Leads are down. Open Figma at this point, and you'll probably find the problem in the design. An outdated form. A weak button. Too many fields. Convenient: you already know what to sell the client.
There's one thing about this approach that bothers me. We don't know what happened to the leads yet.
A website UX audit should help answer that before anyone redesigns the interface. Where are people getting stuck? What's stopping them? And is there any reason to think a new form would change the outcome?
I've left out company names, industries, and identifying details. I care about my clients and the trust they put in me. So I'll stick to what matters here: the original request, what we found, and why the work needed to go in a different direction.
The site looked old. That wasn't why the leads disappeared
One recent audit started with a familiar request: fewer leads, time to freshen up the site, nothing too drastic. The interface certainly gave us things to criticize. It would have been easy to start on a new hero section and make a convincing case for it.
We started with the traffic history instead. That showed when search traffic dropped and what had happened to the site beforehand. An update had changed headings and introduced a lot of similar pages. We compared the analytics, an archived version, and the history of those changes. “The site isn't working anymore” became a specific period to investigate and a list of things to check.
That didn't prove any single change had caused the drop. But one thing was clear: far fewer people were arriving from search. A new hero section wasn't going to help the people who never reached it.
Then we found problems affecting the people who did arrive. Some URLs weren't working properly. The form handler accepted empty submissions. Lead tracking needed a closer look. The client also confirmed that inquiries weren't coming through reliably. A form reporting success didn't establish that the inquiry had reached the person who needed to handle it.
In that situation, I'd first work on getting people back to the site and making sure their inquiries reached someone. Then check what we were actually counting as a lead. After that, we could come back to the interface with something solid to work from. Otherwise, we could end up with a refreshed exterior and the same problems underneath.
Sometimes the main way into the site dries up
On another platform, image search was the main entry point into the funnel. When that traffic fell sharply, the site lost the people its whole process started with. Search wasn't a bonus channel to deal with after the redesign.
The next question was who we'd lost. Someone might arrive looking for an image, view it, and leave. Someone choosing a service provider has a different task. Those visits can sit next to each other in a report even though the visitors need very different things from the product.
So the loss of search traffic and the difficulty of turning visitors into customers needed separate investigation. Where did the decline start? Which source changed? Which pages stopped bringing people in? What was happening to registrations, inquiries, and payments at the same time? Reduce all of that to a button color, and half the story disappears.
In this case, product activity had started declining before the sharp drop in search traffic. Restoring the main entry point needed its own work, but it wouldn't answer every question about the product. We had to untangle both developments: what started falling first, what happened to search, and how losing that traffic made the decline worse.
Before fixing conversion, find out what you're counting
Another product's website appeared to perform much worse than its app. A closer look showed that similar labels in the reports referred to different actions. The app's funnel started with an install. The website comparison included every visit, including people who had only come to read content.
Then we had to untangle the tracking itself. Visits to a page for adding funds to an account were being used to assess the path to payment, even though people could pay another way. There were event-tracking problems, too. For example, a function for sending a Meta payment-related event existed in the code, but we found no calls to it in the code we inspected.
The pixel was installed. The function was written. The event names looked familiar. And the funnel step still wasn't being counted. You wouldn't catch that by checking whether analytics was installed.
We also had to revise some of our own findings during the audit. At first, certain events appeared to be missing entirely. A broader check showed that they existed; we'd been looking in the wrong part of the flow. That's part of an auditor's job, too: test your own explanation before handing it to a developer as a confirmed bug.
Even after that check, I wouldn't rush to announce the “real” conversion rate. First, agree on the action that counts as a result. Verify that its event is sent. Reconcile those events with actual transactions. Otherwise, you can spend a long time improving a metric that was wrong to begin with.
Now we can open the site
I'd start where the data points and walk through the task on a phone. Logged out. No saved addresses. No insider knowledge of where everything lives.
That's harder than it sounds on your own product. We know the shipping information is in the footer. We know which phone number format the form accepts. We stop noticing the button hidden behind the keyboard. Nobody gave the visitor that training.
Try getting something wrong. Enter a phone number in the wrong format, go back from a product page to the catalog, submit a form on a poor connection. See what survives, where the message appears, and whether you can continue without filling everything out again. This is where an interface has to do more than look good in a presentation.
The usual usability heuristics help here. Can you see the status of an action? Do you know what the button will do? Can you recover from an error? I just wouldn't turn them into a standalone report with dozens of pages of observations. Each finding needs somewhere to go: here's the task, here's the obstacle, and here's what we should test.
If you already have a suspicious part of the journey, send me the link for a review. Tell me what isn't adding up: leads, orders, support requests. That's a more useful starting point than “make it look more modern”
It's too easy to explain your own interface for the user
Once the measurement makes more sense, you can test the task with people. It's easier to let go of a decision while it's still a prototype. Once the screens, components, and some of the code are done, “does this work?” quietly picks up a second question: “but what about all the work we've put in?”
For a typical qualitative usability study involving one user group, NN/g recommends starting with five participants. That helps you find problems. It doesn't turn observations into statistics for the entire business. If people have different roles and tasks, testing one group won't tell you how everyone else gets along.
The task wording matters, too. “Find the shipping terms” already tells someone what to look for. Give them a situation instead: you need the item by a certain date. Can you get it in time? Then try to stay quiet. Especially when they sail past the thing you designed so carefully for them.
What I want an audit to leave behind
You can't immediately put a dollar amount on every problem you find. That's fine. Replacing an unknown with a confident number just to finish the spreadsheet isn't.
In the first audit, we could already separate the work: check whether inquiries were reaching the recipient, fix reproducible technical problems, and investigate lost search traffic. Each task had different evidence behind it and needed a different check. Promising that a new heading or layout would bring everything back would have been premature.
I want a result that makes the next move clear. If the form reports success, follow the inquiry all the way to the recipient. If a graph drops after a release, reconstruct what changed and test possible explanations. If an event only exists on paper, find out when it's actually sent and what action triggers it.
I'd be particularly careful with “lost revenue” calculated by multiplying every abandoned cart by the average order value. That calculation rests on a very comfortable assumption: everyone was going to buy. The audit still has to establish that.
A useful result lets you choose the next piece of work. Fix a reproducible bug. Repair the tracking. Talk to people who decided against buying. If a larger rebuild is necessary, explain why smaller changes won't be enough.
Where I come in
I look at the journey and the implementation together: what people are promised, what they encounter along the way, and what happens after they act. Then we can separate a problem we've established from an explanation we're still testing.
The scope depends on the product, the user roles, the available data, and whether we can test with users. Without that context, a price would be a quote for work we haven't defined. Sometimes walking through one task and reproducing a bug is enough. Sometimes the measurement needs work first.
Send me a link to your product. I'll review what's keeping people from completing their task and what makes sense to fix first
Sources and limitations
These examples come from actual project audits. Company names, industries, exact dates, metrics, and other identifying details have been omitted. These are accounts of diagnosis. I'm not claiming that the recommended changes have already shipped or produced a measured improvement.
For the recommendation on participant numbers in a qualitative study, see NN/g, Usability Testing 101, read November 4, 2026.






