How to Run a Conversion Audit Yourself in One Afternoon

Before you pay an agency for conversion work, spend one afternoon finding out where visitors actually leave. The audit below costs nothing but time, and it tells you whether you need an agency at all. If you do, you’ll hire one with a specific problem instead of a vague hope. (My shortlist of CRO agencies is there for that day.)

Step 1: Write down the one action that matters

Pick a single conversion: a purchase, a booked call, a form submission. Not “engagement.” Then write the conversion rate as a number:

Conversions ÷ sessions × 100

Use the last 90 days, and split it by device. A site that converts 3% on desktop and 0.6% on mobile has a mobile problem, not a “conversion problem,” and the fix is completely different.

Step 2: Find the leaking step in GA4

In GA4, open Explore and build a funnel with three to five steps: landing page, product or service page, cart or form start, completion. You’re looking for the single step with the biggest drop. Ignore everything else for now.

Two checks before you trust the numbers:

1. Confirm the conversion event actually fires once per completion. A double-firing event makes every page look better than it is.
2. Filter out your own traffic and any obvious bot sessions.

If your tracking is broken, fix that first. Optimizing against bad data is how people “win” tests that change nothing.

Step 3: Watch 20 real sessions on the leaking page

Install Microsoft Clarity (free) or Hotjar on the page you found in step 2, wait until you have a few days of recordings, then watch 20 sessions from the leaking segment. Mobile first if step 1 pointed there. Write down every time someone:

• scrolls past the main button without touching it
• taps something that isn’t a link
• opens the form, then leaves
• zooms in to read text

Ten minutes of this beats a week of dashboard staring. You’ll usually see the same two or three problems repeat.

Step 4: Check the five usual suspects

Before inventing a clever test, check the boring things that cause most leaks:

1. Speed. Run the page through PageSpeed Insights on a mobile profile. If the main content takes more than a few seconds to appear, nothing else matters yet.
2. Message match. The headline should repeat the promise of the ad or search result that sent the visitor. If the ad says “free audit” and the page says “Welcome,” you lose them.
3. One clear action. One primary button, visible without scrolling on a phone.
4. Friction in the form. Every field you can delete, delete. Ask for the rest later.
5. Proof near the button. A real review, a client name you’re allowed to use, a guarantee, or the actual price. Whatever answers “why should I trust this?” belongs next to the ask.

If you build pages with a template tool, Instapage is one option for faster landing page variants. I compare the rest in landing page optimization tools.

Step 5: Change one thing and test it properly

Fix the biggest problem you found and change nothing else on that page at the same time.

How you test depends on your traffic:

• Under roughly 1,000 visits a month to the page: an A/B test will never reach a trustworthy answer. Make the change, record the date, and compare the same-length periods before and after, adjusting for obvious seasonality.
• Several thousand visits and a few hundred conversions: run a real split test in a tool like VWO, Optimizely or AB Tasty. Google Optimize was shut down in September 2023, so any guide still pointing you at it is out of date.

Decide the success metric and the minimum run time before you start. Stopping the moment the line goes up is the most common way to fool yourself.

Let Claude do the tedious part

Once you have your funnel numbers and session notes, paste this into Claude or ChatGPT:

I run a [type of business]. My goal conversion is [action]. Over the last 90 days: desktop converts at [x]%, mobile at [y]%. My biggest funnel drop is between [step A] and [step B]. From watching 20 session recordings I noticed: [paste your notes]. Give me a ranked list of the five most likely causes, the single cheapest change to test for each, and what result would tell me the change worked. Do not invent benchmarks; say when you are guessing.

Check the answer against your recordings before acting. These models are good at ranking hypotheses and bad at knowing your audience, so treat the output as a list of things to look for, not a diagnosis.

What to do next

Run the audit on your highest-traffic money page first. If the numbers show a clear leak you can’t fix yourself, that’s when an agency earns its fee. For more ideas on what to test, see ways to improve your ecommerce conversion rate and user optimization principles.