation is shifting, and the worlds of finance and talent are up for grabs, Workday stepped up to hold its AI and ML Innovation summit to show a way forward for the customers of its software platform, including a majority of the Fortune 500 that use Workday already as a system of record.

The timing of this summit will be remembered as a time of rapid AI change, with new major announcements happening daily. OpenAI’s near-daily announcements regarding working with Microsoft, launching ChatGPT, supporting plug-ins, and asking for guidance on AI governance are transforming the general public’s perception of AI. Google and Meta are racing to translate their many years of research in AI into products. Generative AI startups already focused on legal, contract, decision intelligence, and revenue intelligence use cases are happy to ride the coattails of this hype. Universities are showing how to build large language models such as Stanford’s Alpaca. And existing machine learning and AI companies such as Databricks are showing how to build custom models based on existing data for a fraction of the cost needed to build GPT.

In the midst of this AI maelstrom, Workday decided to chase the eye of the hurricane and put stakes in the ground on its current approach to innovation, AI, and ML. From our perspective, we were interested both in the executive perspective and in the product innovation associated with this Brave New World of AI.

Enter the Co-CEO – Carl Eschenbach

Workday’s AI and ML Innovation Summit commenced with an introduction of the partners and customers that would be present at the event. The Summit began with a conversation between Workday’s Co-CEOs, Aneel Bhusri and Carl Eschenbach, where Eschenbach talked about his focus on innovation and growth for the company. Eschenbach is not new to Workday, having been on its board during his time at Sequoia Capital, where he also led investments in Zoom, UIPath, and Snowflake. Having seen his work at VMware, Amalgam Insights was interested to see Eschenbach take this role and help Workday evolve its growth strategy from an executive level. From the start, both Bhusri and Eschenbach made it clear that this Co-CEO team is intended to be a temporary status with Eschenbach taking the reins in 2024, while Bhusri becomes the Executive Chair of Workday.

Eschenbach emphasized in this session that Workday has significant opportunities in providing a full platform solution, and its international reach requires additional investment both in technology and go-to-market efforts. Workday partners are essential to the company’s success and Eschenbach pointed out a recent partnership with Amazon to provide Workday as a private offering that can use Amazon Web Service contract dollars to purchase Workday products once the work is scoped by Workday. Workday executives also mentioned the need for consolidation, which is one of Amalgam Insights’ top themes and predictions for enterprise software for 2023. The trend in tech is shifting toward best-in-suite and strategic partnering opportunities rather than a scattered best-in-breed approach that may sprawl across tens or even hundreds of vendors.

These Co-CEOs also explored what Workday was going to become over the next three to five years to take the next stage of its development after Bhusri evolved Workday from an HR platform to a broader enterprise software platform. Bhusri sees Workday as a system of record that uses AI to serve customer pain points. He poses that ERP is an outdated term, but that Workday is currently categorized as a “services ERP” platform in practice when Workday is positioned as a traditional software vendor. Eschenbach adds that Workday is a management platform across people and finances on a common multi-tenant platform.

From Amalgam Insights’ perspective, this is an important positioning as Workday is establishing that its focus is on two of the highest value and highest cost issues in the company: skills and money. Both must exist in sufficient quantities and quality for companies to survive.

The Future of AI and Where Workday Fits

We then heard from Co-President Sayan Chakraborty, who took the stage to discuss the “Future of Work” across machine learning and generative AI. As a member of the National Artificial Intelligence Advisory Committee, the analysts in the audience expected Chakraborty to have a strong mastery of the issues and challenges Workday faced in AI and this expectation was clarified by the ensuing discussion.

Chakraborty started by saying that Workday is monomaniacally focused on machine learning to accelerate work and points out that we face a cyclical change in the nature of the working age across the entire developed world. As we deal with a decline in the percentage of “working-age” adults on a global scale, machine learning exists as a starting point to support structural challenges in labor structures and work efforts.

To enable these efforts, Chakraborty brought up the technology, data, and application platforms based on a shared object model, starting with the Workday Cloud Platform and including analytics, Workday experience, and machine learning as specific platform capabilities. Chakraborty referenced the need for daily liquidity FDIC requests as a capability that is now being asked for in light of banking failures and stresses such as the recent Silicon Valley Bank failure.

