A chatbot got shoved to the front of the product so the release notes could say “now with AI.” Adoption tells the real story: a couple of clicks, and then nothing. It made someone’s daily job harder, not easier, and the spend is hard to justify at the next budget review. Nobody stopped to ask what problem it was actually solving.
The feature is not failing because it was built badly. It is failing because it was aimed at a headline, not a user.
Why this happens
When AI becomes a mandate instead of a tool, the question flips. Instead of “what do our users struggle with, and would a model help,” it becomes “where can we put AI so we can say we have it.” That question has easy answers and no value. A chatbot on the homepage is visible, demos well, and solves nothing anyone was asking for.
The deeper cause is skipping discovery. Nobody defined the user pain, tied it to a number, or asked whether the pain was even the kind of problem a model is good at. So the feature launched pointed at a press release.
What it’s costing you
The obvious cost is the build itself, plus the inference bill for a feature nobody uses. The quieter cost is worse: a product surface that now feels gimmicky, a team that spent a cycle on it, and a data point that makes your next real AI investment harder to fund because the last one flopped. Features that make a daily job harder also carry a churn cost you may not have traced back to their source yet.
What good looks like
AI applied only where a real user pain is genuinely non-deterministic and tied to a number that matters: time saved, tickets deflected, conversion moved. Everything else built as software that just works, quietly, without a badge. One feature people actually use beats five that ship with a launch tweet and die in the analytics.
How Density fixes it
The fix is to build the one AI feature that earns its place and ship it properly, instead of another badge nobody uses. A Forward Deployed AI Engineer ($9,500 a month, deployed in 7 to 10 days, 120 day replacement guarantee) embeds in your team, finds the genuine user pain a model would move, and builds it to production inside your codebase, with the discovery and the numbers to prove it belongs.
We advocate using as little AI as possible: a model only where the work is truly non-deterministic, plain software everywhere else. We have shipped for US companies since 2016 at 96 percent retention. If you would rather locate the one real opportunity before committing to a build, the AI Readiness Assessment ($2,500, two weeks, credited toward the engagement) ranks your workflows by ROI first.
Let’s talk
Stop shipping AI for the badge and start shipping it for the outcome. Book a 30-minute call and we will find the one place it earns its keep. See the Forward Deployed AI Engineer.
Keep reading: where plain software beats a model outright, or the feature you built that nobody uses. Back to all AI implementation challenges.