SurfingBearHome › Blog › Restarting after a failure
Restarting AI adoption after it failed once
The second attempt is harder than the first for reasons of trust, not technology. Budget and organisational attention were already spent once.
Classifying the cause Second attempt Rebuilding trust
The short answer
The first step in restarting is classifying the failure accurately. Most "AI adoption failures" are not AI failures but target-selection failures, data failures or operational failures.
Misclassify the cause and you repeat it. Change the tool while the target process stays the same and the result stays the same.
Classifying the failure — from symptom to cause
| Symptom | Real cause | What to change next time |
|---|---|---|
| Nobody used it | No integration, or unclear use cases | Inside a tool they already use, with three named uses |
| The output was not usable | The target process was unsuitable | Move to work with stable rules and consistent input |
| Review took longer | Output format and prompts never fixed | Fix the format first, then re-measure |
| It was confidently wrong | Stale or unorganised data | Narrow the document scope and restart |
| Security review blocked it | Terms checked too late | Move terms checking into candidate selection |
| We could not prove the benefit | No baseline recorded | Record the baseline numbers first |
Design principles for the second attempt
Rebuilding organisational trust
Start by writing the first failure up explicitly. Pass over it quietly and all the organisation retains is "AI does not work here". Documenting the cause turns that into "that approach did not work".
Then a small result. Scope it so something measurable appears within four to six weeks. Returning with another large plan simply overlays the memory of the first attempt.
And report the result in the language of the people who do the work. "Month-end can start two days late and still hit the deadline" travels further inside an organisation than "40% reduction in processing time".
Frequently asked questions
Can we reuse the tool that failed?
If the tool was not the cause, yes. Where the target process or the data was the cause, the same tool produces a different result. But if organisational perception is poor, the same tool carries a persuasion cost.
Leadership will not approve another attempt.
Present a small scope, an explicit stop condition, and the cause analysis of the first failure together. "Last time the cause was this; this time we confirm it against this number within four weeks" passes more often than a budget request.
We have no record of the failure.
You can reconstruct it through interviews: what was tried, where it stopped, why users did not adopt it. Mark it as reconstructed, but write it down.
What if the second attempt fails too?
If the cause is repeatedly in target selection, a process assessment comes before tool adoption. You are most likely skipping the step that establishes which work suits automation.
Diagnose the cause
Find which axis failed
The AX readiness assessment scores work, data, technology and organisation. It shows which axis the first failure came from — and that is where the second attempt starts.