Web Design
9 CRO Audit Checks That Find Lost B2B Leads

Joshua Jackai
21 min

Published 19 November 2026 ยท Last updated and fact-checked November 2026
Short answer: A CRO audit is a structured review of where a website loses buyers between the first visit and a sales conversation. For a B2B site it covers nine checks, from tracking to lead follow-up. Use three to five reviewers, not one: in Nielsen and Molich's 1990 study, single evaluators found only 20% to 51% of usability problems.
A US B2B site can have steady traffic and still lose buyers after they arrive. That is a leak, not a traffic problem. Visitors arrive from search, ads and referrals, read two pages and leave, or fill in a form that nobody follows up. A CRO audit finds where that happens and puts the fixes in order of value.
I am Joshua Jackai, founder of Demand Designer. I design and build conversion-focused B2B websites on fixed prices, and every project starts with the nine checks below. Each one rests on published research about how people judge pages, fill in forms and respond to delay, and each says where that research stops. A broader guide to the whole discipline lives in my post on B2B conversion rate optimization.
Key takeaways
Nielsen and Molich (1990) found single evaluators caught 20% to 51% of usability problems, while pooled reviews from three to five people did well.
Lindgaard and colleagues (2006) showed people judge visual appeal within 50 milliseconds, so the first screen gets checked before anything else on the page.
Fogg and colleagues (2003) collected credibility comments from 2,684 people; "design look" appeared in 46.1% of them.
Galletta and colleagues (2004) tested page delays from 0 to 12 seconds with 196 people and found losses began to level off from about 4 seconds.
Johari and colleagues (2017) showed that checking a running A/B test again and again makes ordinary p-values unreliable.
Sabnis and colleagues (2013) studied 461 sales reps at four firms and found the share of time reps gave to marketing leads depended on prequalification, managerial tracking, lead volume and rep experience.
Check | What it finds | Key number | Source type |
|---|---|---|---|
1. Tracking and funnel data | Leads you are not counting | Causal tests beat opinion | Peer reviewed |
2. Expert review by 3 to 5 people | Usability problems | 20% to 51% found by one reviewer | Peer reviewed |
3. First screen and offer | Visitors who bounce at once | 50 ms first impression | Peer reviewed |
4. Proof and credibility | Buyers who doubt you | 46.1% of comments, 2,684 people | Peer reviewed |
5. Page speed | Visitors lost to delay | Delays tested 0 to 12 s | Peer reviewed |
6. Forms and booking | Drop-off at the last step | 65 people, 20 form guidelines | Peer reviewed |
7. Five-user test | Problems data cannot explain | Best ratio at 4 in a medium example | Peer reviewed |
8. Test plan | False wins from A/B tests | Peeking breaks p-values | Peer reviewed |
9. Lead follow-up | Leads sales never calls | 461 reps studied | Peer reviewed |
What is a CRO audit?
A CRO audit is a review of a website's conversion path that finds the places where visitors stop moving toward a sale, explains why, and ranks the fixes. For a B2B company the conversion is usually a demo request, a booked call or a qualified form fill, not a checkout.
The audit mixes three kinds of evidence. Numbers from analytics and the CRM show where people drop. Expert review and short user tests show why. Research on credibility, speed and forms tells you which fixes are likely to matter, before you spend weeks testing them. An audit is not a list of opinions about colours. Every finding should point to a page, a number and a fix.
What is included in a CRO audit?
A CRO audit includes nine checks: tracking and funnel data, an expert review, the first screen, proof, page speed, forms, a short user test, the test plan and lead follow-up. The deliverable is a ranked list of fixes, each tied to a page and a number.
Most agency audits stop at the website. I add the ninth check because, in B2B, a lead that waits days for a reply is, in my experience, a lead sales has effectively dropped. The website's job ends when a qualified buyer is talking to sales, so the audit follows the lead that far.
1. Tracking and funnel data: best for the first day of any audit
The tracking check confirms that every form, booking and call is recorded once, with its source, in both analytics and the CRM. It suits every audit, because every later check depends on the numbers. Kohavi and colleagues call controlled experiments "the best scientific design" for linking a change to behaviour.
What to compare: Form submissions in analytics against new contacts in the CRM, same 30 days
Funnel steps: Landing page, key page, form view, form start, submit, booked meeting
Common miss: Thank-you page views counted as leads, so refreshes and spam inflate the total
Time needed: About half a day for a site with under 10 forms
Before I judge any page, I check that the numbers are real. I submit every form on the site myself, then look for each test lead in analytics and in the CRM. A common failure is a form that sends an email but creates no CRM record, or records a lead with no source. Either one makes every later number in the audit wrong. Then I build the funnel from landing page to booked meeting and look for the step with the steepest drop. That step, not the homepage, is where the audit spends most of its time.
Kohavi, Longbotham, Sommerfield and Henne (2009) wrote the standard practical guide to online experiments. Their point for an audit is simple: decisions should follow measured behaviour, not the most senior opinion in the room.
"Controlled experiments embody the best scientific design for establishing a causal relationship between changes and their influence on user-observable behavior," stated Ron Kohavi and colleagues (Data Mining and Knowledge Discovery, 2009).
Their guide is about large sites that can run experiments. Most B2B sites cannot, which is why the audit leans on the other eight checks too.

