What Is Free-Range AI™ for Supply Chain, and Why Should You Care?

I spent 42 years running supply chains, the last several as Chief Supply Chain Officer at Clorox. I sat through more AI vendor demos than I can count, and almost every one of them used the word “intelligent”.  Almost none of this so-called “intelligence” enabled the vendor to tell me what would happen if I asked about a scenario they had not prepared for in advance.

That gap is the whole story.

Most of what gets called AI in supply chain planning today is scripted. It works inside the boundaries of what someone configured for. Ask it something inside those boundaries, and it looks impressive. Ask it something outside of them, during a real disruption, and it goes quiet or gives you an answer built on assumptions frozen at implementation. I did not initially understand this distinction. I judged AI by the demo, and the demo is built to show you exactly what the vendor prepared for you to see. Free-Range AI™ is different, and it is worth being precise about what that actually means, because “different” is not a claim. It is an architecture question, and there are only two answers: bolted on, or built in. Chris Amet, ketteQ’s CTO, and I are writing a longer paper on that question called Bolted On vs. Built In, which will be available soon. This is the short version, from the buyer’s side of the table.

Three drivers led me to serve as Chairman of ketteQ’s Executive Advisory Board. First, it is the only AI I have evaluated that meets four Must-Have requirements. Second, the PolymatiQ™ agentic solver runs thousands of scenarios simultaneously rather than one at a time. Third, Quintus™ is built in, not bolted on, from day one. I do not say that as a paid endorsement. I say this because I looked for built-in architecture before putting my name behind anything, and ketteQ is the only place I found it.

The four Must-Have requirements are as follows - and I wish I had known to ask about all four in every vendor conversation I ever had.

1. It must be purpose-built for supply chain on an AI-native architecture from the ground up. Not a general-purpose model retrofitted for our industry. Supply chain has its own language, constraints, and decisions, and an AI that does not already understand it cannot reason over it, no matter how good it looks on stage.

2. It must be adaptive end-to-end. It needs to reason across the whole chain, supplier to customer, and rewire its own logic in real time as conditions change, not sit inside one module working from a static snapshot.

3. It must solve in real time. I have watched planning engines run one scenario overnight and call it analysis. A solver that takes hours cannot power a recommendation, and a real-time recommendation is exactly what you need when your plant goes down at 6 a.m., and you need an answer in an hour.

4. It must run on your existing planning systems. Most of us don’t have the appetite for a multi-year, risky planning platform replacement. If the intelligence only works on one specific stack, it is not free-range. It is fenced in, just like everything else on the market.

Miss any one of those four, and what you are looking at is scripted AI, bound up inside its own legacy architecture while dressed up as Free-Range.

Now, why should you care? Because I lived through what happens when you do not have this.

I have watched this movie before, just in different theaters. ChatGPT and Claude did not make software engineers, support teams, and marketers slightly more productive. They broke the ceiling on what one person could do in a day, almost overnight. Supply chain has not had that moment yet. I believe Quintus™ is it, and most CSCOs have not noticed, because the AI they have been shown so far was “fenced in” and never built to break the ceiling.

During COVID, my team needed to replan constantly as conditions changed hour by hour, and we did it through sheer manual effort. In 2023, Clorox endured a five-month cyberattack, and I watched what happens when every layer of intelligence sitting atop a compromised system goes dark right along with it. You cannot separate the AI from the architecture it lives on. When the architecture fails, the AI fails with it.

The moment you need real intelligence most is the moment conditions stop being normal. That is precisely when scripted AI has nothing to offer, because it was never built to handle what it was not configured for.

I sat through all those AI evaluations without knowing how to ask the one question that mattered: how is your AI actually connected to your planning system? That is the real test. AI that is merely connected to your planning system cannot operate as successfully as AI that is embedded in it.  

Ask that question now. Next demo, ask the vendor to run an unscripted scenario. If you are a CSCO, ask: our top supplier just told us their plant is down for two weeks, which customer commitments are at risk, and what are our options? If you lead Sales, ask: can I promise this customer the full order by Friday, or should I split the shipment? Nobody on the vendor team is prepared to answer those questions. Watch what happens. If you get a fully reasoned answer in seconds, you have found real architecture. If you get silence, a caveat, or a report to review later, you have found a demo built on old architecture dressed up in new language.

Don’t wait – ask these questions the next time you get demo’d - the answers (or lack thereof) may surprise you.