Workday has four areas of differentiation in machine learning: data management, autoML (automated machine learning, including feature abstraction), federated learning, as well as a platform approach. Workday’s advantage in data is stated across quantity, quality associated with a single data model, structure and tenancy, and the amplification of third-party data. As a starting point, this approach allows Workday to support models based on regional or customer-specific data supported by transfer learning. At this point, Chakraborty was asked why Workday has Prism in a world of Snowflake and other analytic solutions capable of scrutinizing data and supporting analytic queries and data enrichment. Prism is currently positioned as an in-platform capability that allows Workday to enrich its data, which is a vital capability as companies face the battle for context across data and analytic outputs. 

Amalgam Insights will dig into this in greater detail in our recommendations and suggestions, but at this point we’ll note that this set of characteristics is fairly uncommon at the global software platform level and presents opportunities to execute based on recent AI announcements that Workday’s competitors will struggle to execute on.

Workday currently supports federated machine learning at scale out to the edge of Workday’s network, which is part of Workday’s differentiation in providing its own cloud. This ability to push the model out to the edge is increasingly important for supporting geographically specific governance and compliance needs (dubbed by some as the “Splinternet“) as Workday has seen increased demand for supporting regional governance requests leading to separate US and European Union machine learning training teams each working on regionally created data sources.

Chakraborty compared Workday’s approach of a platform machine learning approach leading to a variety of features to traditional machine learning feature-building approaches where each feature is built through a separate data generation process. The canonical Workday example is Workday’s Skills Cloud platform where Workday currently has close to 50,000 canonical skills and 200,000 recognized skills and synonyms scored for skill strength and validity. This Skills Cloud is a foundational differentiator for Workday and one that Amalgam Insights references regularly as an example of a differentiated syntactic and semantic layer of metadata that can provide differentiated context to a business trying to understand why and how data is used.

Workday mentioned six core principles for AI and ML, including people and customers, built to ensure that the machine learning capabilities developed are done through ethical approaches. In this context, Chakraborty also mentioned generative AI and large language models, which are starting to provide human-like outputs across voice, art, and text. He points out how the biggest change in AI occurred in 2006 when NVIDIA created GPUs, which used matrix math to support the constant re-creation of images. Once GPUs were used from a computational perspective, they made massively large parameter models possible. Chakraborty also pointed out the 2017 DeepMind paper on transformers to solve problems in parallel rather than sequentially, which led to the massive models that could be supported by cloud models. The 1000x growth in two years is unprecedented even from a tech perspective. Models have reached a level of scale where they can solve emergent challenges that they have not been trained on. This does not imply consciousness but does demonstrate the ability to analyze complex patterns and systems behavior. Amalgam Insights notes that this reflects a common trend in technology where new technology approaches often take a number of years to come to market, only to be treated as instant successes once they reach mainstream adoption.

The exponential growth of AI usage was accentuated in March 2023 when OpenAI, Microsoft, Google, and others provided an unending stream of AI-based announcements including OpenAI’s GPT 4 and GPT Plugins, Microsoft 365 Copilot and Microsoft Security Copilot, Google providing access to its generative AI Bard, Stanford’s $600 Alpaca generative AI model, and Databricks’ Dolly, which allows companies to build custom GPT-like experiences. This set of announcements, some of which were made during the Workday Innovation Summit, shows the immense nature of Workday’s opportunity as one of the premier enterprise data sources in the world that will both be integrated into all of these AI approaches.

Chakraborty points out that the weaknesses of GPT include bad results and a lack of explainability in machine learning, bad actors (including IP and security concerns), and the potential Environmental, Social, and Governance costs associated with financial, social, and environmental concerns. As with all technology, GPT and other generative AI models take up a lot of energy and resources without any awareness of how to throttle down in a sustainable and still functional manner. From a practical perspective, this means that current AI systems will be challenged to manage uptime as all of these new services attempt to benchmark and define their workloads and resource utilization. These problems are especially problematic in enterprise technology as the perceived reliability of enterprise software is often based on near-perfect accuracy of calculating traditional data and analytic outputs.

Amalgam Insights noted in our review of ChatGPT that factual accuracy and intellectual property attribution have been largely missing in recent AI technologies that have struggled to understand or contextualize a question based on surroundings or past queries. The likes of Google and Meta have focused on zero-shot learning for casual identification of trends and images rather than contextually specific object identification and topic governance aligned to specific skills and use cases. This is an area where both plug-ins and the work of enterprise software companies will be vital over the course of this year to augment the grammatically correct responses of generative AI with the facts and defined taxonomies used to conduct business.