2. Expert review by three to five people: best for finding usability problems fast
An expert review, also called a heuristic evaluation, means trained reviewers walk the site against a list of usability principles and log each problem. It suits sites that need findings within a week. In Nielsen and Molich's 1990 study, single evaluators found only 20% to 51% of the problems.
Reviewers: Three to five, working alone first, then pooled
What they log: Page, problem, severity from 1 to 4, suggested fix
Pages to cover: Homepage, top 3 landing pages, pricing or services page, every form
Output: One merged list, duplicates removed, sorted by severity
One reviewer is not enough, and that includes me. Nielsen and Molich (1990) ran four experiments and found that individual evaluators were "mostly quite bad" at the task. Pooling the reviews from a few people changed the picture.
"Individual evaluators were mostly quite bad at doing such heuristic evaluations and that they only found between 20 and 51 % of the usability problems," stated Jakob Nielsen and Rolf Molich (CHI 1990).
In practice I review the site myself, then ask two or three other people to do the same without seeing my notes: a designer, someone from the client's sales team and, when possible, a recent customer. Each logs problems separately. When I merge the lists, the overlap shows which problems are obvious and the unique entries show what one person would have missed. Sales staff are the best extra reviewer I know, because they hear the questions buyers ask on calls and know which pages prospects complain about.
The study tested interface designs of its time, not modern B2B sites. The pooling effect is the part I rely on, and it costs only a few extra hours of reviewer time.

3. First screen and offer: best for pages with high bounce rates
The first-screen check asks whether a visitor can tell, without scrolling, what you sell, who it is for and what to do next. It suits landing pages and homepages that lose most visitors at once. Lindgaard and colleagues found people judge visual appeal within 50 milliseconds.
Questions it answers: What is this, is it for me, what do I do next
Test: Show the screen for five seconds, then ask a colleague to describe it
Common miss: A clever headline that names no product, no buyer and no outcome
Fix order: Headline, subhead, one primary button, one proof line
The first screen is where the most visitors are, so it is where small changes move the most people. Lindgaard, Fernandes, Dudek and Brown (2006) showed that people form a stable judgement of a page's visual appeal within 50 milliseconds, and that their ratings at that speed matched their slower ratings closely.
That first impression is visual. In my five-second tests, the reason people give for staying or leaving is usually the wording, though the Lindgaard study did not test that. I run a five-second test on every key page: I show the screen to someone outside the company, hide it, and ask what the company sells and who for. If they cannot say, the headline fails, however good it looks. For B2B sites the fix is usually plain: name the buyer, the job the product does and the result, then give one button with one action.
"If they cannot say, the headline fails, however good it looks," stated Joshua Jackai, founder of Demand Designer.
The Lindgaard study measured visual appeal, not conversion. It explains why the first screen is checked first, not how much a new headline will lift leads.

