Start With the Decision, Not the Tool
Most AI projects begin the same way. Someone sees a demonstration, the company buys licenses, and only then does anyone look for something useful to do with the tool. That puts the starting point in the wrong place.
Ruru Hoong, a marketing professor at MIT Sloan, argues that AI adoption is “as much a problem of organizational and information design as it is one of technology procurement.”¹ Her advice is to begin with four questions: which decision you want to improve, which errors matter, what your people know that the AI does not, and how the AI’s answer should reach the person using it.
The decision is the right place to start because the people making it shape the result as much as the software does. How they use the AI’s answer, alongside what they already know, decides whether it helps. And while a tool is chosen once, a decision is made hundreds of times a year.
This approach fits AI that helps someone make a call, such as a ranking or a recommendation. Tools that mainly save time, like drafting proposals, are a topic for a future article, although many of them contain a decision, such as which email needs a reply first.
Take the weekly question of which open quotes your sales team should follow up on, and apply her four questions.
First, which decision are you improving? It is the weekly follow-up list, not “sales” in general.
Second, which errors matter, and what does each one cost? Chasing a dead quote costs a few hours. Missing a live one costs the margin on a job. Only you know the economics, and the tool should favor the cheaper mistake, even if that means a longer list. Where exactly to draw that line deserves its own article.
Third, what do your people know that the system does not? Your estimator heard on Tuesday that the customer’s project is delayed. Your system does not have that information, so the list must leave room for people to override it.
Fourth, how should the answer reach the person using it? A short ranked list in your CRM on Monday morning will get used. A separate dashboard will not, as I explained in AI Belongs in the Workflow.
These questions change the conversation with whoever is selling you the tool, whether that’s your CRM provider, a specialist AI firm or an implementation partner. The conversation moves from what the product can do to what the decision requires. No vendor can answer these questions alone, which keeps you and your people in charge.
Before you approve your next AI purchase, ask the people in the room which decision or piece of work it will improve, and roughly by how much. Measuring AI Performance describes one way to put a number on it. If nobody can answer, the project is not ready to fund yet.
Sources
¹ Sara Brown, “What 3 new MIT Sloan professors have learned about AI,” AI at Work newsletter, MIT Sloan School of Management, September 10, 2026 (comments by Ruru Hoong, assistant professor of marketing).