Use it where it Belongs

There is a version of AI that works almost every time.

Ask it to summarize a contract, extract key terms from a document, generate a product description, or organize a listing of attributes. It delivers. Fast, accurate, genuinely useful. Most people who use AI daily have found this version and built real efficiency around it.

Then there’s the other version.

A senior retail and CPG executive draws a sharp distinction between what he calls straight-line tasks (visual, textual, descriptive, linear) and the complex, multi-variable environments that define most of the real work in big retail and CPG. Product assortment decisions. Pricing interactions with promotions. Digital signals layered across a fragmented media landscape. In these environments, he has seen AI produce recommendations that were nearly the opposite of correct.

Practitioners are learning this the hard way. A senior analytics leader at a major global BevAl company describes going at pace early, getting stuck, and having to reverse course entirely. The lesson she drew: build the data infrastructure first, then define the problem you are solving, then move. The AI meant nothing without that foundation and died without it.

Another senior global intelligence executive at a major CPG company frames it this way: everyone wants the agent, everyone wants the toy. But if you skip the semantic layer and the knowledge graph underneath it, the model will hallucinate, misfire, and erode trust in every output that follows.

The foundation is not sexy. It is also not optional.

The problem isn’t the technology. It’s the deployment. AI performs well when the path from input to output is relatively direct. It struggles, sometimes badly, when the variables are numerous, the data sources are disparate, and the domain knowledge required to interpret the output isn’t baked into the model.

Most organizations are deploying AI as if all problems are straight-line problems.

They aren’t. Knowing the difference is not a technology decision. It’s a business judgment that has to happen before the model is built, not after the wrong answer lands in a leadership deck.

Use AI aggressively where it wins. Be disciplined about where it doesn’t.