2026-09-10 · article
Before building an AI tool, find where the money is
A frustrating step is not automatically the valuable one. Check where time and cost actually sit before funding a new AI tool.
An AI feature can make an annoying task feel easier and still be the wrong investment. The useful question comes before the solution: where do people actually spend time, money and attention in the work around it?
Start with the full job
Take the work from its first input to the thing the customer receives. Name the stages, who owns each one, how long it takes, and what happens when it goes wrong. The painful step is often easy to point at because everyone remembers it. That does not mean it carries the largest cost.
When considering a product for bid workflows, I examined whether faster qualification was the part worth selling. The research pointed toward bid preparation as a larger cost to investigate. That changed the question: how much of the total cost could a faster first decision actually remove? This was a product assessment, not a measured saving from a customer deployment.
Check the alternatives before the model
The first answer may be existing software, a simple rule, a clearer handoff, or stopping a piece of work that has no return. An AI build is worth investigating when it changes a costly constraint and the people who carry that constraint can use the result in their real workflow.
- What decision or output gets better?
- What does the current version cost in time, errors or missed revenue?
- What information is available at the point of work?
- What would make a simple alternative sufficient?
Leave uncertainty in the recommendation
A good recommendation says what is known, what is assumed, and what has to be tested next. It does not promise that a prototype will solve a problem nobody has measured yet.
That is the useful part of an external assessment. It gives a team a path they can take forward, including the answer that a build should wait.