UX Audit for SaaS: Report, Cost and What AI Misses

Key takeaways

• A finding without evidence is an opinion: each one needs a screen, a heuristic, a 0–4 severity score, evidence and a fix.
• In Baymard's 2023 test, GPT-4 found 14% of the issues experts found, and 80% of its suggestions were false positives. AI tools tested in 2025 were 50–70% accurate.
• As of September 21, 2026, most UX agencies on Clutch charge $25–$49 an hour, US agencies $100–$149. Clutch's pricing guide lists no separate price for audits.
• In B2B SaaS, audit every role, every screen state (empty, loading, error, partial success) and every bulk action.
• Fix severity 3–4 issues that take little effort first. Redesign only when findings cluster in one flow or point at the structure.

A UX audit is an expert review that checks a product against proven usability principles and shows where people will get stuck. You get a ranked list of problems, each tied to a screen, with evidence, a severity score on Jakob Nielsen's 0–4 scale and a fix. As of September 21, 2026, Clutch's UX pricing guide puts most UX agencies at $25–$49 an hour.

AI can't do this job for you yet. When Baymard Institute gave GPT-4 screenshots of 12 e-commerce pages in 2023, it found 14% of the issues Baymard's experts had found on the live pages, and 80% of its suggestions were false positives: advice that didn't fit the page.

This guide is for founders and product leads about to commission an audit of a SaaS product or a website. Inside: a sample finding card, market rates and examples from our work on a fintech payouts platform and a cosmetics store.

What is a UX audit, and how is it different from usability testing?

A UX audit is an inspection by experts, while usability testing watches real users try to complete tasks. The two catch different problems, so a growing product needs both.

Audits rest on heuristics: broad rules of thumb for interface design, such as Nielsen's 10 usability heuristics. A heuristic evaluation checks an interface against such a list, while an expert review also applies usability guidelines and the reviewer's experience. In practice, a UX audit is an expert review with a written report.

Four ways to find usability problems. Sources: Nielsen Norman Group and Baymard Institute, 2000–2026.

Nielsen Norman Group (NN/g), the research firm Nielsen co-founded, states the limit plainly: heuristic evaluations cannot replace user research. A good audit report marks which findings need a test before you pay for a fix.

When does a SaaS product need a UX audit?

Commission one when a decision is coming and you need to know what's broken before you spend money on it. The usual triggers:

• Before a redesign. Many organizations run expert reviews before a major redesign project, NN/g notes, to map what works and what doesn't. The strengths tell designers what not to break.
• Before a demo to partners or investors. The product works but still looks like an internal tool, and the date is fixed.
• After a metric drops. Analytics shows where users leave, for example before activation (the moment a new account gets its first real result), but not why. An audit gives you likely causes to test.
• After features pile up. Similar actions now look and behave differently from screen to screen, which breaks the consistency and standards heuristic.
• On a schedule. Even with no redesign planned, NN/g recommends an expert review every 2 to 5 years.

Skip the audit when the open question is whether anyone wants the product: an audit tells you how to make something usable, not whether it's worth building. Skip it too when you already know the problem and only need the fix.

What should a UX audit report include?

A useful report lists each problem where it occurs, with evidence, a severity rating and a fix, plus what already works. NN/g lists the same core parts and adds examples of good practice from other products.

Severity tells you how urgently a problem needs fixing. Nielsen combines three factors: how often the problem occurs, how hard it is to overcome, and whether it keeps bothering users after they know about it. His scale runs from 0 to 4:

• 0: not a usability problem at all
• 1: cosmetic, fix only if time allows
• 2: minor, low priority
• 3: major, high priority
• 4: usability catastrophe, fix before release

Every finding in a report should look like this one. This is an illustrative example written for this article, not a finding from a client project.

Ask who rated severity and how many people did. Nielsen found the mean of three evaluators' ratings good enough for many practical purposes, and his 1994 warning still applies:

"My experience indicates that severity ratings from a single evaluator are too unreliable to be trusted." — Jakob Nielsen, Nielsen Norman Group

How does a UX audit work, step by step?

A solid audit narrows the scope, gathers data, has several people review each flow independently, then merges and rates what they found. The steps below follow NN/g's method for heuristic evaluation, plus the product data most SaaS teams already have.

