“AI gave us the wrong answer.” Did it? Or did we give it half a picture?

Or did we give it half a picture? Most enterprise AI failures have nothing to do with the model that produced them. They are caused by missing business context.

You ask a new hire on day one: "Why did our biggest customer stop buying?" What you give them: sales records. What you withhold: contracts, support tickets, shipment history, production issues.

You wouldn't trust their answer. Don't trust the model's either.

ERP holds finance and orders. CRM holds customers. HR holds employees. Supply chain holds suppliers and shipments. Documents hold contracts and terms. Email holds the conversations around both.

AI sees fragments. People connect them from memory.

Ask it why revenue declined and it can answer with reduced demand, seasonal changes, or market competition — all reasonable. What actually happened: a supplier shutdown, production ran late, orders shipped late, customers held payment.

The failure wasn't reasoning. It was missing evidence.

Set-up one: the model, one sales database, general knowledge, a best guess. Set-up two: the same model, connected systems, business relationships, an evidenced answer. Nothing about the model changed. Everything about the input did.

Most programmes spend their first six months choosing the right model, writing better prompts, fine-tuning the responses — and skip the larger question: can it actually understand our business? A brilliant model with partial context is still guessing.

Enterprise AI shouldn't answer questions. It should explain them: what happened, why it happened, who it affected, what happens next, and what should happen next. Reliability doesn't come from bigger models. It comes from better enterprise understanding.