Rob Collie P3 Adaptive is a Microsoft Solutions Partner for Data and AI, serving hundreds of clients in the mid-market and Fortune 1000. Rob was a former Microsoft engineer on Excel, Bing and Power BI Teams. He led the Power BI Wave after leaving Microsoft. Rob has written three business technology books with over 92,000 copies in print. He hosts the Raw Data With Rob Collie podcast. This is his fourth book. Fair Game: Customizing AI to Your Business Is Easier Than You Think The AI Moment (August 2026) will turn the practitioner’s credibility around.
P3 Adaptive was founded in 2013 by you after more than a ten-year stint at Microsoft, where you helped to develop the business intelligence capabilities in Excel and Power BI. What shaped your view on enterprise AI as you transitioned from developing software at Microsoft to solving clients’ data challenges?
Microsoft’s product development teams were building universal software, which could not be customized to the needs of any one customer. It was called “the’stuff of dreams. “ordering a pizza whose toppings were acceptable to 300 million people.” There’s necessarily a lowest-common-denominator vibe to that job, as well as a certain distance from individual customers.
It was certainly a great honor to work on such a large stage. However, it didn’t feel as rewarding as being able to help clients achieve their individual goals. When we work closely with clients, it gives us the opportunity to invest in their success and find creative solutions that wouldn’t fit the Big Software model. In many ways it’s intellectually stimulating, and having a direct connection to our clients is rewarding.
It’s not just about the money. Microsoft treated a single unhappy client as a mere statistic. Every day, I would ignore complaints and continue to do my job. P3 adaptive is not satisfied with a single client. No statistics exist. Every relationship is important to us.
Microsoft taught me many things and I wouldn’t change it for anything. But I still refer to myself often as an employee. “recovering software engineer,” Because success today means operating in a very different way.
That’s the exact lens that I use when it comes to enterprise AI. Off-the-shelf AI is the ultimate 300-million-person pizza – a genuine marvel, engineered to be individually valuable to everyone while tailored to no one. But It is a good way to get organized. AI wins will come from the tailoring – from getting close to one specific company and customizing the AI to You can find out more about it here. data, You can find out more about it here. processes, You can find out more about it here. definitions. My career took me to both sides, which left me in no doubt as to the winner.
You can also find out more about the following: Fair GameYou argue that companies are approaching artificial intelligence the wrong way by issuing general-purpose licenses for chatbots, rather than creating systems that can understand their business operations. What are the limits of off-the shelf AI assistants, and how can a company tell if it needs something more customized?
AI that is available off-the-shelf has an MBA in all but your business. Your company is not represented on the Internet. “active customer,” Your pricing logic, operational processes and the systems you trust to resolve disagreements. The information you have will never become public. It’s frustrating to see that the AI for businesses is not as good as the one for individuals.
The answer is a simple yes or no. “what to do about AI” The term “Has been” has been used to describe the time when the word “buy subscriptions and find out.” It seems like a logical first step. I don’t criticize anyone for doing it. Instead I am sympathetic – no one is really taking the time to explain that the off-the-shelf subscriptions aren’t enough, nor why. So I think that businesses are basically right where we should expect them to be – trying the thing that’s available and starting to learn that it’s insufficient.
The fix isn’t touching the AI model itself – you don’t have to become an LLM researcher. The model is only as good as the data you provide, plain English instructions, and software that’s available. This is the sign that you’re typing the exact same text into the chatbot for the 5th time in a week. Whatever you keep re-explaining is exactly what a custom system should already know – every time it wakes up.
When you use the term “Crafters” Data-savvy professionals can create custom AI without the need to be traditional software developers. How can managers identify Crafters within their workforce and what characteristics make them a crafter?
The Crafter, as the name suggests is someone who was born to use tools. My experience has shown that roughly 1 in 16 knowledge-workers have this condition. There were Excel power users and Power BI people, followed by IT professionals. “shadow IT.” They’re your analysts, your finance modelers, your ops leads – people who grew up in the business and discovered a knack for tooling.
They are ideal candidates for AI projects because they possess two traits. As professionals, they are able to decompose complex processes into simple inputs, outputs and rules. The second is a firm grasp of business. They know what numbers the CFO watches, and the real question being asked. Both of these skills cannot be taught in a bootcamp.
Find yours by following the spreadsheets. There are many spreadsheets, dashboards and automations that your business uses to manage critical processes. They were not built by IT and they all have an author. Begin there. Then, start to assess how you can use their talent for customized AI solutions.