4. Proof and credibility: best for long B2B sales cycles
The proof check maps each claim on the site to evidence a buyer can verify: named results, client logos, case studies, security pages and real people. It suits B2B deals with several reviewers. In Fogg's 2003 study, design look appeared in 46.1% of credibility comments from 2,684 people.
Claims to audit: Results, speed, expertise, security, price
Evidence types: Named case study, measure and time period, logo, review, certificate
Placement rule: Proof sits next to the claim it supports, not on a separate page
Common miss: Testimonials with no name, company or result
In Fogg's data, design look was the most frequently mentioned credibility cue. Fogg and colleagues (2003) asked 2,684 people to compare live websites and comment on why they trusted one over another.
"The 'design look' of the site was mentioned most frequently, being present in 46.1% of the comments," stated B. J. Fogg and colleagues (DUX 2003).
In my audits, design earns the read and proof earns the next step, but the study counted comments and did not test that order. In the audit I list every claim on the key pages and mark whether evidence sits beside it. Most B2B sites I review make strong claims on the homepage and keep the proof three clicks away on a case study page nobody opens. I move it next to the claim. My own proof follows the same rule, and these are first-party results from my projects, not independently audited: for Snappy Kraken, a Figma to HubSpot build shipped in 21 days and contributed to a 30% rise in MQLs and a 28% higher click-through rate.
The study used general web users judging sites on everyday topics, and it counted how often each cue was mentioned. It shows what people notice, not what closes a B2B deal.

5. Page speed: best for mobile and paid traffic
The speed check measures how long key pages take to show their main content on a mid-range phone, and finds what slows them. It suits sites buying paid traffic, where every slow second is paid for. Galletta and colleagues tested delays from 0 to 12 seconds with 196 people.
Tools: PageSpeed Insights, Chrome DevTools, field data from the Chrome User Experience Report
Pages to test: Top 5 landing pages by traffic, plus every page with a form
Usual causes: Large hero images, video backgrounds, chat widgets, tag managers loading 20+ scripts
Budget: Main content visible in under 2.5 seconds, Google's guideline for Largest Contentful Paint
Delay costs attention before it costs leads. Galletta, Henry, McCoy and Polak (2004) gave 196 people search tasks on sites with delays of 0 to 12 seconds. Performance, attitudes and intentions fell as delay rose, and the losses began to level off from about 4 seconds. A flat curve after 4 seconds does not make long waits safe. It means the study saw no further steep fall, and it cannot tell you how many visitors a 4 second page loses.
In audits I test the pages that carry money, not the homepage alone. Paid landing pages often load slowest, because they collect tracking scripts from every campaign that ever ran. I list each script, its owner and whether it still feeds a report anyone reads. Removing the dead ones is often the fastest speed fix on the site.
"Paid landing pages often load slowest, because they collect tracking scripts from every campaign that ever ran," stated Joshua Jackai, founder of Demand Designer.
The 2.5 second figure is Google's Largest Contentful Paint guideline, a vendor threshold, not a study result. The Galletta study used 2004 sites and tasks, so treat 4 seconds as a rough marker, not a safe limit.

6. Forms and booking: best for sites with traffic but few leads
The form check reviews every field, label, error message and step between "I am interested" and a booked meeting. It suits sites where people reach the form and stop. In Seckler's 2014 eye-tracking study of 65 people, forms rebuilt with 20 guidelines were completed faster with fewer submission attempts.
Fields: Count them, then justify each one against what sales uses on the first call
Labels: Above the field, never only as placeholder text
Errors: Shown next to the field, in plain words, without clearing other answers
Booking: Calendar offered on the thank-you page for qualified leads
Forms are where intent meets friction. Seckler, Heinz, Bargas-Avila, Opwis and Tuch (2014) took forms from real company websites, applied 20 usability guidelines and tested both versions with 65 people. The improved forms were faster to complete, needed fewer submission attempts and left people more satisfied.
"Improved web forms lead to faster completion times, fewer form submission trials, and fewer eye movements," stated Mirjam Seckler and colleagues (CHI 2014).
Length matters as well. Galesic and Bosnjak (2009) found that the longer a web questionnaire said it would take, the fewer people started and finished it, and Liu and Wronski (2018) saw the same pattern across 25,080 real surveys. Those are surveys, not lead forms, so I use them as direction, not as a conversion forecast.
What a form asks, and how, also changes what people share. John, Acquisti and Loewenstein (2011) found across four experiments that people's willingness to disclose responded to cues that bore little relation to the real risk.
"Disclosure of private information is responsive to environmental cues that bear little connection, or are even inversely related, to objective hazards," stated Leslie John and colleagues (Journal of Consumer Research, 2011).
So I check what surrounds the form: a privacy line, the reason for each qualifying question, and a clear statement of what happens after submit. My guide to how many form fields a B2B lead form needs goes deeper.

