Most advice about doomed AI projects is written for people evaluating a project before it starts — a pre-launch checklist of red flags to watch for. That's useful, but it misses a more common situation: you're already inside the project. It's been running for months. There was a kickoff, a budget, maybe even a demo that got applause. And yet, if you're honest with yourself, something feels off. The question isn't "should we start this?" anymore — it's "is this one already lost, and I just haven't admitted it yet?"
Here are the tells that show up from the inside, after a project is already underway — not before it starts, but while everyone involved is still describing it as "on track."
The success metric keeps quietly changing
Early on, the project had a specific target: fewer support tickets, faster processing time, less manual data entry. A few months in, that target has softened into something nobody can quite point to — "adoption," "engagement," "it's helping." Nobody announced the change. It just happened, meeting by meeting, as the original number stopped being hit. When a project's definition of success keeps drifting toward whatever is easiest to claim, that's not progress — that's the project quietly renegotiating with reality.
The demo is always great; production use is always "coming soon"
There's a specific kind of AI project that's excellent at demos and strangely absent from daily work. Someone can always pull it up and show a clean, scripted example. But ask how many real transactions it handled last week, or how many people actually opened it without being prompted, and the answer gets vague fast. A tool that only performs well when someone is driving it for an audience isn't a tool yet — it's a prototype wearing a launch announcement.
Questions about the data get redirected, not answered
If you ask where the underlying information comes from, how current it is, or who's responsible for keeping it accurate, and the answer is consistently "that's being worked on" or gets handed to someone else in the room — pay attention to how long that's been the answer. Early in a project, that's a normal gap. Months in, it's a sign that the hardest, least glamorous part of the work has been quietly avoided the entire time, and it isn't going away on its own.
The only person who understands how it works is outside the company
This one is easy to miss because it often looks like efficiency. A vendor or consultant built the thing, and they're great — responsive, competent, always available for a call. But if nobody inside the organization could explain how it actually works, what it depends on, or how to change it without that outside person, the project hasn't actually transferred into the business. It's being rented, not owned, even if the invoice says otherwise.
Usage requires a reminder
Watch for how the tool gets used without a nudge. If people only touch it after a manager mentions it in a meeting, or during a compliance check, or right before a status update is due — that's not adoption, that's compliance theater. Genuine usefulness doesn't need a reminder system. People return to tools that save them time without being told to.
Status updates describe effort, not outcomes
Listen for the shape of the language in check-ins. "We made great progress on the integration" and "the team worked through several edge cases" describe activity, not results. A healthy project's updates sound like "tickets are down" or "the team stopped manually entering this data." When updates consistently describe what was worked on rather than what changed, the project may be busy without being effective — and busy is a much easier story to keep telling.
None of this means pull the plug today
Recognizing these signs doesn't automatically mean a project should be killed. Sometimes it means the goal needs to be re-anchored, the data work needs to finally happen, or ownership needs to move in-house. But recognizing them early — while there's still room to adjust course rather than just extend the timeline again — is what separates a project that gets rescued from one that quietly becomes permanent overhead nobody wants to be responsible for cancelling.
If you're in the middle of an AI initiative and want an outside, no-stakes read on whether it's still on track, book a free 30-minute call or reach out through our contact page. Sometimes the most useful thing is just a second set of eyes before another quarter goes by.