Episode Transcript
[00:00:00] Today I'm going to tell you the story of the Jupiter release of Galileo. And it's a slightly long story, but I think you'll find it interesting and educational and a little promotional, but not too much so. Most of you know that the history of our business is research, advisory services, education and consulting. And so going back to 1998 when I first started doing this, what we basically have been doing for close to 30 years is studying the human capital practices of management, leadership, human resources, and all of the related tools and technologies that allow companies to manage and grow their people. It started originally with training and then we got into leadership development and then we got into succession management and performance management and talent management and executive development and pay equity and diversity and inclusion and well being and employee experience and the pandemic and org design and many, many other topics. And over this 28 year period, we identified 95 functional capability areas of human capital. And you could say they're functional capabilities of hr, but they're really functional capabilities of management in general.
[00:01:20] And in each of those 95 areas we've developed maturity models, benchmarks where they're appropriate assessments, and hundreds and hundreds and hundreds of case studies. And we have more than 1800, almost 1800 case studies. And the reason we've done all these case studies is the purpose of this research is to give you actionable solutions to problems in your industry, in your location, in your company. So size. If I tell you how Google or Walmart does performance management and you're a fast growing tech company or a small retailer or a retail distribution company in the uk, it may not apply at all. I mean, it might be interesting and you might learn something from it, but it may not be relevant. You'd probably rather hear a story from a company like yours and what happened to them. And over the years, what we found is that the way to produce and monetize this IP is through what we called a research membership. This is a very common model that Gartner uses in other research companies. Where we take the research, we produce it into reports and models and frameworks and graphics. By the way, this includes analysis of vendors, which is more of a vendor analyst work. But we do the same thing for vendors. So we have pretty much the same examples and same process where we talk to vendor customers, we talk to the vendors, we track their roadmaps and their success in the market and so forth. So we're kind of business analysts from that standpoint and HR analysts and management analysts at the same time. Anyway, we produce these reports Spend a lot of time making them beautiful, writing them, editing them. It's actually quite fun if you like writing. But it's, you know, a business, like a publishing business. And then we sell or have sold access to this library, like a library card, where you kind of just go in and read whatever you want. And of course, over the last decade, this has turned into videos, podcasts, and then educational materials. Because what we found is that nobody reads everything and everybody who's a consumer of our materials has a particular job and they want to know what's relevant to that job. They don't want to know what everything else is going on in management and hr.
[00:03:39] So we needed to build search tools and subscription models that that allowed us to sell access to this research to the appropriate people in a company. And that includes C level executives all the way down to recruiters or L and D specialists or technical specialists who are buying and selling, or buying and selling technology. And by the way, we also have a lot of clients who are consultants and vendors who are trying to keep up with the market too. And there's a whole bunch of intricacies to doing all this that I won't take you through. But, but the crux of it is being able to move quickly so that we can identify trends rapidly and make sense of them through research and not look like a pundit or a journalist creating a listicle or a position opinion on a topic. There's plenty of that out there and we have plenty of opinions too. And occasionally we take very strong opinions about things that we think are going in the wrong direction or need to be addressed, but we do that through research. And so most of our time is literally talking to companies, surveying companies, looking at data, looking at meta research from other people and making sense of this complex world of human capital in the business world. Now, the importance of what we do is very high right now because AI is threatening to change the way humans add value in companies, as you know.
[00:05:12] And we all need to learn how to use AI and superpower ourselves. And that changes the structure and the pay levels and the way we do performance management and the way we do development and training, the way we do employee support and employee experience, as well as the functional areas of hr, how we do recruiting, how we do L and D. I mean, if you fast forward AI two or three years from now into our HR 2030 model, all of these human capital agonizing decisions that people make all day is going to be made by AI. A lot of it's going to be either done exclusively by AI or supported by AI, because the AI will know a lot about every single person and what they've done and what they're working on and what their strengths and weaknesses are. So this. These practices are very dynamic. So when ChatGPT first came out, I knew there was an opportunity for us. And I originally thought to myself, wouldn't it be nice to have a search engine for where somebody could search our entire corpus of knowledge and get the answer to a question and find the report they want to read? So we built a prototype of Galileo, which we called the Copilot. And it worked extremely well because it did much more than we expected. Not only did it find information and answer questions, but it could generate information. You could use our intellectual property to build an implementation plan or an RFP checklist, or a hiring guide, or a. A behavioral interview guide or a skills model. I mean, it's really good. And we moved it to sana, which was a private company at the time. It was not owned by Workday yet. And they helped us expand it and it became Galileo. And we built it on a platform that allows you to add your own data to it and your own policies. And so we have more than probably 2,000 companies using Galileo for different things. And it was originally positioned as a tool for HR professionals. But I always knew from the beginning, when I first started doing all this work, that we were writing for business people. And I used to read the Wall Street Journal in the days that it was print, and I would read every single article and I would say, wow, that's the kind of thing we should be writing. Because sometimes the Wall Street Journal does very useful information, not just news information about how to apply different principles of management and leadership and HR in a business. So, anyway, so we're sort of this business and HR research company moving to AI and Galile is like miraculously powerful tool. People love it. It's very exciting. It does amazing things. We've written many, many documents on how to use it use cases, hundreds of use cases in every domain of management and hr. However, what happened, of course, is that the agent and AI market has become huge, and people don't want to necessarily buy a new agent for a specialized purpose. They might. We're still selling Galileo pretty regularly, by the way. We also added to Galileo a learning compone built again on sana, so that you would not only be able to read and generate information and answer questions, but it would teach you things. So the sana, the Galileo interface, actually brings teaching into the user experience. Which you don't get through Claude or ChatGPT today. But what happened, of course, is you guys went out and bought a lot of tools and you said to us, well, I don't really want to use your AI. I want you to do it in our AI. I have ChatGPT or I have Gemini, or I have a homegrown thing, or I have Claude, or I have Microsoft Copilot. So we said to ourselves, we need to build a more portable version of Galileo that we can move to these other platforms. And we started doing deals with ServiceNow and SAP and other companies to try to figure out how to move our corpus of knowledge from platform to platform. And it was. We didn't have MCP yet, so it was a project, but we got it going. So we have a version of Galileo that runs on now Assist. We have a version that runs on Joule that's in sort of prototype with SAP. And then, of course, Workday embedded Sana into their platform. So we work within Workday. But we realized that the power of Galileo was much, much bigger than the platform was letting us deliver because the platform was getting, in a way.
