90-Day AI Feasibility Review
Re-test previously impossible workflows every 90 days as AI capabilities advance
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 96%
The 90-Day AI Feasibility Review treats every conclusion about AI capability as temporary. Teams maintain a backlog of useful tasks that produced unacceptable results, then revisit the most valuable cases once per quarter. Each review repeats a comparable test, evaluates quality against a predefined threshold, and either advances the use case into a controlled pilot or returns it to the backlog with documented limitations. The mechanism works because model quality, modalities, integrations, and tooling can improve materially within a few months. Instead of chasing every product release, the team uses a predictable review cadence that converts rapid technical progress into operational opportunities while keeping experimentation focused on real business value.
Origin
Extracted from Marketing Against the Grain during a discussion of how quickly AI capabilities were expected to improve in 2024.
Core principles
- 01AI capability assumptions expire quickly
- 02Previously failed use cases deserve periodic reassessment
- 03Small experiments reveal progress better than speculation
- 04Business value matters more than novelty
How to run it
- 1
Build the Reassessment Backlog
Record valuable tasks that AI attempted but could not complete at an acceptable quality level. Include the failed output and the limitation that blocked adoption.
Pro tip Prioritize repetitive, expensive, or high-volume work.
Watch out Do not add speculative use cases with no clear business value.
- 2
Define the Success Threshold
Specify what accuracy, quality, speed, cost, and human oversight would make each use case viable.
Pro tip Use a representative example that can be repeated during every review.
Watch out Changing the benchmark between tests makes progress difficult to judge.
- 3
Re-Test Every 90 Days
Repeat the task using current models and tools, including newly available modalities or integrations.
Pro tip Start with the highest-impact use cases rather than testing the entire backlog.
Watch out Do not assume a polished demo proves production reliability.
- 4
Compare and Classify
Compare the new result with the previous attempt and classify the use case as viable, promising but limited, or still unsuitable.
Pro tip Preserve outputs so improvement is visible over time.
Watch out Avoid grading solely on novelty or subjective excitement.
- 5
Pilot Newly Viable Workflows
Run a controlled production pilot when the output clears the success threshold. Measure operational results before expanding adoption.
Pro tip Keep a human review step until performance is consistent.
Watch out Do not scale immediately from one successful test.
In the wild
A marketing team previously rejected AI video localization because translated speech and lip synchronization looked unnatural. It retains the campaign as a quarterly benchmark, re-tests it after new tools launch, and begins a limited two-language pilot once the output clears its quality threshold.
→ The team captures a newly viable localization opportunity without constantly chasing AI releases.
Common mistakes
Treating One Failure as Permanent
A poor result reflects the tested model and moment, not necessarily the long-term feasibility of the workflow.
Testing Without Acceptance Criteria
Without a defined threshold, teams can mistake an impressive-looking result for an operationally useful one.
Chasing Tools Instead of Outcomes
Reviewing every new product creates noise; the cadence should begin with valuable business tasks.
Is it for you?
Best for
Teams with a backlog of automation or content ideas that earlier AI tools could not execute reliably.
Not ideal for
Stable problems where AI capability is irrelevant or repeated testing would cost more than the potential benefit.
From the transcript
“try what couldn't be done a few months ago can be done now”
“you're going to have to keep revisiting your use cases every 90 days because it's moving so fast that every 90 days it could now…”
“you have to really look at every problem through the lens of this potentially could be possible now with AI and like recalibrate whether it's…”
From the episode
10 AI Predictions That Will Completely Change Life In 2024 (#186)