7. Five-user test: best for explaining what the numbers show
A five-user test asks a handful of people from the target audience to complete real tasks on the site while you watch. It suits audits where analytics show a drop but not the reason. Nielsen and Landauer's 1993 model, built on 11 studies, put the best benefit-to-cost ratio at four evaluations in their worked medium example.
Participants: 4 to 5 people who match the buyer, recruited per round
Tasks: Find pricing, compare plans, check a feature, request a demo
Format: Remote, recorded, about 30 minutes each, no help from the moderator
Output: Clips of each failure, linked to the funnel step that dropped
Analytics tell you where people leave. They rarely tell you why. Nielsen and Landauer (1993) modelled how usability problems are found across 11 studies and showed that each extra tester finds fewer new problems. For their medium example, they estimated 16 evaluations would be worth the cost, with the best ratio of benefit to cost at four.
That is why I run small rounds. Four or five people per round, one fix cycle, then another round if needed. For B2B sites, recruiting is the hard part: the right testers are busy managers, not general panel members. I use the client's own prospects and recent customers when I can, and I pay for their time. Each clip goes into the findings next to the funnel step it explains.
"Analytics tell you where people leave. They rarely tell you why," stated Joshua Jackai, founder of Demand Designer.
The model estimates how many problems a test finds, not how much a fix will lift conversions. Large sites with several distinct audiences need one small round per audience.

8. Test plan: best for teams that run A/B tests
The test plan check reviews how changes are tested and judged: sample size, run time, the metric that decides and when results are read. It suits teams with an A/B testing tool. Johari and colleagues showed that checking a running test repeatedly makes ordinary p-values unreliable.
Before launch: One primary metric, a minimum effect worth finding, a fixed sample size
During: No stopping early because a dashboard turned green
Low traffic: Use sequential methods, or ship research-backed fixes and watch the trend
Record: Every test, including losers, in one shared log
A testing programme that stops tests early can produce false wins. Johari, Koomen, Pekelis and Walsh (2017) worked with Optimizely data and described the problem: users of A/B testing tools "continuously monitor" their results, and traditional p-values and confidence intervals give unreliable answers when they do. Their fix, always-valid p-values, was deployed in that platform.
For most US B2B sites the bigger issue is traffic. A page that gets 40 demo requests a month cannot detect a modest lift in any reasonable time. When I find that, I recommend shipping the fixes that the research and the user tests already support, then watching leads and pipeline in the CRM month over month. Honest trend reading beats a test that was never going to reach an answer.
The paper is about the statistics of one platform. The rule it supports, decide your stopping point before you start, applies to every tool.

9. Lead follow-up: best for companies where sales and marketing disagree on lead quality
The follow-up check traces what happens after a form is submitted: routing, response time, qualification and whether the lead ever becomes a meeting. It suits teams where marketing reports leads and sales reports none. Sabnis and colleagues studied 461 sales reps across four firms.
Trace: Five recent leads from submit to first reply, with timestamps
Routing: Each form mapped to a named owner, with an alert on submit
Qualification: The fields sales actually uses on the first call
Feedback loop: Sales marks each lead good or bad, and marketing sees it monthly
A website can produce leads that nobody calls. Sabnis, Chatterjee, Grewal and Lilien (2013) studied 461 sales reps employed by four firms. They found that how much time reps gave to marketing leads depended on how leads were prequalified and tracked, and on each rep's experience. Their paper opens from a widely cited claim that 70% of marketing leads are never pursued; that figure is background in their paper, not their own finding.
"The proportion of time that sales reps devote to marketing leads depends on organizational lead prequalification and managerial tracking processes," stated Gaurav Sabnis and colleagues (Journal of Marketing, 2013).
In the audit I pick five recent leads and follow each from the form to the first reply. I note the timestamps, who got the alert and whether the CRM record had what the rep needed. The fix is usually not on the website at all: a routing rule, an owner, a better qualifying question. I still list it in the audit, because the website gets blamed for it. My guide to website lead tracking covers the tools.