[00:09:33] And we also realized that a lot of the use cases of Galileo were the interleaving of Galileo's intelligence with the intelligence of the company, the policies of the company, of course, but also the human capital data of the company. So we built a layer on top of Galileo that understands the data models of SAP, Oracle, Workday, and now hibob, so that if you're aware, if you're using an agent that comes from or has connection to those systems, you can ask it a question like, what is the span of control of our sales department? What's the average salary compared to the benchmarks for salaries in that tenure in that region? And it'll do all that, which is pretty unbelievable, to be honest. But we also found, as we were doing this, is that the way our corpus works was not as optimized or efficient as we'd like. And so in the middle of thinking about how to make Galileo more portable and turn it into an enterprise intelligence service, we realized we could rethink the actual content itself. So we went back and looked at the way our content was written. And of course, the way it's written is the way any other book is written. It has an introduction, it has a table of contents, it has a bunch of stuff in different chapters and then has a conclusion at the end and then a bunch of boilerplate around it. Well, the AI has to make sense of all of that and it doesn't need the boilerplate, it doesn't need the introduction, it doesn't need the table of contents, it just wants the content.
[00:11:07] So we essentially built a new architecture which we call ARC Agent Ready Corpus that takes the essence of Galileo and all of the data. By the way, there's a lot of data in here too. There's numbers, there's salary data, there's data about turnover, there's data about pricing for different products, many, many numeric information, the data about maturity models, and we chunked that up into machine readable form and we carefully indexed it and tagged it in eight different dimensions of tagging so that the LLM can read it extremely well. And it does not have to hallucinate, it does not have to guess, it does not have to make things up. Because if you ask ChatGPT a typical question about management, it's probably going to give you an answer, but you can't really tell where it came from. It's not really clear why it came to the conclusion it did. And a lot of times, as I'll mention in a minute, it's incorrect. And so this agent Ready Corpus, we realized after we got it working, was 10 to 100 times more efficient at token usage for running these complex questions that people want to run on hr. So we've essentially built a new architecture for the content itself. Now, while we were doing all this, of course, we needed to make sure that it worked. And so we spent some time with some engineers from anthropic and OpenAI and others. We kind of picked our brains and they gave us some advice on how to do this.
[00:12:37] And we then after we got it all working, we ran benchmarks. We have 30 golden prompts, which we call them, which are complex questions about human capital.
[00:12:47] And we, we benchmark these prompts on all of the platforms we're running on. And what we found is that when you run those 30 prompts on ChatGPT or Claude alone, it hallucinates and makes mistakes, it misattributes information, it's not very good. So this new corpus, which is the highly tuned, highly optimized version of our corpus, including, by the way, there's podcasts, there's videos, there's interviews, there's a lot of sources of information here is very fast, very secure, very reliable, predictable, truthful and efficient. And that is what is going on with Jupyter. Jupyter is the same information we've had for years, but highly optimized for the consumption and delivery by AI. It's as if we took all of the 30 years of printing and layout and fonts and colors that we spent on our library and regenerated it so that the machine could read it perfectly and make completely accurate and detailed analysis of answers by industry, by company size, by domain, by problem.
