Who to interview when you are scoping an AI feature
A product lead scoped a feature from the sponsor's description alone. It was wrong in ways only two other people could have told her. Here is who to talk to.
Who to interview when you are scoping an AI feature
A product lead at an HR-tech company scoped a feature the way most people do. She talked to the executive who asked for it. He described what he wanted, clearly and confidently, and she wrote it down and built toward it. The feature was supposed to draft candidate outreach messages that recruiters would send. He knew exactly how it should sound and what it should say.
It shipped, and it was quietly useless. Not broken. Useless. The recruiters rewrote almost every message before sending, and the hiring managers who received the resulting replies could not tell the AI had touched anything, because by the time a message went out the recruiter had redone it. She had scoped a real feature from a real request and still missed, because she talked to the wrong person. Or rather, to only one of the right people.
The sponsor knows why, not what
The sponsor is the person who requests and funds the feature. They are worth talking to. They own the reason it exists, the outcome the business wants, the money behind it. What they usually do not own is the texture of the actual work. The executive wanted “great outreach messages.” He had not personally written a recruiting message in years. His description was a goal wearing the costume of a specification.
There are two people who own the texture, and you have to interview both.
The first is the person who does the work. In this case, the recruiter. She knows why the sponsor’s clean description falls apart in reality. She knows that the message is different for a passive candidate than an active one, that certain phrasing gets flagged as spam, that she personalizes off a detail in the profile that no template captures. The doer knows the exceptions, the shortcuts, and the reasons the work is not as simple as it looks from above. Scope without her and you build the sponsor’s imagined version of the job.
The second is the person who lives with the output. The one downstream who receives what the feature produces and has to act on it. For the outreach feature, that is partly the candidate and partly the recruiter’s own reply-handling. Whoever consumes the output defines whether it is good, because they are the ones who feel it when it is not. If you never ask them what “good” means, you will optimize for the sponsor’s taste and discover too late that the consumer had a different bar.
Three roles, three different truths
So the roster for scoping is at least three people, and they are rarely the same person.
- The sponsor tells you why the feature should exist and what outcome justifies it.
- The doer tells you what the work actually involves, including everything that makes it harder than the sponsor thinks.
- The consumer of the output tells you what “good enough” means, because they are the one who lives with it.
When all three are one person, scoping is easy and you are lucky. Usually they are three people who have never compared notes, and the gaps between their versions of the truth are exactly where your feature will succeed or fail.
An operations manager at a healthcare provider learned this the hard way with an AI feature that summarized patient intake for clinicians. He interviewed the clinicians, the consumers, thoroughly. He never interviewed the intake staff, the doers, so he did not learn that half the important information arrived by phone and never made it into the system the model read. The consumers told him what good looked like. Only the doers could have told him the input was incomplete before the model ever saw it.
Make the interview list before the spec
The practical habit is small. Before you write a line of the spec, name the three roles for this specific feature. Who asked for it. Who does the work today. Who receives and acts on the output. Then go talk to a real human in each role, separately, and listen for where their descriptions disagree. The disagreements are not noise. They are the scope.
How we approach it at Density Labs
In the AI Opportunity Assessment, our fixed two-week engagement at $2,500, we insist on talking to the doer and the output’s consumer, alongside the person who signed the engagement. The sponsor’s version of the work is where projects start. The doer’s and the consumer’s versions are where projects survive contact with reality. Reconciling the three is most of what those two weeks buy you.
Scope from the sponsor alone and you build a confident answer to a question only one of three people was asking. Interview the work and the output, along with the request.