The pattern
Most AI initiatives do not fail in production. They fail earlier, in the weeks where a company decides what to build. The demo looked great, the vendor sounded confident, the budget got approved, and six months later there is a chatbot nobody uses while the process that actually bleeds money is still run by hand.
We have watched this pattern from inside enterprise operations for over a decade, and the discovery mistakes are remarkably consistent.
Mistake one: starting from the demo, not the ledger
The question that opens most AI conversations is "what could we do with AI?" That question generates demos. The question that generates returns is different: where does this operation spend skilled hours on work a system could carry? The first question starts from the technology and searches for a home. The second starts from the cost structure and searches for leverage.
Mistake two: ignoring what the data can actually support
Every AI capability rests on the data underneath it, and discovery is where that reality check belongs. If the source documents are inconsistent, if systems disagree about the same fact, if nobody can say where a number came from, that is not a reason to stop. It is the first thing the roadmap has to fix, and finding it in week two costs a fraction of finding it in month six.
Mistake three: nobody owns the "no"
A discovery effort that only produces green lights is a sales document. Real discovery kills ideas: the use case with no path to payback, the automation that would create more review work than it removes, the model that cannot be audited in a regulated workflow. The most valuable line in an assessment is often "don't build this."
What a working assessment looks like
- Weeks, not quarters. Two to four weeks inside the operation: workflows, data, systems, and the people who run them.
- An opportunity map ranked by payback, not by how impressive the demo would be.
- A reference architecture that respects the systems already in place.
- A roadmap the company can execute with anyone. If the plan only works with the firm that wrote it, it is a brochure, not a plan.
Discovery is not the paperwork before the real work. Done honestly, it is the part that decides whether everything after it pays.