How We Think About Trust in AI

The past few months I've been on the road a lot, meeting companies to show them MintIQ. The reactions have been overwhelmingly positive. It's clear to me that many product based companies are actively looking for the best, and least disruptive, way to start bringing AI into their business.

One thing I keep hearing, especially in production and distribution, is that people see their industry as conservative. And the question that follows is almost always the same: how do you build trust in AI in an industry like ours?

It's something we think about a lot. We had the same hesitation when we started building what eventually became MintIQ. What it came down to for us was a few things.

The first one is being very deliberate about where and how we use AI. When we do use it, we use it narrowly, trained on one specific decision. For example, deciding whether an order can go into a batch pick or not. Used that way, you get same input, same output. For things where you can't afford the latency of an LLM, like the validation that needs to happen in milliseconds when someone picks an item, we use deterministic code. And anything that touches the database, your inventory administration for instance, is 100% boring on purpose.

Second, your data stays yours. We don't train models on it and we don't share it. For a lot of the companies we talk to this is non-negotiable, and honestly it should be.

Third, humans stay in the loop where it matters. Not every decision needs a person looking at it, but the ones that do get routed back to your team. We're pretty deliberate about where that line sits, and we'd rather be too careful than not careful enough.

Of course no technology is perfect. Local systems had issues with service being down. So did cloud systems. Is AI always perfect? No. But the rate of improvement is astonishing, and with the guardrails around it, it works.

You can read more about how we use AI in our platform here: How MintIQ uses AI