Amalgam also found it interesting that Chakraborty mentioned that the future of models would include high-quality data and smaller models custom-built to industry and vertical use cases. This is an important statement because the primary discussion in current AI circles is often about how bigger is better and how models compete on having hundreds of billions of parameters to consider. In reality, we have reached the level of complexity where a well-trained model will provide responses that reflect the data that it has been trained on. The real work at this point is on how to better contextualize answers and how to separate quantitative and factual requests from textual and grammatical requests that may be in the same question. The challenge of accurate tone and grammar is very different from the ability to understand how to transform an eigenvector and get accurate quantitative output. Generative AI tends to be good at grammar but is challenged by quantitative and fact-based queries that may have answers that differ from its grammatical autocompletion logic.

Chakraborty pointed out that reinforcement learning has proven to be more useful than either supervised or unsupervised training for machine learning, as it allows models to look at user behavior rather than forcing direct user interaction. This Workday focus both provides efficacy of scale and takes advantage of Workday’s existing platform activities. This combination of reinforcement training and Workday’s ownership of its Skills Cloud will provide a sizable advantage over most of the enterprise AI world in aligning general outputs to the business world.

Amalgam Insights notes here that another challenge of the AI discussion is how to create an ‘unbiased’ approach for training and testing models when the more accurate question is to document the existing biases and assumptions that are being made. The sooner we can move from the goal of being “unbiased” to the goal of accurately documenting bias, the better we will be able to trust the AI we use.

Recommendations for the Amalgam Community on Where Workday is Headed Next

Obviously, this summit provided Amalgam Insights both with a lot of food for thought provided by Workday’s top executives. The introductory remarks summarized above were followed up with insight and guidance on Workday’s product roadmap across both the HR and finance categories where Workday has focused its product efforts, as well as visibility to the go-to-market and positioning, approaches that Workday plans to provide in 2023. Although much of these discussions were held under a non-disclosure agreement, Amalgam Insights will try to use this guidance to help companies to understand what is next from Workday and what customers should request. From an AI perspective, Amalgam Insights believes that customers should push Workday in the following areas based on Workday’s ability to deliver and provide business value.

  1. Use the data model to both create and support large language models (LLMs). The data model is a fundamental advantage in setting up machine learning and chat interfaces. Done correctly, this is a way to have a form of Ask Me Anything for the company based on key corporate data and the culture of the organization. This is an opportunity to use trusted data to provide relevant advice and guidance to the enterprise. As one of the largest and most trusted data sources in the enterprise software world, Workday has an opportunity to quickly build, train, and deploy models on behalf of customers, either directly or through partners. With this capability, “Ask Workday” may quickly become the HR and finance equivalent of “Ask Siri.”
  2. Use Workday’s Skills Cloud as a categorization to analyze the business, similar to cost center, profit center, geographic region, and other standard categories. Workforce optimization is not just about reducing TCO, but aligning skills, predicting succession and future success potential, and market availability for skills. Looking at the long-term value of attracting valuable skills and avoiding obsolete skills is an immense change for the Future of Work. Amalgam Insights believes that Workday’s market-leading Skills Cloud provides an opportunity for smart companies to analyze their company below the employee level and actually ascertain the resources and infrastructure associated with specific skills.
  3. Workday still has room to improve regarding consolidation, close, and treasury management capabilities. In light of the recent Silicon Valley Bank failure and the relatively shaky ground that regional and niche banks currently are on, it’s obvious that daily bank risk is now an issue to take into account as companies check if they can access cash and pay their bills. Finance departments want to consolidate their work into one area and augment a shared version of the truth with individualized assumptions. Workday has an opportunity to innovate in finance as comprehensive vendors in this space are often outdated or rigidly customized on a per-customer level that does not allow versions to scale out in a financially responsible way as the Intelligent Data Core allows. And Workday’s direct non-ERP planning competitors mostly lack Workday’s scale both in its customer base and consultant partner relationships to provide comprehensive financial risk visibility across macroeconomic, microeconomic, planning, budgeting, and forecasting capabilities. Expect Workday to continue working on making this integrated finance, accounting, and sourcing experience even more integrated over time and to pursue more proactive alerts and recommendations to support strategic decisions.
  4. Look for Workday Extend to be accessed more by technology vendors to create custom solutions. The current gallery of solutions is only a glimpse of the potential of Extend in establishing Workday-based custom apps. It only makes sense for Workday to be a platform for apps and services as it increasingly wins more enterprise data. From an AI perspective, Amalgam Insights would expect to see Workday Extend increasingly working with more plugins (including ChatGPT plugins), data models, and machine learning models to guide the context, data quality, hyperparameterization, and prompts needed for Workday to be an enterprise AI leader. Amalgam Insights also expects this will be a way for developers in the Workday ecosystem to take more advantage of the machine learning and analytics capabilities within Workday that are sometimes overlooked as companies seek to build models and gain insights into enterprise data.
Posted on by