[00:13:58] And you know, a lot of what we do in our advisory work is we don't just kind of spout out a bunch of HR stuff. We start with the business problem. Because ultimately the reason you select an approach or a tool or a strategy for solving a human capital problem is because you have a growth or profitability problem in the company or a competitive problem in the company. And that's where this all starts. So we now have a system that will easily create a chain of thought between here's my business problem and here's my human capital issue and here's my solution. And that, by the way, that chain of thought process is embedded in our HR 2030 model, which we don't sell, but you can get access to it with us and we'll use it with you on a consulting basis at this point. We may productize it later. So Galileo Jupiter is this new highly optimized, high performance, highly accurate, highly responsible, trusted system for all aspects of hr. We also added, by the way, a whole bunch of new salary data. We added executive compensation data from Ecuilar into the system. So it has job titles of every major job in every functional area, skills associated with those jobs, salary benchmarks from around the world, which we now have from Wagescape and other sources, practices From I think 120 countries, benchmarks on turnover, benchmarks on all these different business practices, including HR and hundreds of data sets on various aspects of the vendor market itself and different technologies that you buy for the HR and management issues you have in your company.
[00:15:44] And now that we have this high performing corpus, we can feed it faster and faster and faster because we don't have to write PDF reports or do giant publishing projects every time we want to get information to you. We still do that and we're going to continue to do that, but honestly, over time we're going to do less of that because you can consume it as needed. And we're building a whole new website that'll be also part of this. And you know, you get what you need and then you dive in if you want and read the PDF if you feel like it, or not, or not, read the PDF and just get your answer and go back to work. And then go back and ask a couple more questions and we also abstracted away the HR jargon into the model through the metadata, so that now Galileo is a tool or a system or an intelligence for managers and employees, not just HR people. It doesn't just talk HR speak.
[00:16:36] And I know that sounds a little bit odd, but. But if you think about the kinds of questions your employees are asking of you as a company or each other or looking up on their portals, a lot of them have to do with management issues or human capital issues or culture issues or training issues or skills issues. So this is a body of intelligence, an intelligence layer in a sense, that now plugs into any AI agent you have. We also, by working very closely with Microsoft, and I mean really closely, for almost 18 months, have a very, very highly optimized version of Galileo designed for the Microsoft Copilot. And the Microsoft Copilot is very unique for one big reason. It has access to the Microsoft graph, it has access to emails, SharePoint, Viva skills, work IQ, all sorts of information about what's really going on in your company that's not located in the HCM system at all. In fact, it's not even close. So you can go into the Copilot implementation of Galileo and I've done this, and you can say, please read my last two years of emails and tell me what my top skills are relative to this new project I have to do and what you think I should learn to become ready for this project. And it does that and it works. And you could just think of the thousands and thousands of other things you can do now with the Copilot. And because the Copilot is such an easy to use tool set, you can build agents in the Copilot to do things like onboarding or checklists or project management or weekly status reviews. One of our clients actually built his own personal management coach that goes through his meetings from the prior week, uses his own rubric for management and gives him coaching on what he could be doing better this week versus last week. And Galileo informs that.
[00:18:33] So there's just hundreds of things you can do here with the Copilot because of the Copilot's footprint and Surface they call it, to access so much enterprise data. And most of you have IT departments that are connecting your corporate agents to many, many things. So you know, if you're connecting them to the time and materials system, the scheduling system, the payroll system, the training system, whatever it may be, the recruiting system, Galileo understands the context of all of that and makes it all smarter, more interesting, more relevant, more benchmarkable and more Problem solution oriented. So it's really exciting for us. You're going to see statements of support from ServiceNow, Workday, Microsoft, Hibob, who by the way has integrated Galileo into Hibob, which is where a lot of this is going and gloat. And we've also tested it with Glean, Gemini, Claude, ChatGPT and a few other smaller vendors. So whatever AI infrastructure you have, the Ark corpus of Galileo can make it smarter with no cost in tokens. The token cost of using Galileo is minuscule because it is so highly optimized for LLM consumption. So not only does it not hallucinate or make things up or guess, but it's very fast because the corpus is designed to be read by an LLM.
[00:19:59] And in the rest of our business this means that we can produce content even faster than ever before. We don't have to publish it if we don't want to. We can literally produce it in a form that goes into the R corpus that day and you'll see it within an hour. So this is a really powerful tool. We are really excited about it. We have hundreds of customers waiting for the Galileo port On these other platforms we have benchmarks, there's a whole series of white papers. We put together a technical white paper, an explainer video that explains how Ark works, a page of information on each of the platforms with support from the vendors, and there's a lot more to come. This is an amazingly fascinating business transformation for us and what it does for you is it. And by the way, we license this on a license basis so you don't have to pay token costs for any of this. What this does for you is it gives you a way to deliver a single pane of glass employee experience for every employee, every manager, every HR professional with a deep level of human capital intelligence that is independent, trusted and supported with authoritative background research on every single topic.
[00:21:13] And I think we could go through use cases for any one of you that have various ideas here. There are thousands and thousands of applications for Galileo and the Jupyter release really does warrant the name Jupyter because it's one of the biggest projects we've ever done and one of the most exciting things I've ever done in my career. We look forward to talking to you about this. I think that's enough for the podcast. Take a look at the materials and we will show you how it works in webinars coming up soon. Thanks everybody. Have a great week.
