31 Mar 2026
When you open GA4 and look at your traffic, you're looking at an average. Organic, direct, paid, and job-seekers all lumped together. For a B2B SaaS company - or an agency like us - that average is close to meaningless.
Here's what we found when we actually cleaned the data: 67% of all traffic to the Polar Hedgehog website in January was going to the careers page. Not the services page. Not the homepage. The jobs section.
That means every channel metric, every engagement rate, every session count was being dragged down by people who had zero interest in hiring an agency. The real qualified traffic was 525 sessions - not the 1,597 GA4 was showing.
This is not a Polar Hedgehog problem. It's a standard B2B website problem. Career traffic, internal visits, bot traffic - they're all in your numbers by default. The first job is to take them out.
Once we excluded careers traffic, qualified sessions dropped from 1,597 to 525. Every channel metric, every engagement rate, every benchmark comparison changed. That was the real picture - and it was the accurate one.
We developed a five-step analysis process for B2B SaaS websites. It works whether you're analyzing your own site or a client's.
Step 1 - Segment and clean the data. Exclude career pages, identify anomaly spikes (viral posts, holidays, product launches), and separate new from returning visitors. You can't analyze what you haven't cleaned.
Step 2 - Compare against B2B benchmarks. Not generic web benchmarks - B2B SaaS-specific ones. A 25% engagement rate might be fine for a media site. For a company selling software to enterprise buyers, it means your messaging isn't landing.
Step 3 - Identify root causes, not symptoms. Low return visits is a symptom. No reason to come back is the cause. Zero conversions is a symptom. CTA appearing before any trust is established is the cause. The framework pushes you to one level deeper.
Step 4 - Map the B2B buying journey. B2B buyers need 3-5 touchpoints before they convert. The data should show whether users are progressing through that journey or dropping after visit one. Usually it's the latter.
Step 5 - Layer in Hotjar data. Heatmaps and session recordings answer the where and how. Users clicking on non-clickable client logos. Scrolling 60% down before seeing a CTA. Abandoning forms at the company size field. These patterns don't show up in GA4.
These are the numbers we use to evaluate a B2B website's health. Most of them are different from what you'll find in generic website benchmarks content.
| Metric | Weak | Average | Strong |
|---|---|---|---|
| Return visitor rate | <15% | 15-30% | >30% |
| Avg. sessions per user | <1.3 | 1.3-2.0 | >2.0 |
| Organic search engagement rate | <40% | 40-60% | >60% |
| Avg. engagement time | <20s | 20-45s | >45s |
| Demo / contact conversion rate | <0.5% | 0.5-2% | >2% |
| Career traffic as % of total | >50% | 20-50% | <20% |
We didn't use Claude to interpret exported files or pasted data. We used Claude in Chrome - Anthropic's browser extension - to give Claude direct access to the live GA4 and Hotjar interfaces. Claude navigated the reports itself: switching views, applying filters, reading retention charts, and analyzing the Hotjar heatmaps in real time.
That changes the workflow significantly. Instead of exporting data, cleaning it, and figuring out what to include, we just opened the right tabs and let Claude do the navigation. It moved through the Acquisition overview, applied the careers exclusion filter, pulled the retention report, and then switched over to Hotjar to read the heatmaps - all in one session.
The key is still the prompt. Claude in Chrome gives it eyes on the data - but without a structured framework telling it what benchmarks to use and how to connect observations to recommendations, you still get a summary rather than a diagnosis. The skill file below is what turns browser access into a proper CRO report.
Here's an example of the kind of prompt that gets results:
The Polar Hedgehog analysis surfaced four core problems.
Career traffic is masking everything. Once excluded, qualified sessions dropped from 1,597 to 525. Every channel metric looked worse - but that was actually the accurate picture.
Almost nobody comes back. 88.7% of visitors were new users who never returned. B2B average is 60-70% new. With a 14-30 day buying cycle, users need to return multiple times before they're ready to contact you. If they only visit once, the site has no mechanism for pulling them back into the funnel.
Users are clicking on things that don't go anywhere. The Hotjar heatmap showed heavy clicks on client logos - users expected to land on case studies and got nothing. This is a missed conversion moment that's invisible in GA4.
The CTA is asking for too much too fast. "Let's Talk" is a high-commitment ask for someone on their first visit. Changing it to something lower-friction - "See our work" or "Get a free audit" - lowers the barrier to that first micro-conversion.
The pattern we see repeatedly: sites aren't failing because of bad traffic. They're failing because there's no nurture mechanism for the traffic they have. Users arrive, don't know what to do next, and leave.
We've packaged the entire analysis framework as a Claude Skill - a structured prompt file you can use with your own website's data. If you have Claude in Chrome, you can give Claude direct access to your GA4 and Hotjar tabs and let it navigate the reports itself. If not, you can paste in screenshots or exported data and the framework still applies.
It includes the five-step framework, all the B2B benchmarks, the report structure, and instructions for interpreting Hotjar data alongside GA4. It works for agency owners, in-house marketers, and anyone running a B2B website who wants to make sense of their analytics.
A CRO (Conversion Rate Optimization) report analyzes your website data to understand how visitors behave, where they drop off, and what's stopping them from converting. For B2B websites, this is especially important because the buying cycle is long - buyers need multiple visits before they're ready to contact you. A monthly CRO report tells you whether your site is actually supporting that journey or losing people before they even enter the funnel.
GA4 shows averages across all your traffic by default - including job seekers visiting your careers page, internal team visits, and bots. For many B2B companies, career page traffic alone can represent 40-70% of total sessions, which inflates session counts and distorts every engagement metric. The first step in any honest B2B analysis is segmenting out this noise to see only the traffic that actually represents potential buyers.
Generic web benchmarks don't apply to B2B SaaS sites. The benchmarks that matter: return visitor rate should be 15-30% or higher (below 15% means people aren't coming back for the multiple visits B2B buying requires), engagement rate from organic search should be above 40%, average engagement time above 20 seconds, and demo or contact form conversion rate between 0.5-2%. Career traffic should be under 20% of total sessions - above 50% is a red flag.
The most efficient way is with Claude in Chrome, Anthropic's browser extension, which gives Claude direct access to your live GA4 and Hotjar interfaces. You open the relevant tabs, give Claude a structured prompt with the framework and benchmarks to follow, and it navigates the reports itself - applying filters, reading retention charts, switching to Hotjar heatmaps - without you needing to export or paste anything. If you don't have the extension, you can still use the skill by pasting in screenshots or CSV exports. The Claude Skill file included in this post gives Claude the full B2B CRO framework either way, so the output is a structured report with root causes and prioritized recommendations, not just a data summary.
GA4 tells you what happened - pages visited, sessions, bounce rate. Hotjar tells you why. Heatmaps show where users click (including clicks on things that aren't clickable, like logos or section headers), how far they scroll before leaving, and which CTAs are being ignored. Session recordings let you watch real user journeys. Combined with GA4 data, Hotjar turns a quantitative problem - low conversion rate - into a visual, actionable diagnosis.
A Claude Skill is a structured prompt file that teaches Claude how to approach a specific type of task - in this case, B2B website CRO analysis. Instead of starting from scratch each time, you load the skill into a new Claude conversation and it applies a consistent framework: clean the data, compare against B2B benchmarks, identify root causes, map the buying journey, and produce a prioritized action plan. The skill we're sharing in this post is free to download and works with any Claude plan.
Norta works with a small number of startups to ensure focus, quality, and true partnership.
Limited spots available