14 Key Trends for Surviving 2023 as an IT Executive

2023 is going to be a tough year for anybody managing technology. As we face the repercussions of inflation and high interest rates and the bubble of tech starts to be burst, we are seeing a combination of hiring freezes, increased focus on core business activities and the hoary request to “do more with less.”

Behind the cliche of doing more with less is the need to actually become more efficient with tech usage. This means adopting a FinOps (Financial Operations) strategy to cloud to go with your existing Telecom FinOps (aka Telecom expense) and SaaS FinOps (aka SaaS Management) strategies. And it means being prepared for new spend category challenges as companies will need to invest in technology to get work done at a time when it is harder to hire the right person at the right time. Here is a quick preview of our predictions.

 

14 Key Predictions for the IT Executive in 2023

To get the details on each of these trends and predictions and understand why they matter in 2023, download this report at no cost by filling out this quick form to join our low-volume bi-monthly mailing list. (Note: If you do not wish to join our mailing list, you can also purchase a personal license for this report.)

Posted on by

What To Watch Out For As GPT Leads a New Technological Revolution

2022 was a banner year for artificial intelligence technologies that reached the mainstream. After years of being frustrated with the likes of Alexa, Cortana, and Siri and the inability to understand the value of machine learning other than as a vague backend technology for the likes of Facebook and Google, 2022 brought us AI-based tools that was understandable at a consumer level. From our perspective, the most meaningful of these were two products created by OpenAI: DALL-E and ChatGPT, which expanded the concept of consumer AI from a simple search or networking capability to a more comprehensive and creative approach for translating sentiments and thoughts into outputs.

DALL-E (and its successor DALL-E 2) is a system that can create visual images based on text descriptions. The models behind DALL-E look at relationships between existing images and the text metadata that has been used to describe those images. Based on these titles and descriptions, DALL-E uses diffusion models to start with random pixels that lead to generated images based on these descriptions. This area of research is by no means unique to OpenAI, but it is novel to open up a creative tool such as DALL-E to the public. Although the outputs are often both interesting and surprisingly different from what one might have imagined, they are not without their issues. For instance, the discussion around the legal ownership of DALL-E created graphics is not clear, since Open AI claims to own the images used, but the images themselves are often based on other copyrighted images. One can imagine that, over time, an artistic sampling model may start to appear similar to the music industry where licensing contracts are used to manage the usage of copyrighted material. But this will require greater visibility regarding the lineage of AI-based content creation and the data used to support graphic diffusion. Until this significant legal question is solved, Amalgam Insights believes that the commercial usage of DALL-E will be challenging to manage. This is somewhat reminiscent of the challenges that Napster faced at the end of the 20th century as a technology that both transformed the music industry and had to deal with the challenges of a new digital frontier.

But the technology that has taken over the zeitgeist of technology users is ChatGPT and related use cases associated with the GPT (Generative Pre-Trained Transformer) autoregressive language model trained on 500 billion words across the web, Wikipedia, and books. And it has become the favorite plaything of many a technologist. What makes ChatGPT attractive is its ability to take requests from users asking questions with some level of subject matter specificity or formatting and to create responses in real-time. Here are a couple of examples from both a subject matter and creative perspective.

Example 1: Please provide a blueprint for bootstrapping a software startup.

This is a bit generic and lacks some important details on how to find funding or sell the product, but it is in line with what one might expect to see in a standard web article regarding how to build a software product. The ending of this answer shows how the autogenerative text is likely referring to prior web-based content built for search engine optimization and seeking to provide a polite conclusion based on junior high school lessons in writing the standard five-paragraph essay rather than a meaningful conclusion that provides insight. In short, it is basically a status quo average article with helpful information that should not be overlooked, but is not either comprehensive or particularly insightful for anyone who has ever actually started a business.

