Everyone loved it. Nothing shipped.
The demo went well enough that people talked about it for a week. Then it needed real data, an owner, a security review, and someone accountable for what it got wrong. It has been nearly there ever since.
Now it is awkward. Nobody wants to kill it and nobody is moving it.
A stalled pilot is not a failure. It is a finding.
In the stalled pilots we get asked about, the model is rarely the problem. They stall on the things nobody assigned: who owns the problem, whose knowledge the thing needed, who is accountable when it is wrong, and whether the problem was worth solving in the first place.
That is uncomfortable, but it is useful, because every one of those is fixable and none of them is fixed by a better model.
The two honest endings
Restart it properly. Name the problem in operating terms, give it an owner, and get the context it needs into a form it can actually use. If the problem is real, this works.
Stop it cleanly. Say what you learned, retire it, and spend the attention on a problem that deserves it. Stopping well is cheaper than a pilot that limps for another two quarters, and it buys you credibility for the next one.
We will tell you which one you are looking at. Sometimes that answer costs you nothing but the conversation.