We Use Our Own Tools. Now We Are Ready for Yours.

Posted on November 20, 2026

0


A year, more than 2,000 hours, and a public record. Our challenge to the market.


In 2004, Hewlett-Packard moved one of its largest North American divisions onto a centralized SAP system. The migration went badly. HP’s chief executive put the financial impact at $160 million, several times what the project itself cost. At the same time, HP was selling its services as an SAP implementer. Analysts asked the obvious question: how do you sell clients a capability you could not make work for yourself?

That question has come back.

Deloitte delivered a government report in which an outside researcher found unsupported references and a fabricated quotation, after it was published. Recently, Accenture published an AI paper containing hallucinations and called it human error. In both cases, the AI was not the failure. The release process was. Nothing in the review stopped the errors before they reached the client.

These are firms that sell AI transformation. Their own publications could not get through their own review.

What we did instead

For more than a year, and over 2,000 hours, every Procurement Insights post, every Hansen AI Labs™ lab and every assessment we have produced has gone through ARA™ RAM 2025™ running on SLAP OS™.

That means the same question goes to multiple AI models from different makers at the same time. A separate Shadow Panel’s only job is to find what a critic would attack. Claims are checked against their primary sources. The model that writes an expectation never scores it. And a human, me, makes the final call with every objection on the record.

We keep the full transcript of every exchange with every model, including the disagreements, the challenges and the corrections. Nothing is summarized away. That record is how we refer back to the reasoning behind any result, and how anyone we work with can see where the human stepped in and what it changed.

Then we publish the result as it stands. It is fingerprinted and never edited, so anyone can check whether it changed. We let the world decide whether we got it right.

Hallucinations, inaccuracies and drift still happen. They happen in every model. The difference is that our process catches them, before publication or on the record afterward.

Here is a small example from today. One AI model graded the same post A+ inside a long-running conversation that knew our work, and B- in a fresh chat that didn’t. Most people would have kept the A+. We went looking for the cold read, recorded the B-, and learned something about how these systems drift toward the people using them. The same afternoon, a reviewer’s claim about a research paper was checked against the paper itself and corrected before it went anywhere.

That is what we mean when we say we use our own tools. If you can’t make a method work for yourself, you can’t make it work for anyone else.

The receipts

This did not start with AI.

  • 1983. Programming in dBASE II, I noticed that changing one line of code could break another line that seemed unrelated. That was the first time the pattern crossed my mind.
  • 1998. With partial funding through Canada’s SR&ED program, I formulated that observation as a theory and applied it for the parts operation serving Canada’s Department of National Defence. Next-day delivery went from 51 percent to 97.3 percent within three months, against a 90 percent requirement, and held for seven years. The architecture is documented in a published international patent application, WO 01/65428.
  • 2007 to today. Nineteen years of Procurement Insights, dated and public.
  • 2026. A year of multimodel work on the record, and four live labs that lock their expectations before the outcome is known and score them afterward.

What the market is saying

For years, agent-based modeling, Metaprise and measuring outcomes instead of activity drew skepticism. Now those ideas are coming back to us in other people’s words.

Saurabh Mishra, who helped create the Stanford HAI AI Index, told me, “You have a better idea of what I’m doing than I do.” Michael Lamoureux of Sourcing Innovation, never shy about criticism, now places our readiness and fit models in the buying sequence. And today, Chandrakumar Pillai, an AI governance advisor in Brussels, answered our argument about AI agent supervision this way: “Independent checks plus an accountable human is the right shape.”

A current client put our principle back to us better than I could: “We don’t want to be right; we want to get it right.”

We also have references, people who saw the work produce results, who will take your call directly. Ask, and we will put you in touch.

Why it works

The integrity of AI comes down to the human orchestrator. There is no other way around it.

AI models follow the framing they are given. A person who sets out to prove a theory will get an echo. A person who sets out to get it right will get a test. Most AI governance discussion asks whether there is a human in the loop. The better question is what that human is trying to do: confirm, or challenge.

I have spent 43 years learning what happens when technology meets real organizations, and the last year putting the leading AI models to work against each other on live problems, on the record. The people at the AI companies build the engines, and they know things I don’t. But on the question that decides whether your AI investment pays off, how AI performs in the real world and how to make it work, I will put my experience, and the consistent results this method has produced over 12 months and more than 2,000 hours, against anyone’s.

That second part matters more than the first. If the results depended on me alone, they would end with me. They don’t. The method is documented, it runs on a system, and every step of it is on the record, which is what allows it to be taught, run by others and scaled well beyond one person.

Our challenge to the market

We are not asking you to take our word for it. The record is public. The references will take your call. The labs are running now.

We are ready.

If you’re ready to get it right, we’re ready to help you do it.

Contact: HPT@hansenprocurement.com

-30-

What a Hansen AI Lab is

A Hansen AI Lab is a real-world test, run on a live event while it is still unfolding. Before the outcome is known, we name the strands we expect to decide it and lock our expectations in a dated, fingerprinted record. Then we follow the event, measure what actually happens, and score every expectation against the evidence, hits and misses alike. Nothing is edited after the fact, and every exchange with every model is kept.

The labs are not case studies written after the result is known. They are the method working in public, on situations nobody controls, scrutinized for accuracy and for the value of the guidance they give a decision-maker while there is still time to act.

These are live labs, not experiments in a hermetically sealed environment. We went out into the world and put our reputation on the line. Using ARA™ RAM 2025™ running on SLAP OS™, we fingerprinted and published what we were tracking, why we were tracking it, and the outcomes of that tracking, whatever they turned out to be.

HP couldn’t make its system work for itself, so it couldn’t credibly make it work for anyone else. We have made ours work for ourselves, in public and on the record.

These are the four labs on the record so far:

Table titled "The labs behind the record," listing four Hansen AI Labs: the Isaias Strand Lab (opened 10 October 2026, completed 12 October), the Isaias Reverberation Lab (opened 17 October, running with weekly sweeps), the Cold Chain Strand Lab (opened 4 November, reckoning 16 December) and the FAA Ground Stop Strand Lab (opened 16 November, awaiting the FAA's findings).

Across the labs, the system’s output accuracy, meaning what the labs put on the record checked against the evidence as it came in, is holding at an average of around 91 percent, and it fluctuates. Our prediction record is tracked separately, because the misses and what caused them teach us as much as the hits do.

Lab records are available to qualifying organizations under NDA. Contact us at HPT@hansenprocurement.com.

Posted in: Commentary