1. Set the scope. Pick the flows the decision depends on, such as sign-up, the core task and billing. NN/g suggests narrowing by task, section, user group or device.
2. Collect data. Analytics funnels, session recordings, support tickets and past research turn opinions into evidence.
3. List screens and states for each role. Every screen a role sees, in every state it can be in.
4. Review independently. NN/g recommends three to five evaluators working separately, each going through a task twice if the product is new to them: once to learn it, once to look for problems.
5. Merge and rate. Duplicates are combined, each evaluator rates severity alone, and the ratings are averaged.
6. Report and walk through. Findings and strengths go to whoever will act on them, in a live session.

What do generic UX audits miss in B2B SaaS?

Generic checklists are written for public websites with one kind of visitor. They have no rows for the roles, screen states and bulk operations of B2B products.

Roles

An admin, an approver and an accountant see different screens and do different jobs. Audit each role's main tasks separately, with a test account for each. Testing follows the same logic: NN/g advises testing additional users when a product has several highly distinct groups of users.

States

Every screen has several states: empty (no data yet), loading, partial success, error and success. An audit of screens full of demo data misses most of them. The first heuristic adds a rule worth checking on every action: nothing with consequences for users should happen without telling them.

Tables and bulk actions

Most B2B work happens in tables: filter, select, then act on many rows at once. Check that the confirmation names the count and the amount, that progress stays visible, and that failures come back row by row with a way to retry.

We designed for this kind of bulk work on NVP, a fintech platform that onboards, checks and pays self-employed contractors, sole traders and civil-law workers. Its working prototype looked like an internal tool and had to face partners in May. Before moving a pixel, we mapped all 18 prototype screens.

The single-payout form got empty, filled, error and success states. Bulk payouts arrive as Excel registries (payout lists), so the upload has four states: drop zone, processing, result and an error table that names the row, the field and the fix. Clean rows move on to approval while the rest wait. For the dashboard side of that project, see our fintech dashboard design guide.

Case study · NVPPayouts and paperwork for companies hiring freelancers15 daysfrom brief to a partner-ready dashboard

Can ChatGPT or another AI tool do a UX audit?

Not on its own. In published tests, ChatGPT-4 missed most issues, and even newer AI tools mixed good advice with bad that only an expert can tell apart.

Baymard Institute's test of ChatGPT-4, published in October 2023, shows why. Six trained UX benchmarkers spent 2–10 hours on each of 12 product, listing and checkout pages at sites such as Lego, REI and Zalando, and logged 257 issues. Baymard then uploaded a screenshot of each page to ChatGPT-4, asked what UX improvements could be made and got 178 suggestions:

• It caught 14.1% of the issues on the live pages, and 25.5% of those visible in the screenshots.
• Of its suggestions, 80.1% were false positives, and 8.9% would likely have hurt the experience.
• For all 12 pages it suggested making the site mobile responsive, although every screenshot showed a desktop layout.

Baymard's explanation: a very large share of UX issues are interactive, and a screenshot can't show a click, a hover or a move between pages.

Newer tools do better, but not well enough to trust blindly. In a January 2026 article, NN/g cites two Microsoft researchers who tested four unnamed AI tools in March 2025 and measured accuracy of 50% to 70%. The catch, as NN/g explains it: if 7 of 10 suggestions are good, the other 3 may make the product worse, and you can't tell which.

"GenAI-powered systems may struggle to identify deeper, subtler issues, like a fundamental mismatch between the site's information architecture and the expectations of its target audience." — Kate Moran, Nielsen Norman Group

Information architecture is how content and features are grouped and labeled, and the Barber & Co example below turned on exactly that kind of problem. Even Baymard's own AI tool, UX-Ray, covers only the guidelines where its AI is at least 95% accurate, about a quarter of the total.

How much does a UX audit cost in 2026?

Clutch, a B2B review platform, gives no separate price for UX audits in its UX Services Pricing Guide, so you build the estimate from hourly rates and scope. As of September 21, 2026, the guide puts most UX agencies at $25–$49 an hour, and the rate depends heavily on where the agency is.

Hourly rates of UX agencies by location, from verified client reviews. Source: Clutch, as of September 21, 2026.

The guide has no line for the UAE. UX/UI design projects reviewed on Clutch typically cost $10,000–$49,999, but that range is for UX/UI projects in general, not for audits.

These are market ranges from Clutch reviews, not our quote. For your scope, a partner gives an estimate on a call.

To sanity-check a quote, multiply the flows in scope by the hours per flow, the number of evaluators and the Clutch hourly rate for the agency's location, then add time for data, rating, writing and the walkthrough. Depth moves the hours most: Baymard's experts spent 2–10 hours on a single page. Clutch names the scope of work as the biggest cost factor, and in SaaS the scope grows with every role, every platform and any usability testing you add.

