A heuristic evaluation is an expert review where a few evaluators each inspect a site alone and judge it against a fixed set of usability rules, most often Jakob Nielsen's 10 usability heuristics.
It is fast and cheap and catches obvious friction, but it produces hypotheses, not proof: every problem it finds still needs an A/B test or user research to show it costs conversions.
This guide covers where Nielsen's heuristics come from, what each of the 10 looks like on a store or lead-gen funnel, and what real A/B tests in our database say when teams fixed those problems. Some fixes won big. Some lost.
What a heuristic evaluation is, and where it fits in CRO
Nielsen Norman Group defines it as "a method for identifying design problems in a user interface" in which evaluators judge the design against guidelines that make systems easy to use (NN/g).
No users are involved. That is both its speed and its weakness.
In a CRO program it usually sits at the start of an audit, next to analytics review and customer research.
Its job is to fill the test backlog with specific, fixable issues, ranked so the team knows what to test first.
It is not the same as usability testing.
A usability test watches real people try tasks.
A heuristic evaluation is an informed guess about where they will struggle.
The two work best together; see qualitative vs quantitative research for how they combine with test data.
Nielsen's heuristics: origin and how well the method works
Jakob Nielsen and Rolf Molich introduced heuristic evaluation in a 1990 CHI paper, "Heuristic Evaluation of User Interfaces."
Nielsen refined the list in 1994 using a factor analysis of 249 usability problems, and NN/g says the 10 heuristics have stayed unchanged since then; the 2020 update only rewrote the descriptions and examples (NN/g).
The key finding behind the method is that one reviewer is not enough.
Averaged over six of Nielsen's projects, single evaluators found only 35% of the usability problems, so NN/g recommends about five evaluators, and at least three (NN/g).
The model Nielsen cites (Nielsen and Landauer, 1993) is:
Problems found(i) = N × (1 − (1 − λ)^i)
Here N is the total number of problems in the interface, λ is the share one evaluator finds, and i is the number of evaluators.
In Nielsen's six case studies λ averaged 34% (range 19% to 51%).
| Evaluators (i) | Calculation with λ = 0.34 | Share of problems found |
|---|---|---|
| 1 | 1 − 0.66 | 34% |
| 3 | 1 − 0.66³ = 1 − 0.287 | about 71% |
| 5 | 1 − 0.66⁵ = 1 − 0.125 | about 87% |
| 10 | 1 − 0.66¹⁰ = 1 − 0.016 | about 98% |
The curve flattens fast, which is why three to five is the usual range.
The model assumes evaluators find problems independently, so they must work alone and only compare notes after everyone is done, as NN/g's method requires.
Note what "problem" means here: a usability issue an expert flags.
It says nothing about how many visitors hit it or whether fixing it changes revenue.
That gap is where A/B testing comes in.
The 10 usability heuristics, applied to conversion
Below, each heuristic uses NN/g's name, a short plain-English version, what to look for on a commercial site, and tests from our database where a team acted on that kind of issue.
1. Visibility of system status
Tell people where they are, what just happened and what comes next.
On a funnel: progress indicators, add-to-cart confirmation, delivery dates, "what happens after you submit."
Tinkoff Bank added a progress bar to a four-step credit card application; the green filling version raised completed submissions by 12.8% across 190,000+ visitors.
HomEquity Bank added a tracker explaining next steps after its inquiry form and reported +20% conversion rate.
Rylee + Cru showed estimated delivery dates site-wide and reported +22.4% conversion rate in a test run by Fenix Commerce, though that is a vendor story without a stated sample size. More on step indicators in multi-step forms.
2. Match between the system and the real world
Use the customer's words, not internal or category jargon.
Check nav labels, button copy and product names against how customers search and talk.
Frette relabeled a nav item from "Bath Linens" to "Bath Linens & Robes" after seeing robes were the top site search; bathrobe page visits rose 101% and category clicks 29%, though those are engagement metrics, not sales.
MarketingExperiments reported a three-word link change from "learn more" style wording to "Open an Account" that lifted funded accounts 43%. Both are cases of information scent.
3. User control and freedom
Give people a clear exit, undo and a way to come back.
On a funnel: editable carts, back links that keep entered data, saved applications, guest checkout.
IMB Bank added an exit popup offering to save a loan application, and saved applications rose 52%, but completions did not clearly move during the test.
That is a common pattern: the heuristic fix improves the behavior it targets without proving revenue impact.
4. Consistency and standards
Follow the conventions people learned on other sites: search in the header, cart top right, standard checkout order.
Nielsen's point is that users spend most of their time on other sites.
Many stores in our database tested an always-visible search bar instead of an icon.
Blend Commerce saw +22% conversion rate for Titan Casket on mobile.
But Clean Commit's test of a more prominent, guided search with suggestions and banners lost 12.0% conversion (not significant).
Conventional placement helped; extra layers on top of it did not. See site search.
