Think Build Implement Repeat
London, UK +44 7367 067226
WhatsApp FOLLOW f in X
AI & Machine Learning

Evaluating Whether an AI Feature Is Good Enough

Last updated:

“It seems good” is not a measurement

The most common way an AI feature goes wrong is that nobody can tell it has. Quality drifts — a provider updates a model, your documents change, the input distribution shifts — and the first signal is a complaint.

The fix is unglamorous: a fixed set of real examples with known-correct answers, run automatically, with the score recorded every time.

Building the set

  1. Take 50–200 real examples, sampled across the range you actually see
  2. Have someone who knows the domain write the correct answer for each
  3. Include the hard cases deliberately — the ambiguous, the malformed, the unusual
  4. Include a handful that should be refused or escalated
  5. Store it in version control alongside the code

Two days of work, and it is the difference between engineering and guessing. Every AI project we run has one before any prompt is written.

If you cannot say what the right answer is for fifty real examples, you do not yet understand the task well enough to automate it.

Choosing a metric that matches the stakes

TaskMetricWhy
ClassificationPer-class precision and recallOverall accuracy hides a class you always get wrong
ExtractionField-level exact matchOne wrong field spoils the record
RetrievalWas the right passage in the top resultsThe answer cannot be right if the source was missed
DraftingHuman accept-without-edit rateThe only thing that correlates with time saved
AgentsTask completed correctly, end to endPer-step accuracy compounds misleadingly

For anything with asymmetric cost, weight the metric accordingly. Missing a fraudulent transaction is not the same as flagging a legitimate one, and a single accuracy figure obscures the difference.

Regression testing on every change

The evaluation set runs whenever the prompt changes, the model changes, the retrieval changes or the provider ships an update. A drop beyond a threshold blocks the deployment, exactly like a failing unit test.

This catches the two failure modes that would otherwise be invisible: a prompt improvement that helps one case and breaks four, and a provider update that quietly changes behaviour on your edge cases.

Watching production, not just the test set

  • Sample a percentage of live outputs for human review weekly
  • Track the confidence distribution — drift shows up here first
  • Track how often users edit, reject or escalate
  • Alert on rate changes rather than on individual events

Human review of live samples is what finds the failures your evaluation set never imagined. Those cases then get added to the set, which is how it stays useful rather than becoming a museum of last year's problems.

Deciding when it is good enough

Good enough is defined by what happens to the errors, not by a number in the abstract. With a human review step, a lower bar is fine because mistakes are caught. Fully automated, you need the residual error rate to be genuinely tolerable at your volume.

So the honest way to state a threshold is: at this accuracy and this volume, we will produce roughly this many errors a week, and here is who handles them. If that sentence is acceptable when written down, the feature is good enough to ship.

Frequently asked questions

How big should an evaluation set be?

Fifty is enough to catch obvious problems; a few hundred gives confidence on the edge cases. Coverage of the variety you see matters far more than raw size.

Who should write the correct answers?

Whoever does the task today. Not the developer — they will unconsciously write answers the system can produce.

How often should we re-run it?

On every change, and monthly regardless. Provider-side updates happen without notice, and the monthly run is what catches them.

What if the correct answer is subjective?

Then score on acceptability rather than exactness, with two reviewers and an agreed rubric. If reviewers cannot agree with each other, no model will satisfy them either — and that is worth knowing early.

Keep reading

Need to know whether a model is good enough to ship?

We build the evaluation set first on every project. Tell us the task and we will tell you how we would measure it.

Book a free 30-minute call Get a project estimate WhatsApp us

Related services

What we build for problems like this one

AI AgentsMachine LearningAI Integration