CRO audit checklist by company stage
A CRO audit checklist is the nine checks in order: tracking, expert review, first screen, proof, page speed, forms, a five-user test, the test plan and lead follow-up. The matrix marks each as required, recommended or skip by monthly visits. Under 2,000 visits a month, start with tracking, the first screen and forms.
Tracking and funnel data: every form tested in analytics and the CRM
Expert review by three to five people, pooled
First screen: what, who for, what next, why believe it
Proof beside every claim
Page speed on the top 5 landing pages
Forms and booking: fields, labels, errors, privacy line
Five-user test on real tasks
Test plan with a fixed stopping point
Lead follow-up traced for five recent leads
Check | Effort | Under 2,000 visits a month | 2,000 to 20,000 | Over 20,000 |
|---|---|---|---|---|
1. Tracking and funnel data | Half a day | Required | Required | Required |
2. Expert review | 1 to 2 days | Required | Required | Required |
3. First screen and offer | Half a day | Required | Required | Required |
4. Proof and credibility | 1 day | Required | Required | Required |
5. Page speed | Half a day | Recommended | Required | Required |
6. Forms and booking | Half a day | Required | Required | Required |
7. Five-user test | 2 to 3 days | Recommended | Required | Required |
8. Test plan | Half a day | Skip | Recommended | Required |
9. Lead follow-up | Half a day | Required | Required | Required |
The effort figures are my own planning estimates from past audits, not a published benchmark. The traffic bands are also my own cut-offs, not research thresholds. A site with many forms or several audiences takes longer.
Calculator: which CRO audit checks come first?
The calculator ranks the nine checks for your site from four inputs: monthly visits, monthly leads, the share of leads sales calls back within a day, and whether you run A/B tests. It uses only your numbers and the stage rules in the matrix. The default values are placeholders, not benchmarks.
The scoring rule is simple: tracking always comes first, lead follow-up rises as the callback share falls, the first screen and forms rise when fewer than 1% of visits become leads, and speed, user tests and the test plan rise with traffic.
Should you hire a conversion rate optimization agency for the audit?
Hire a conversion rate optimization agency for the audit when nobody in-house can be neutral about the site or has time to run user tests. Keep it in-house when tracking is sound and a marketer can spend a week on it. Either way, insist on three to five reviewers, not one.
A conversion rate optimization agency is a service business that finds why visitors do not become leads or customers, then changes the site, offer and tests to raise that rate. I run Demand Designer, a B2B web design studio, so read this as one vendor's view. When you compare agencies, ask three things:
Who reviews the site, and how many people?
Do they follow leads into the CRM, or stop at the form?
Will they hand over a ranked list you can act on without them? An audit that only makes sense with the same agency's retainer attached is a sales document. My guide on how to choose a B2B web design agency has a longer checklist.
Frequently asked questions
How much does a CRO audit cost?
Four things set a CRO audit's cost: pages, forms, audiences and whether user tests are included. Vendor prices are rarely published, so I cite no market range. I run the audit as the first stage of a fixed-price build ($3,250 to $14,999 USD), not as a separate product.
How long does a CRO audit take?
In my projects, a CRO audit for a typical B2B site takes one to three weeks. Tracking, expert review, the first screen, proof, speed and forms take about five working days together. Two rounds of user tests add one to two weeks, mostly waiting for busy buyers' calendars.
What is the difference between a CRO audit and an SEO audit?
An SEO audit asks whether search engines can find, read and rank your pages. A CRO audit asks whether the people who arrive become leads and meetings. They overlap on speed and page structure, but a CRO audit adds forms, proof, user tests and lead follow-up.
How often should you do a CRO audit?