What are your thoughts on why you think Crafters rather than only developers should be in charge of many AI internal projects? How can responsibilities best be distributed amongst business experts, IT departments, data teams and software engineers as well as security teams and information technology departments?
Because the hard part of custom AI isn’t code – it’s context. In an AI project, the single most important activity is to decide what your system should know about you and your company. Crafters have this knowledge in-house. The brilliant engineer that parachutes in three org chart hops away will have to interview your Ops lead for several months before he can learn the information.
But this is emphatically not a developers-are-obsolete story. My recommendation for dividing labor is based on three factors. Seniority and personality are not among them. The work tends to gravitate toward professionals as developers. reusability, ComplexityThen, Sensitivity rise. Anything customer-facing, anything touching sensitive data, anything making autonomous decisions – that’s developer territory, and as agents multiply, those scarce engineering skills become more valuable, not less. The work gravitates towards Crafters, where the nuance of business processes dominate.
Crafter builds are audited by developers. IT and security shouldn’t be gatekeepers who approve projects into existence – they should own the paved road. Let the closest people to the issues do the construction. Provide them with the approved platforms, rules of data access, and review checkpoints. The whole process should be treated as a maturation model rather than a fence.
AI must have access to specific company terminology, metrics, business processes and institutional knowledge. How do existing business intelligence and semantic models help AI gain a better understanding of the company?
The decoder rings are the key. Right now your company’s definitions – what counts as an active customer, which costs belong in gross profit – live in people’s heads and in a thousand slightly inconsistent spreadsheets. AI agents cannot reliably analyze your data until these definitions are in a reliable format. It has become a discipline in the AI industry. “context engineering,” It’s the process of structuring your company’s knowledge so that an AI could actually utilize it. It was made to sound like a new concept by the analysts. BI practitioners use a form of this for 15 years.
The good news is that you may already have a leg up if your investment in Power BI was made in the BI era. The machine-readable representation of meaning is what agents require. Companies who treated semantics as a secondary layer are now discovering the importance of it. “boring” Tolls on AI’s road are now charged for definitional work, which they skipped. And critically, this work is deeply specific to your business – which is precisely why it’s the durable advantage. Each vendor will sell the same model. You cannot sell yourself.
Eddie was the name of your custom AI editor that you created to aid in developing Fair Game. What exactly did Eddie do in the course of writing, and how did you learn from its success and failures about creating AI that works with a personal workflow?
Eddie waited and mostly watched as I composed every sentence of the book. Sometimes I would spend hours writing a whole chapter, before even asking. “him” You can read it. At other times, I’d bounce ideas off him every couple of minutes. Eddie is always available. It was easy to get his feedback, even at 3 am. Eddie probably read through the entire manuscript thirty or more times. The job was impossible for a person to do, since no one would have wanted to. want it.
He reminded me of promises made to him in Chapter Three when I forgot about them. He learned my writing style and then enforced it – holding me to the best version of my own voice instead of letting me drift into Humorless Business Author mode. He pointed out when I was laziness and when I was wasting my time. He won some of our disagreements.
Eddie’s design is the biggest lesson. “brain” It is in English, and it lives in a file. Every time he gave feedback that missed – too generic, wrong register, forgetting a rule I’d already stated – the fix was to write the correction down and make it part of his permanent context. It wasn’t that AI failed, it was just that I hadn’t taught him enough. That loop – notice the miss, encode the lesson, watch it stick – is the entire craft of custom AI in miniature. This is why I created specialized Eddies, for website messages, competition research and publicity. Same LLM underneath. But different specialists.
Before implementing custom AI, many organizations feel they have to clean up and centralize all their data. How ready is data to get started, and can businesses start generating value before completing the foundation?
The following are some of the most effective ways to reduce your risk. The perfect way to improve your life? It’s a good thing that perfection is never achieved. Many consulting firms will tell you to build the perfect data estate first. This is what I refer to as a “perfect” system. “plumbing for its own sake” – expensive pipes running everywhere, but when you finally get around to installing a faucet, you find that there’s no pipe where you need it.
We are a company that advocates for a “faucets first” approach. Choose a use case, and then work backwards rather than working forward from the infrastructure. Create an MVP based on the use case and with minimum new infrastructure. Then, iterate until the MVP is production-ready and step back to evaluate whether you can harden your existing infrastructure. That delivers business impact more quickly, minimizes cost, and informs future projects – at both the faucet and the plumbing levels.
