Restart a stalled AI pilot
Your first attempt taught you something. Use it.
The demo worked, but the project did not stick. Before choosing another model or rebuilding the pilot, use Canvas to separate the original assumptions from what your team actually learned.
On the home page, choose “Restarting an attempt” to set your starting point.
Illustrative workflow to plan
A small first version. A clear human decision.
Select a step to follow the handoff. People keep the decisions marked for human review.
Reconstruct the attempt
Record the intended outcome and what actually happened.
Plan the evidence and ownership for this handoff.
Example: a support assistant drafted plausible answers, but staff spent too long correcting outdated policy references. The next step may be fixing knowledge ownership before another pilot. Canvas helps turn that lesson into a plan; it does not diagnose a failed deployment from a short description.
Find the reason it stalled before starting again.
The obstacle might be unreliable sources, an unclear owner, a handoff nobody wanted or a value case that ignored review effort. Revisit the real workflow with someone who tried to use the first version. Keep what worked; make the unresolved gap explicit.
Planning context: the Canvas guide on stalled AI projects examines the gap between a promising idea and a workable operating model. Use it to structure the review, not as a diagnosis of your team or a failure-rate claim.
Read: Why AI projects stallBring the evidence behind the idea.
- The original goal, intended users and actual workflow.
- Sanitized examples of failures, feedback and review effort.
- What has changed since the attempt, and what is still unknown.
Turn discovery into a reviewable plan.
Use discovery to capture the earlier attempt in a shared Agentic Brain. Review readiness gaps, narrow the first version and prepare a revised blueprint. A useful conclusion can be to improve the process first or stop the idea.
Define what would make it worthwhile.
Set a baseline from the previous attempt: corrections, usage, time and operating cost. Define what the next test must demonstrate, who will judge it and when to stop. A restart is a decision to test, not a promise of success.
Your first conversation
Start with a brief like this.
We tried an AI support assistant, but outdated guidance created too much review work. Help us understand what failed, what needs to change and whether a narrower pilot is worth attempting. We want evidence for a decision, not another demo.
Adapt this example and bring it to the home-page conversation. Choose “Restarting an attempt” in the starting-point options. The link does not select a chip or fill in your brief.
Start planningBefore you start
Do we need to abandon the original idea?
Not necessarily. Separate a useful objective from a weak assumption or implementation choice. The review may support a smaller version, preparation work or a decision to stop.
What if we do not know why it failed?
Start with observations and mark explanations as hypotheses. Bring the people who used and reviewed the pilot. Canvas can structure the questions; it cannot replace their evidence.
How do you restart a stalled AI pilot?
Reconstruct what happened, name the gap that stopped it and decide what is different this time. Then test one smaller step with a named owner and a number to measure before rebuilding anything.
Should we kill the pilot or revive it?
The decision owner decides, based on what has changed in the evidence. If nothing has, a documented pause with the prerequisite named is a sound result.