Synonyms: false positives vs. false negatives, precision-recall tradeoff, hit rate vs. coverage, exactness vs. completeness
Do not index
Definition
Precision and recall measure how an AI model gets things right in two different ways. Precision asks: of the items the model flagged, how many were correct? Recall asks: of all the items it should have flagged, how many did it catch? They trade off, so pushing one up usually pulls the other down.
For product work, this is the choice between annoying users with false alarms and quietly missing real cases.
Use cases
Pick the wrong balance and the feature fails in a way users feel, even when the "accuracy" number looks fine.
Spam filter, too much recall: It catches every spam message but also flags real emails. Users miss an invoice or a job offer and stop trusting the inbox.
Fraud check, too much precision: It almost never false-alarms, so it feels clean, but it misses real fraud and the company loses money.
Search and recommendations: High recall returns everything loosely related and buries the good results in noise. High precision returns a few exact matches but skips options the user wanted.
How it's used in practice
Decide which error costs more: For this feature, is a false alarm or a miss worse? Set the threshold toward the side you can least afford to get wrong.
Design for the error you chose: If you favor recall, make "not spam" recovery one tap. If you favor precision, give users a clear "report a miss" path.
Show the tradeoff with real numbers: Bring stakeholders precision and recall separately, not a single accuracy figure that hides the choice.
Make the threshold context-aware: Let high-stakes flows lean toward catching everything and low-stakes flows lean toward staying quiet.
🪄
Pro-tip: "Accuracy" hides the tradeoff. A model that's 99% accurate can still miss almost every rare case when the thing you're detecting is rare. Ask for precision and recall separately, and tie the call to the mistake your users can least afford.
Challenges & limitations
You can't max both: Improving precision usually costs recall and the other way around. There's no setting that wins on both at once.
The right balance shifts: What's correct depends on context and changes as usage and stakes change, so it needs revisiting.
Recall vs Precision in an AI Product or Feature (Matt Szaszko, Medium) — frames the precision-recall choice as a product and UX decision driven by risk, with low- vs high-risk examples and a Google Photos walkthrough (People tab, NLP search) arguing the balance should shift by feature and available data.
Share this post
Get free UX resources
Get portfolio templates, list of job boards, UX step-by-step guides, and more.