Custom AI prototypes can look impressive in a demo but be unreliable once they are exposed to actual employees, data changes, or edge cases. Before an AI system is operational, what evaluation, monitoring and human oversight are required?
Demos have less value in the AI age than in the previous software age, with a few notable examples. All of us knew software demos were over-promising. However, AI demos may be further from the reality of your life.
AI is workflow. The thousands of custom workflows that drive the organization’s operations are the most customized. Return to the “new hire with a PhD in everything” metaphor. How much training – and hands-on experience working at your company – does a new hire require before they are effective at your company? What can be accounted for by a demonstration?
Demonstrations are used to stimulate people’s thinking. They are shown the possibilities. It’s not about selling them something. A prototype of your custom solution is the first step in the real demo. It is the MVP. Then we improve and iterate. Rapidly.
It will be ready at some point for a pilot or soft launch. And again, we learn – together – and rapidly improve based on that learning. During this phase, monitoring, evaluation and oversight become more apparent. You may find that the things you actually need are very different from your initial expectations.
How can Crafters be empowered to experiment by companies without creating shadow AI, redundant workflows, vulnerabilities in security, or tools that no one is accountable for maintaining?
Shadow IT was not malicious, but rather the result of unmet needs. Crafters build because problems bother them – that’s the gene. If the sanctioned path means waiting a year, shadow AI will fill the gap – and fill it below the radar, where it’s most dangerous.
Make the lane that is sanctioned the easy lane. Give Crafters an approved platform with the security guardrails already baked in – identity, data access, logging – so the compliant choice is also the convenient one. A lightweight registry is a good idea: Anything from an experiment that becomes something on which another person can rely, should be recorded with its owner. The orphaned tool problem is largely eliminated by this rule, as tools that have names are not abandoned silently.
Then apply the escalation model: experiments run free, but as something becomes mission-critical – more users, more sensitivity, more autonomy – it earns progressively more engineering review. Crafter owns the business logic. A developer will harden what needs to be hardened. Not a permissions process, but a maturity pipe is the goal. The same thing happened with spreadsheets. And the people who won were not the ones that banned Excel.
What should be the first use case for a custom AI initiative? And how can a company decide whether it wants to continue, expand or redesign it once it has determined that it is providing meaningful business benefits?
Two different starting points are used with clients.
Option one, look for the jobs nobody’s doing – not the jobs you’d like to eliminate. It’s one of my favorite questions to ask managers: Where have you thought? “if I had one person constantly watching this and thinking about it, things would get meaningfully better – but I could never justify a whole hire for it”? These are usually the best places to start. These are safe and build confidence. Nobody feels targeted, nor is anyone a threat. The counterfactual, on the other hand, was that no one did it (like Eddie, my editor friend).
The second option is to replace dashboards with agents that provide data. Dashboards, however simple they appeared in theory, failed to deliver on their promises. It’s difficult to convert a question about a company into a dashboard. What is the best way to get in touch with you? What is it? What exactly is the dashboard? Name? What is the purpose of such a dashboard? There are many ways to get in touch with us.? If you find it, then that’s a good thing. “right” Is it easy to understand and use? You may have to manipulate the image repeatedly, capturing screenshots or writing them down to get an overall view.
In the AI era, you just take your business question – in your own words – and type it (or dictate it!) to a data agent who then handles all of that for you, and returns a certified, well-researched answer – visuals included – in a minute or two. When you have a follow-up question, it’s happy to quickly answer that, too – in the meeting while decisions can still be made.
What is the common theme behind these two starter options? The two options both target pain points which will be embraced by employees rather than resisted. Early AI initiatives shouldn’t cause distrust. Instead, you want to bring your employees into the conversation. Your employees should suggest new projects and improvements. You have thousands of different workflows in your company, but your employees will know these better than you.
On expand, redesign, or abandon – be kind to yourself, because the research on this is genuinely comforting: most successful AI deployments had failures before them. The fact that a first project produces more lessons than rewards is not proof that AI does not work. If people use it, then expand it. You need to find out why people don’t use it. The reasons can range from “because it doesn’t work well” The following are some of the ways to get in touch with us: “because I don’t understand it” The following are some of the ways to get in touch with us: “it frightens me.” Answering this question will determine whether or not you should redesign, improve, or give up. It’s not necessary to know where everything will end up. Just start with the truth.
You should read the interview. Fair Game: Customizing AI to Your Business Is Easier Than You Think.