After the audit: quick fixes, iteration or redesign?

Let severity and the pattern of findings decide. Even a deep redesign doesn't have to mean a new platform.

• Severity 3–4 with small effort: fix now, before the next release or demo.
• Several findings in one flow: redesign the flow, test it with five users, fix it and test again. NN/g recommends three studies with five users each over one study with 15.
• Findings about structure, meaning how products, features or content are grouped and found: rework the information architecture first and the screens after.
• Severity 1: batch them into a polish sprint, or let them go.

Barber & Co, a Minsk retailer of professional hair, face and body care, is a case of the third kind. It wasn't a formal audit, but the problems we found in its store on Tilda, a website builder, were structural, and each one led to a specific change.

The decision behind it all: gender is a storefront, not a filter, and product type is the catalog. We kept Tilda and did the work in three weeks on the live store.

Case study · Barber & CoA faster, mobile-first store for a barber brand30×lighter catalog images

If the redesign also changes page addresses, read how to redesign without losing SEO first. If the platform itself is in question, see our Shopify vs custom vs Tilda comparison.

What to prepare before a UX audit: a checklist

The more context you hand over, the more audit hours go into findings. Have these ready before kickoff:

• The decision and the metric. What the audit should inform, and the number you watch: activation, trial-to-paid conversion or time to complete the core task.
• Flows and roles in scope. The flows that matter most and who uses each one.
• Test accounts for every role, filled with realistic data, plus an empty account and one with errors.
• Analytics and recordings: funnels in Google Analytics 4 or your product analytics, plus session recordings if you collect them.
• Support tickets and sales-call notes from recent months, tagged by topic.
• Past research and earlier audits, so nobody pays twice for the same finding.
• Known constraints: platform, design system and what can't change this quarter.
• Design files or the prototype, if the audit runs before launch.
• One owner for the fixes, who joins the walkthrough and decides what ships.

If you can't fill in the first item yet, start there. An audit with no decision behind it turns into a long list nobody acts on.

Questions we get asked

How long does a UX audit take?

There is no standard length: it depends on the flows, the roles and the depth. NN/g suggests reserving about 1–2 hours per evaluator for a heuristic evaluation session, while Baymard's experts spent 2–10 hours on a single page. Calendar time also goes to access and data, independent passes, merging, rating, writing and the walkthrough. Ask for a schedule that names each step.

Can you run a UX audit on a Figma prototype before launch?

Yes. NN/g says expert reviews can happen at any stage, provided there is a prototype with enough detail. Before launch there is no analytics, so findings rest on heuristics, guidelines and the reviewers' experience. That makes a short test with five users before you build even more useful: it checks the riskiest findings while a change is still a design edit, not a rebuild.

Should our own designer do the UX audit?

Your designer should check their own work, but the audit itself works better from outside. NN/g recommends a reviewer who was not involved in creating the design: an outsider is not emotionally invested in it and is unaware of internal team politics. Your designer then gets the report, joins the walkthrough and owns the fixes.

Does a UX audit include an accessibility (WCAG) check?

Not by default. WCAG, the W3C's Web Content Accessibility Guidelines, is a separate standard with its own success criteria. A UX audit may flag low contrast or small tap targets along the way, but a conformance check goes through the criteria one by one. If you sell to the public sector or large enterprises, ask for it as a separate line in the scope.

How do we know the UX audit paid off?

Pick the metric before the audit starts, such as activation or time to finish the core task. After the fixes ship, compare the same funnel step over a comparable period, and confirm with a five-user test that the problem is gone. A change in one week is a signal, not proof, so look across several weeks and note anything else that changed.

Sources

1. UX Expert Reviews · Nielsen Norman Group, 2018
2. Severity Ratings for Usability Problems · Nielsen Norman Group, 1994
3. How to Conduct a Heuristic Evaluation · Nielsen Norman Group, 2023
4. 10 Usability Heuristics for User Interface Design · Nielsen Norman Group, 2024
5. Why You Only Need to Test with 5 Users · Nielsen Norman Group, 2000
6. Demand Accuracy in Your AI Tools: Lessons from Baymard Institute · Nielsen Norman Group, 2026
7. Testing ChatGPT-4 for ‘UX Audits’ Shows an 80% Error Rate & 14–26% Discoverability Rate · Baymard Institute, 2023
8. UX Services Pricing Guide 2026 · Clutch, 2026

Share this article

About the author

Artsiom Yankovich

Head of Design & Strategy

Leads design and strategy, from product logic to brand. 9+ years, 150+ projects.

Meet the leadership