5. Error prevention
Remove the conditions for mistakes before they happen: clear labels, sensible defaults, explanations of why a field is needed.
Wingify Services made form labels on Qualicorp's quote form more legible and reported +16.93% visits to the request-received page over 12 days.
Surefoot grouped BioLite's checkout email and phone fields into a labeled block with helper text on how each would be used: +18% transactions. More in form fields.
6. Recognition rather than recall
Do not make people remember things between pages.
Show recently viewed items, keep filters visible, label icons.
Goat Milk Stuff added short text labels to its mobile header icons and saw +10.37% conversion rate (Blend Commerce).
A recently viewed carousel lifted Stone Creek Coffee's revenue per visitor 13%, yet Clean Commit's recently viewed section on another store was +1.0%, inconclusive across 97,250 visitors. Same heuristic, different results.
7. Flexibility and efficiency of use
Offer shortcuts that help people who know what they want without slowing down first-time visitors: product finders, quick filters, autocomplete.
Aeromotive pinned its Pump Finder as a sticky button on mobile collection pages; Blend Commerce reported +69% conversion.
Fragrance Hub's instant search suggestions were flat (+1.92% RPV, not significant).
8. Aesthetic and minimalist design
Every extra element competes with the important ones.
This is the heuristic CRO teams over-apply most, because "clutter" is easy to spot and hard to measure.
Removing a "Track Order" link from main navigation raised revenue per visitor 30% for Peter Christensen, and moving a chat pop-up off key buttons gave Clean Commit +44.5% conversion.
But Clean Commit's stripped-back mobile layout cut conversion 4.2%, and MarketingExperiments found a plain, typo-filled signup page beat two polished versions, about 14% vs 7-9%.
Minimalism means removing what distracts, not removing information people use. See best practices that failed.
9. Help users recognize, diagnose and recover from errors
When something goes wrong, say what happened in plain language and offer a way forward.
On a store: 404 pages, out-of-stock pages, empty search results, payment failures.
Clean Commit added best-selling collections to a plain 404 page: +6.8% conversion rate on 140,427 visitors, though revenue per visitor rose only 0.3% because order value fell.
The Oodie added similar-product recommendations to out-of-stock pages and Visually reported +9.5% conversion rate. More on 404 pages.
10. Help and documentation
Ideally people need no help, but when they do it should be easy to find and focused on their task.
Tinkoff Bank tested a "More details" link on a card application: an inline box of benefits lifted applications 15.5%, while sending people to a separate info page underperformed the control.
Convertibles' "Help me choose" comparison modal was flat at +2.6% RPV.
Help works best in place, at the moment of doubt.
Surefoot
Blend Commerce
Clean CommitClean CommitClean CommitHow to run a heuristic evaluation for CRO
- Pick the scope. Choose the pages that carry the most traffic and revenue: homepage, collection, product page, cart, checkout, or the lead form. Review mobile and desktop separately.
- Define two or three tasks. For example, "find a gift under $50 and check out as a guest." Evaluating against tasks keeps findings tied to conversion.
- Use three to five evaluators working alone. NN/g suggests timeboxing each session to about 1-2 hours (NN/g). Each evaluator logs the issue, the heuristic it breaks, the page and a screenshot.
- Merge and rate severity. Nielsen's scale runs from 0 (not a problem) to 4 (usability catastrophe), based on frequency, impact and persistence (NN/g).
- Add traffic. Multiply severity by how many visitors reach the page. A severity-3 issue on checkout beats a severity-4 issue on an About page.
- Turn findings into test hypotheses. Fix obvious bugs directly. Test the rest, especially anything under heuristic 8.
| Finding type | What to do | Example |
|---|---|---|
| Broken or misleading (bug, dead link, wrong price) | Fix without testing | Plain 404 with no way back |
| Clear friction with a known fix | A/B test, expect it may be small | Unlabeled icons, missing delivery date |
| Taste or "clutter" judgment | A/B test, and expect losses | Removing content, redesigning layout |
| Unsure why users behave a certain way | Research first | High drop-off with no visible cause |
When heuristic evaluation is the wrong tool
NN/g itself says heuristic evaluations "cannot replace user research" (NN/g). Use something else when:
- The problem is motivation, not usability. Heuristics cannot tell you that the offer is weak or the price is wrong. Customer interviews and surveys can.
- You need to know if it matters. An expert can flag 40 issues; only traffic data and testing show which ones move revenue. The mixed results above for minimalism and search show how often a sensible fix goes flat or loses.
- The audience is specialized. Evaluators who do not know the domain miss problems and flag false ones. Watch real users instead.
- Only one person is reviewing. At about a third of problems found, a solo review is a starting point, not an audit.
Used well, a heuristic evaluation is the cheapest way to fill a test backlog with sharp ideas.
Treat each finding as a hypothesis and check it against real test results before you assume a lift.