Quick answer
Validation is confirming, with evidence, that something meets its intended use. Verification asks whether it was built to spec; validation asks whether it meets a real need. For a feature, validation needs behavior: how many people tried it, used it again, and came back more than before.
What is validation?
Validation is the act of confirming, with evidence, that something does what it is meant to do for its intended use. The word covers several fields (data, software, medicine, research), but they share one test: not "was it built to the plan?" but "does it work for the purpose it exists for?"
In product work the purpose is a person's need, so the evidence has to come from people: what they chose, used and came back to. A feature that the team believes is useful has not been validated. A feature that 30% of active users tried, and 40% of those used again, has started to be. This page defines validation, separates it from verification, lists the three kinds you meet, and gives a rule for deciding from data whether a new feature passed.
Validation vs verification
The two terms are often swapped, but they answer different questions. Barry Boehm's 1981 phrasing, widely quoted, is that verification asks whether you are building the product right and validation asks whether you are building the right product.
| Verification | Validation | |
|---|---|---|
| Question | Did we build it to the specification? | Does it meet the real need for its intended use? |
| Checked against | The requirements and design documents | The user's or customer's needs |
| Who judges | Usually the team: tests, reviews, inspections | People who use it, or evidence of their behavior |
| Typical evidence | Passing tests, a review sign-off | Adoption, repeat use, retention, acceptance |
| Example | The export button produces a correct CSV in under 5 seconds | After 14 days, people export reports and come back to do it again |
| Can pass while the other fails? | Yes: a perfect build of the wrong thing | Yes: the right idea, built with bugs |
Wikipedia, summarising ISO 9001:2015 and the PMBOK Guide as adopted by IEEE, describes verification as conformity with requirements and validation as assurance that the product meets the needs of the customer and other stakeholders. A product can pass verification and still fail validation if the specification never matched a real need.
The three kinds of validation
| Kind | Question | Evidence | When | Example |
|---|---|---|---|---|
| Need validation | Is the problem real, and do people want this solution? | Interviews, sign-ups, a painted door test | Before building | Visitors click a "Coming soon" button for a feature that does not exist yet |
| Product validation | Does the built thing get used and re-used? | Adoption, repeat use, retention | After launch | Of 1,200 active users, 30% try the new feature |
| Data validation | Does this data follow the rules? | Format, range and consistency checks | Whenever data is entered or moved | An email field rejects text without an @ sign |
Need validation tests a product hypothesis cheaply before a minimum viable product exists; the full process is in product validation. Data validation is a different meaning of the same word and has its own entry: data validation. One point worth knowing: MDN's guide to form validation notes that client-side checks in the browser are a convenience for users, not a security measure, and that applications must validate again on the server because client-side checks are easy to bypass.
This page concentrates on product validation, because that is where behavior data decides the answer.
A rule for deciding whether a new feature passed
The common failure is to launch, watch the numbers move and decide afterwards that they look good. Use a rule that is fixed before launch, made of three checks over the same window (14 days is a common choice for features meant for weekly use):
- Adoption. Users who used the feature at least once ÷ active users who had access to it. Target example: at least 25%.
- Repeat use. Of the people who tried it, the share who used it again on a different day in the window. Target example: at least 35%. One trial shows curiosity; a second use shows the feature earned a place.
- Return against the users' own baseline. The share of feature users who came back in the following 14 days, compared with the same people's return rate in the 14 days before launch, not with the people who skipped the feature.
The third check exists because people who try new features are usually the more engaged. Comparing them with non-users measures who they are, not what the feature did. All three must pass; the numbers in each target are your call, and the point is that you set them before you look.
Worked example: did "Saved reports" pass?