Run a full CRO audit once a year and after any redesign, platform move or big change in traffic source. Between audits, check tracking monthly by submitting every form yourself, and review the funnel each quarter. A drop in leads with steady traffic is the signal to audit sooner.
Can you do a CRO audit yourself?
Yes, you can run most of a CRO audit yourself if you can read analytics and the CRM. Ask two or three colleagues to review separately, because in Nielsen and Molich's study one reviewer missed 49% to 80% of problems. Neutral user tests are the hardest part alone.
How I run a CRO audit at Demand Designer
I run CRO audits as the first stage of a website project for US B2B companies, and I follow every lead from the first visit to the first sales call. The output is a ranked list of fixes with the page, the number and the reason for each.
The fixes then go into a fixed-price build, from $3,250 for a focused landing page set to $14,999 for a full site, so the audit and the work stay in one set of hands. For Advisor Websites, that approach contributed to a 58% rise in organic traffic and a 28% better MQL conversion rate.
If you would rather start alone, the free Pipeline Leak Scorecard walks you through 12 questions in about 15 minutes with your site open. It covers the same leaks as the nine checks, so your team starts from one shared list.
Sources
Peer-reviewed research
Fogg, B. J., Soohoo, C., Danielson, D. R., Marable, L., Stanford, J., and Tauber, E. R. (2003). How do users evaluate the credibility of Web sites? Proceedings of the 2003 Conference on Designing for User Experiences, 1 to 15.
Galesic, M., and Bosnjak, M. (2009). Effects of questionnaire length on participation and indicators of response quality in a web survey. Public Opinion Quarterly, 73(2), 349 to 360.
Galletta, D. F., Henry, R., McCoy, S., and Polak, P. (2004). Web site delays: How tolerant are users? Journal of the Association for Information Systems, 5(1), 1 to 28.
Johari, R., Koomen, P., Pekelis, L., and Walsh, D. (2017). Peeking at A/B tests. Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1517 to 1525.
John, L. K., Acquisti, A., and Loewenstein, G. (2011). Strangers on a plane: Context-dependent willingness to divulge sensitive information. Journal of Consumer Research, 37(5), 858 to 873.
Kohavi, R., Longbotham, R., Sommerfield, D., and Henne, R. M. (2009). Controlled experiments on the web: Survey and practical guide. Data Mining and Knowledge Discovery, 18(1), 140 to 181.
Lindgaard, G., Fernandes, G., Dudek, C., and Brown, J. (2006). Attention web designers: You have 50 milliseconds to make a good first impression! Behaviour and Information Technology, 25(2), 115 to 126.
Liu, M., and Wronski, L. (2018). Examining completion rates in web surveys via over 25,000 real-world surveys. Social Science Computer Review, 36(1), 116 to 124.
Nielsen, J., and Landauer, T. K. (1993). A mathematical model of the finding of usability problems. Proceedings of the INTERACT and CHI 1993 Conference on Human Factors in Computing Systems, 206 to 213.
Nielsen, J., and Molich, R. (1990). Heuristic evaluation of user interfaces. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 249 to 256.
Sabnis, G., Chatterjee, S. C., Grewal, R., and Lilien, G. L. (2013). The sales lead black hole: On sales reps' follow-up of marketing leads. Journal of Marketing, 77(1), 52 to 67.
Seckler, M., Heinz, S., Bargas-Avila, J. A., Opwis, K., and Tuch, A. N. (2014). Designing usable web forms: Empirical evaluation of web form improvement guidelines. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 1275 to 1284.
Industry and first-party sources (not peer reviewed)
Google. Largest Contentful Paint (LCP), web.dev. Vendor threshold, read 19 November 2026.
Search results for "cro audit", read 19 November 2026, including guides from Baymard Institute, Shopify and Invesp, mostly vendors writing about their own services.
Demand Designer price table, November 2026 (first-party prices, USD).
Client results for Snappy Kraken and Advisor Websites are first-party results from my own projects.
Free workbook
Pipeline Leak Scorecard
Find the leaks between your traffic and your pipeline. Score your site in about fifteen minutes, with it open in front of you.
12 questions, 8 pages, 15 minutes. Free.







