The AI Decisions Are Wrong (And Your Data Isn't the Problem)

The AI Decisions Are Wrong (And Your Data Isn't the Problem) | James Taylor

Guest: James Taylor, Founder and Executive Partner of Blue Polaris (formerly Decision Management Solutions)

Host: Kumar Dattatreyan

Episode Date: June 16, 2026 Watch on YouTube

Why You Need to Listen to This Episode If you have sat through one more AI strategy meeting that ended with "we just need better data," this episode is for you.

James Taylor has been in the room for those meetings for twenty years. He helped create the decision management category at FICO. He co-submitted the DMN industry standard. He has watched companies pour money into predictive analytics, then machine learning, then generative AI, and watched most of those investments produce demos that impress executives and change almost nothing about how the business actually runs. He has a pretty clear idea of why.

It is not the technology. It is not the data. It is the question nobody is asking before the work starts: which decisions are we actually trying to improve?

In this conversation, James walks through the discipline he has spent his career building. Why decisions belong outside the process, not buried inside it. Why a language model alone cannot run a regulated business decision. Why "explainable AI" does not actually explain anything a regulator will accept. And the pattern he sees working in the wild, where AI handles the messy parts on either end and prescriptive logic handles the part that has to be auditable. By the end of it, you will probably start noticing missing decisions in your own organization that you have been walking past for years.

What You Will Learn in This Episode The Difference Between a Process and a Decision (And Why It Matters More Than You Think) Most teams build their systems as processes with decisions hidden inside as steps. James makes the case that this is the single biggest source of friction in enterprise software. Processes change slowly. They have lots of owners. They require retraining when they change. Decisions change all the time, often have one owner, and need to flex with the market. When you smear them together, every small change becomes a big project. When you separate them, the process gets dramatically simpler and the business gets dramatically more agile.

The California DMV Example That Settles the Whole Argument James drops one of those examples that you will use in your own conversations for years. The California DMV has had the same process for licensing vehicles since the 1950s. They send you a bill, you send them a check, they send you a sticker. Nothing about that has changed. What changes constantly is the math underneath, because politicians tinker with the fees every legislative session. Same process for seventy years. Different decision every year. This is what decision management is actually about.

The Four Conditions a Decision Has to Meet Before You Should Even Try to Automate It Not everything is a candidate for automation. James walks through the criteria he uses to figure out which decisions are worth the work: the timing has to be predictable, the decision has to be made often enough to justify the effort, the inputs have to be reliably available, and the outputs have to be bounded. If any of those are missing, the decision either stays with a human or needs to be redesigned before you can automate it. This is the filter most AI initiatives skip, and it is why so many of them end up in the slide-deck graveyard.

Why Explainable AI Does Not Actually Explain Anything This is the part of the conversation that should make every executive sit up. James was direct: explainable AI does not know how the model decided. It generates a plausible-sounding story about how the model might have decided. Regulators have figured this out. The CFPB has already said the rules do not change just because you used AI. Canada has legislation that requires you to be able to tell any affected customer exactly how a decision about them was made. A chatbot answer is not going to satisfy that. You need actual structure underneath.

The Architecture That Actually Works in Regulated Industries Here is where the episode flips the typical AI narrative on its head. James describes the pattern he sees succeeding at his clients. AI ingests the messy input — emails, attached documents, unstructured forms. A prescriptive decision engine runs the actual logic against that data. AI generates the explanation back to the customer or call center rep in plain language. To the user it feels like a chatbot. Under the covers, every step is logged, auditable, and defensible. You get the speed of AI and the accountability of a traditional system. This is not theoretical. This is what is shipping right now.

Why the Bottleneck Has Quietly Moved From Data to Decisions There is a moment in the conversation where James says something most AI buyers have not caught up to yet. The old constraint was that you had to digitize and clean your data before you could do anything useful with it. Large language models have largely solved that. They are good at reading documents. The new constraint is upstream of the data. If you do not know what you need to extract from those documents in the first place, the fact that AI can extract anything does not help you. The discipline that tells you what to look for is decision modeling. It used to feel optional. It is now the thing.

Why the Hardest Part of This Entire Field Is the Word "Decision" The closing exchange is the one that will stay with you. James explains that the single biggest obstacle to building this discipline inside organizations is that everyone already thinks they know what a decision is. Executives hear the word and picture strategic choices they make in boardrooms. James means the thousands of small operational decisions their systems make every day, often badly, often invisibly. The website that fails to recognize a returning user. The claim that goes to manual review when it did not need to. The loan that gets declined for a reason nobody can articulate. These are the decisions where the value is, and they are the ones nobody is paying attention to.

Best Quotes from This Episode "My process doesn't change. They still got the same documents. They still go to the same people. They still look at them. They still check the same things. All of that's the same. All that changes is the decision making."

"It used to be for sure you had to digitize the data first before you could do anything else. And now large language models are really good at reading documents. So now what matters is do you know what data you need out of the documents?"

"Even if you look at explainable AI, mostly what it does is it comes up with a plausible explanation of how it might have decided. It doesn't really know how it decided."

"The people we see basically flailing around are the ones who are like, I'm going to throw the baby out with the bath water and start again with large language models. And we're like, why would you do that? You already know a lot."

"It's the D word, decision. On the one hand, everyone knows what decisions mean. The problem is that everyone knows what decisions mean. They think I mean the decisions they make. And generally I don't."

Connect with James Taylor LinkedIn: https://www.linkedin.com/in/jamestaylor/

Blog (JT on EDM): https://jtonedm.com

Blue Polaris: https://bluepolaris.com

Books by James:

Digital Decisioning: Using Decision Management to Deliver Business Impact from AI (MK Press, 2019)

Real-World Decision Modeling with DMN, with Jan Purchase (MK Press, 2016)

Decision Management Systems: A Practical Guide to Using Business Rules and Predictive Analytics

Smart (Enough) Systems: How to Deliver Competitive Advantage by Automating Hidden Decisions, with Neil Raden

Take the Disruptor Method Quiz Are you disrupting or about to be disrupted? Find out in under five minutes: https://www.thedisruptormethod.com/quiz

Work with Kumar Book a 30-minute conversation: https://tidycal.com/coachkumar/30-minute-meeting