A second example of ChatGPT is in providing creative structural formats for relatively obscure topics. As you know, I’m an expert in technology expense management with over two decades of experience and one of the big issues I see is, of course, the lack of poetry associated with this amazing topic. So, again, let’s go to ChatGPT.

Example 2: Write a sonnet on the importance of reducing telecom expenses

As a poem, this is pretty good for something written in two seconds. But it’s not a sonnet, as sonnets are 14 lines, written in iambic pentameter (10 syllable lines split int 5 iambs, or a unstressed syllable followed by a stressed syllable) and split into three sections of four lines followed by a two-line section with a rhyme scheme of ABAB, CDCD, EFEF, GG. So, there’s a lot missing there.

So, based on these examples, how should ChatGPT be used? First, let’s look at what this content reflects. The content here represents the average web and text content that is associated with the topic. With 500 billion words in the GPT-3 corpus, there is a lot of context to show what should come next for a wide variety of topics. Initial concerns of GPT-3 have started with the challenges of answering questions for extremely specific topics that are outside of its training data. But let’s consider a topic I worked on in some detail back in my college days while using appropriate academic language in asking a version of Gayatri Spivak’s famous (in academic circles) question “Can the subaltern speak?”

Example 3: Is the subaltern allowed to fully articulate a semiotic voice?

Considering that the language and topic here is fairly specialized, the introductory assumptions are descriptive but not incisive. The answer struggles with the “semiotic voice” aspect of the question in discussing the ability and agency to use symbols from a cultural and societal perspective. Again, the text provides a feeling of context that is necessary, but not sufficient, to answer the question. The focus here is on providing a short summary that provides an introduction to the issue before taking the easy way out telling us what is “important to recognize” without really taking a stand. And, again, the conclusion sounds like something out of an antiseptic human resources manual in asking for the reader to consider “different experiences and abilities” rather than the actual question regarding the ability to use symbols, signs, and assumptions. This is probably enough of an analysis at a superficial level as the goal here isn’t to deeply explore postmodern semiotic theory but to test ChatGPT’s response in a specialized topic.

Based on these three examples, one should be careful in counting on ChatGPT to provide a comprehensive or definitive answer to a question. Realistically, we can expect ChatGPT will provide representative content for a topic based on what is on the web. The completeness and accuracy of a ChatGPT topic is going to be dependent on how often the topic has been covered online. The more complete an answer is, the more likely it is that this topic has already been covered in detail.

ChatGPT will provide a starting point for a topic and typically provide information that should be included to introduce the topic. Interestingly, this means that ChatGPT is significantly influenced by the preferences that have built online web text over the past decade of content explosion. The quality of ChatGPT outputs seems to be most impressive to those who treat writing as a factual exercise or content creation channel while those who look at writing as a channel to explore ideas may find it lacking for now based on its generalized model.

From a topical perspective, ChatGPT will probably have some basic context for whatever text is used in a query. It would be interesting to see the GPT-3 model augmented with specific subject matter texts that could prioritize up-to-date research, coding, policy, financial analysis, or other timely new content either as a product or training capability.

In addition, don’t expect ChatGPT to provide strong recommendations or guidance. The auto-completion that ChatGPT does is designed to show how everyone else has followed up on this topic. And, in general, people do not tend to take strong stances on web-based content or introductory articles.

Fundamentally, ChatGPT will do two things. First, it will make mediocre content ubiquitous. There is no need to hire people to write an “average” post for your website anymore as ChatGPT and other technologies either designed to compete with or augment it will be able to do this easily. If your skillset is to write grammatically sound articles with little to no subject matter experience or practical guidance, that skill is now obsolete as status quo and often-repeated content can now be created on command. This also means that there is a huge opportunity to combine ChatGPT with common queries and use cases to create new content on demand. However, in doing so, users will have to be very careful not to plagiarize content unknowingly. This is an area where, just like with DALL-E, OpenAI will have to work on figuring out data lineage, trademark and copyright infringement, and appropriation of credit to support commercial use cases.  ChatGPT struggles with what are called “hallucinations” where ChatGPT makes up facts or sources because those words are physically close to the topic discussed in the various websites and books that ChatGPT uses. ChatGPT is a text generation tool that picks words based on how frequently they show up with other words. Sometimes that result will be extremely detailed and current and other times, it will look very generic and mix up related topics that are often discussed together.