Sample data: a product ships a "Saved reports" feature. In the first 14 days, 1,200 active users had access. Targets set before launch: adoption at least 25%; repeat use at least 35%; return no lower than the users' own pre-launch rate.
| Measure | Calculation | Result | Target | Verdict |
|---|---|---|---|---|
| Adoption | 360 ÷ 1,200 | 30.0% | ≥ 25% | Pass |
| Repeat use | 144 ÷ 360 | 40.0% | ≥ 35% | Pass, see below |
| Return after launch, feature users | 234 ÷ 360 | 65.0% | ||
| Return before launch, same 360 users | 198 ÷ 360 | 55.0% | ||
| Change for feature users | 65.0% − 55.0% | +10.0 pts | ≥ 0 | Pass |
The tempting comparison is feature users against everyone who skipped it: 65.0% against 294 ÷ 840 = 35.0%, a 30-point gap. Most of that gap is who chooses to try things. Their own before-and-after, +10.0 points, is the fairer number, and it is still not proof of cause because something else may have changed in those weeks. For the 840 non-users the rate moved from 300 ÷ 840 = 35.7% to 35.0%, down 0.7 points, so the product as a whole moved from 498 ÷ 1,200 = 41.5% to 528 ÷ 1,200 = 44.0%, an increase of 2.5 points.
Then check the noise. A 95% interval for the repeat-use rate is 40.0% ± 1.96 × √(0.40 × 0.60 ÷ 360) = 40.0% ± 5.1 points, about 34.9% to 45.1%. The point estimate clears the 35% target but the lower end falls just under it, so call this a provisional pass and recheck at 28 days. Adoption, 30.0% ± 2.6 points, clears 25% comfortably.
Common mistakes with validation
- Treating launch as validation. Shipping shows the team finished. Validation needs evidence that people use it.
- Treating verification as validation. "All tests pass" says the feature matches its spec. It says nothing about whether anyone wants it.
- Trusting what people say they would do. Interview answers and survey intent are a start, but behavior is the evidence: a click on a painted door beats "yes, I'd use that".
- Setting targets after seeing results. A target chosen to fit the data cannot fail.
- Comparing users with non-users. The people who try features are the ones already engaged; compare people with their own earlier behavior.
- Calling a total a result. "10,000 clicks" is a count. Divide by who could have clicked.
- Mixing the meanings. A project can pass data validation (the events arrive correctly) and fail product validation (nobody uses the feature).
How to run this check in TapCub
Here is the "Saved reports" example worked through in TapCub. TapCub measures the behavior; it does not run A/B tests or send surveys.
- Send an event for the feature. Send a
report_savedevent with custom events, and list it in All-platform analytics › Data › Tracking plan with a status. Turn on Live verification (24 h) to see that the event arrives and that required properties are filled, which is data validation for the event itself. - Read adoption. In All-platform analytics › Behavior › Behavior analysis, use the event analysis tabs for
report_saved. The unique-users aggregation gives the people who used it, and the active-percent aggregation gives the share of active users. - Count repeat use. In All-platform analytics › Users › Segments & tags, create a rule segment of people who did
report_savedat least twice in the last 14 days, and compare its member count with those who did it at least once. Segments need identified users, so calltapcub.loginto recognise people; anonymous visitors have no profile. - Compare people with themselves. In the retention view, set
report_savedas the start event and any activity as the return event, then run the same users' pre-launch window as a separate date range. Retention has no compare-period switch, so run the two ranges one after the other and note the launch date in Configuration › Annotations. - Write the verdict next to the target. Record the three results against the targets you wrote before launch. For the full process, see measure product adoption.
Sources
- Verification and validation, Wikipedia (Boehm 1981 phrasing; ISO 9001:2015 and PMBOK/IEEE definitions)
- Client-side form validation, MDN Web Docs (client-side vs server-side validation)
Was this explanation clear?
We use your answers to fix and expand this page.
Couldn’t send. Please try again in a minute.
Thanks for the feedback. We’ll use it to improve this page.You’ve already sent feedback on this page. Thanks.

Product Analyst
Focuses on event analysis, conversion funnels, and retention, using clear metric definitions and structured analysis to explain product performance.