The scope creep that starts with ‘while we’re at it’

An engineering manager at a proptech company scoped an AI feature that started clean. It would read a rental listing and extract the key facts into a structured form. One input, one output, a bounded job. Everyone understood it. Then the meetings happened.

While we’re at it, could it also flag listings that look like duplicates. And while we’re at it, could it detect the ones that might be fraudulent. And since it is already reading the text, could it rewrite the description to sound better. And could it translate that for the second market too. Each request arrived friendly, each one sounded small, and each one came riding on the phrase “while we’re at it.” By the time he noticed, the tight extraction feature had four passengers, and the project that was going to take weeks was going to take a season.

Every addition multiplies more than the code

The reason “while we’re at it” is dangerous is that the visible cost, a bit more code, is the small part. The hidden cost is the surface you now have to understand and check.

Every capability you bolt on has its own discovery. Duplicate detection needs its own definition of what counts as a duplicate, its own edge cases, its own data. Fraud detection needs a whole risk conversation, examples of real fraud, a decision about false positives, probably a compliance review. Rewriting descriptions needs a quality bar and a consumer who defines it. Translation needs a second language’s evaluation. Each of those is not a feature-and-a-half. It is a new problem with its own interviews, its own examples, and its own way of being wrong.

And every capability you add has its own evaluation. This is the part teams forget. You cannot ship an AI feature you cannot test, and each new job needs its own test set, its own accuracy target, its own watch in production. Four “while we’re at it” additions do not make the eval four steps longer. They make it four separate evals, each with its own labeling effort and its own failure modes to monitor. The surface you have to prove correct grows with every passenger, and it grows faster than the code does.

The additions also blur the answer to “is it working”

There is a quieter cost. A feature that does one thing has one clear measure of success. A feature that does five things has five measures, and they tangle. When the combined thing underperforms, you cannot easily tell which of the five is dragging it down. The clean signal you had at the start, this extraction is accurate or it is not, dissolves into a fog where nothing is clearly good or clearly broken. Debugging a bundled feature is harder than debugging five separate ones, and you gave that clarity away one friendly suggestion at a time.

Guard the scope out loud

The defense is not to say no to every idea. Some of those additions are good and worth building later. The defense is to name what is happening when it happens. When someone says “while we’re at it,” treat it as a proposal for a new project, not a tweak to this one. Ask what its discovery costs, what its evaluation costs, and whether it belongs in this build or the next. Usually the honest answer is that it is a real feature that deserves its own scoping, not a free ride on this one.

A product owner at a fintech told me she keeps a literal “while we’re at it” list in every scoping session. Every add-on someone suggests goes on it, and nothing on it enters the current build without going through the same discovery the original feature did. The list does two things. It makes people feel heard, because their idea is captured, not dismissed. And it protects the shipping thing from quietly becoming five shipping things that never ship.

How we approach it at Density Labs

In the AI Opportunity Assessment, our fixed two-week engagement at $2,500, part of the job is drawing a hard line around the one feature actually worth building first and parking the rest, explicitly, for later. We count the discovery and evaluation surface each proposed capability adds, because that surface is what turns a tight build into an open-ended one. A scoped feature that ships beats a bundled feature that stalls, every time.

The phrase “while we’re at it” is where clean scope goes to die. Treat each one as a new project, not a small favor, and most of them will wait their turn.