Second, this tool now provides a much stronger starting point for writers seeking to say something new or different. If your point of view is something that ChatGPT can provide in two seconds, it is neither interesting or new. To stand out, you need to provide greater insight, better perspective, or stronger directional guidance. This is an opportunity to improve your skills or to determine where your professional skills lie. ChatGPT still struggles with timely analysis, directional guidance, practical recommendations beyond surface-level perspectives, and combining mathematical and textual analysis (i.e. doing word problems or math-related case studies or code review) so there is still an immense amount of opportunity for people to write better.

Ultimately, ChatGPT is a reflection of the history of written text creation, both analog and digital. Like all AI, ChatGPT provides a view of how questions were answered in the past and provides an aggregate composite based on auto-completion. For topics with a basic consensus, such as how to build a product, this tool will be an incredible time saver. For topics that may have multiple conflicting opinions, ChatGPT will try to play either both sides or all sides in a neutral manner. And for niche topics, ChatGPT will try to fake an answer at what is approximately a high school student’s understanding of the topic. Amalgam Insights recommends that all knowledge workers experiment with ChatGPT in their realm of expertise as this tool and the market of products that will be built based on the autogenerated text will play an important role in supporting the next generation of tech.

Posted on by

December 2: From BI to AI, Part 1 (AtScale, Dremio, Matillion, Informatica, Starburst)

Launches and Updates

AtScale’s Semantic Layer Launches on Google Cloud Marketplace

Semantic layer platform AtScale announced Thursday that it was now available on Google Cloud Marketplace. Mutual customers will be able to use AtScale on Google Cloud with services such as Google BigQuery, where they can run BI and OLAP workloads without needing to extract or move data. 

Dremio Announces Updates to Lakehouse Capabilities

Open data lakehouse Dremio announced a number of improvements this week. Among the updates: new SQL functionality, including support for the MAP data type so users can query map data from Parquet, Iceberg, and Delta Lake; security enhancements such as row and column-level policy-defined access control for users; support for INSERT, DELETE, and UPDATE on Iceberg tables, and for “time travel” to query historical data in place; as well as usability and performance improvements. Dremio also added a number of connectors, including dbt, Snowflake, MongoDB, DB2, OpenSearch, and Azure Data Explorer. 

Informatica Reveals AWS-Specific Cloud Data Management Services

At AWS re:Invent 2022, Informatica announced three new capabilities for Informatica within AWS. Informatica Data Loader is embedded within Amazon Redshift so that mutual customers will be able to ingest data from a wide variety of systems, including AWS. The Informatica Data Marketplace now supports AWS Data Exchange, allowing customers to access and use third-party data hosted on the Data Exchange. And Informatica INFACore, INFA’s new development and data science framework, simplifies the process of developing  and maintaining complex data pipelines, which can be shunted over to Amazon SageMaker Studio as a simple function, allowing users to pull prepared data from INFACore into SageMaker Studio for further use in building, training, and deploying machine learning models on SageMaker.

Matillion Accelerates Productivity for Data Teams with Key Ecosystem Integrations | Matillion 

Data productivity platform Matillion announced a number of integrations with technical partners. Most of these integrations are accelerators that speed up some aspect of data processing between Matillion and its partners. FHIR Data, built by Matillion and Hakkoda, is a Snowflake healthcare data integrator that simplifies the process of loading FHIR data in Snowflake, then transforming it into a structured format for analytics processing. AWS Redshift Serverless Scale will let mutual Matillion-AWS customers run analytics without needing to manually provision or manage data warehouse clusters. Matillion One Click within AllCloud automates the setup and maintenance of data pipelines. Finally, the Matillion-Collibra integration creates data lineage, mapping inbound and outbound data flows, and attaches data objects to assets in the Collibra data catalog.

Starburst Grows Galaxy with Data Products Capabilities

Analytics engine Starburst launched new capabilities for Starburst Galaxy, the managed service version of its primary Starburst Enterprise offering.  The new Data Products capabilities include a catalog explorer for users to search through their data more easily and understand what they have; schema discovery, which can help users find new datasets regardless of their storage format; and enhanced security and access controls. 

Starburst Enterprise Now Supports AWS Lake Formation and Further Data Federation

Starburst also announced support for AWS Lake Formation via Starburst Enterprise, which will allow joint customers to more easily implement a data mesh framework across all of an organization’s data sources.