Changing something on your website is the easy part. Knowing afterwards whether it worked is where most people get stuck.
The usual version goes like this. You rewrite a page, cut a form down, move a button. A week later you open your analytics and some numbers are up, some are down, and none of them answer the question you actually have. Time on page fell. Is that bad, or did people find what they needed faster? Enquiries dropped, but the ones arriving feel better. Is that a loss or a win? Nothing in the dashboard settles it, so the call gets made on instinct, and the change gets kept or reversed for reasons nobody writes down.
Two things make this harder than it looks.
The first is that the metrics on your screen were not designed for your page. Time on page, bounce rate and sessions mean different things on a pricing page, a specification table and a blog post. Your analytics reports them identically on all three, and with the same confidence.
The second is that the changes worth making often cost you something visible. If a change filters out people who were never going to buy, your headline number falls first. On a monthly report that reads as damage, and it usually gets undone before anyone finds out it was working.
So rather than add another opinion to the pile, I went and asked the people who have already been through it.
What I asked, and what came back
In September 2026 I put one question to people who build and run websites for a living. Name a user experience change you made that had a real effect, tell me how you measured it, and tell me what you would say to someone about to try the same thing. Thirty three people answered.
I set aside the responses that named no specific change or no way of measuring it, the ones repeating an angle already covered, and seventeen that shared a writing style, a structure and several identical sentences across supposedly unrelated businesses. Twelve are quoted below, as submitted, trimmed for length where marked.
I expected a list of wins where every arrow pointed up. That is not what came back. Four people, in four unrelated industries, independently described the same uncomfortable thing: the change that worked made their headline number go down.
That is the part nobody writes about, so that is where this starts.
The change that looks like a failure
RHILLANE Ayoub runs a digital marketing agency. For years his site did what agency sites do, which is describe the work in detail and then ask you to get in touch to discuss investment. He replaced that with honest price ranges, what sits behind each one, and a plain line about the size of business each suits. He also said who the agency is not right for.
Enquiry volume went down and booked calls went up. That is the part people need to hear, because if you only watch form submissions you will read the change as a failure and undo it.
RHILLANE Ayoub, RHILLANE Marketing Digital
Read the order of that twice. Volume fell. Booked calls rose. A dashboard counting form submissions would have recorded that month as damage.
I have been on the wrong side of the same fight. Most of my clients had an investment page that asked people to get in touch for a full pricing guide, and in every analytics account I opened, that page was where visitors were leaving. I started insisting on a plain starting figure near the top instead, even a rough one.
Enquiries dropped by roughly a third on the first site we tried it on. The number that turned into booked calls went up, because everyone filling in the form had already decided the budget worked. I checked it with form submissions in GA4 set against the client’s own record of which enquiries became paid work, and I looked at the exit rate on that page in Clarity before and after, so I knew the drop was coming from the right place.
Nassira Sennoune found a third version of this at Originn Properties. Listings showed a price, and the fees, registration and notary costs lived on a separate page most buyers never reached. She put an estimated all-in figure on the listing itself, with the components named.
Her enquiry volume did not move at all. What changed was the content of the messages.
Before, most messages were some version of what are the extra costs. After, people arrived asking about the property itself, the orientation, the service charge, when they could visit. Same volume of enquiries, much further along.
Nassira Sennoune, Originn Properties
Three businesses, three countries, same lesson. If you publish the number people are afraid to ask about, you lose the enquiries that were never going to buy. Whether that reads as a win or a disaster depends entirely on what you decided to count, and you have to decide before you make the change, not after.
Choose the metric before you touch the page
Richard Meadows at Streamrise makes the sharpest version of this argument. Their guides used to open the way guides do, with context and why this matters and what we will cover, and the actual specifications sat several screens down. He inverted it. The numbers now sit directly under the first heading as a table, and the context got demoted to underneath.
The interesting part is what he refused to measure.
Not with time on page, which we had been treating as a quality signal and which is close to useless here. A visitor leaving quickly because they got their answer looks identical to one leaving because the page was bad.
Richard Meadows, Streamrise
He used scroll depth to the answer block and the rate of return visits from the same query instead, on the reasoning that a reference page is doing its job when people come back to it.
Work out what a successful visit looks like for that specific page before you choose the metric, because the default metrics quietly assume that longer is better. For a reference page, longer is worse. For a decision page, longer might be better. Using one dashboard definition of engagement across both will point you in the wrong direction on at least one of them, confidently.
That last word is the one that should worry you. The dashboard is not unsure.
Eric Turney at The Monterey Company gets to the same place from the opposite direction. He made order and proofing status visible on the page and in confirmation emails, so customers could see where a custom order stood without emailing to ask. The result showed up as an absence.
Measure impact by what stops happening, not just what starts happening. A drop in repetitive customer questions is often a better signal that a UX change worked than any engagement metric, because it means people found what they needed without friction.
Eric Turney, The Monterey Company
If you want a single habit out of this article, it is that one. Write down what you expect to go up and what you expect to go down before you publish, and whether you are willing to accept the trade. Your analytics will not make that judgement for you.
The changes where the numbers do move the obvious way
Not everything is a trade. Several of these are straightforward subtraction, and the effect is immediate and uncomplicated.
Joe Spisak deleted the homepage at Fulfill.com. The hero image, the mission, the explanation of what a third party logistics provider is, all of it. He replaced the lot with one question, “What are you shipping?”, an input field and four buttons.
Click-through rate to our matching tool went from 12% to 59%… Our Net Promoter Score went from 31 to 68 in three months.
Joe Spisak, Fulfill.com
He measured it three ways: click-through to the tool, completed match requests, and follow-up surveys asking whether people actually found a good provider. That third one matters, because the first two can both rise while the product underneath fails.
Stop explaining and start doing. Most sites over-explain because founders are insecure about whether visitors will understand… The best UX improvement is usually subtraction, not addition.
Jason Bland of Custom Legal Marketing did the same thing to forms. Law firm contact forms had grown to nine fields asking for case type, how you heard about us, preferred contact time. He cut them to three: name, phone number, brief message.
Some clients saw a 40% increase in form submissions within the first 30 days.
Jason Bland, Custom Legal Marketing
He tracked it through Google Analytics goal completions compared month over month, but the reason he made the change in the first place is the better story. He was watching Hotjar session recordings and could see people start typing, hit a confusing field, and leave.
It was painful to watch… Stop asking for information you don’t actually need on first contact. Your intake team can gather details during the initial phone call. The website’s job is to get someone to raise their hand, not to pre-qualify every lead before a human ever speaks to them.
Victor Smushkevich of Tested Media arrived at the same three fields, and then pointed past them. Session recordings on his clients’ sites showed most visitors scrolling straight to the phone number, so he put a click to call button above the fold on mobile.
The form was often just decoration… Measure time from click to first response before touching layout or colors. Fix the follow up speed first, then simplify the ask. Most sites optimize the wrong half of that chain.
Victor Smushkevich, Tested Media
Callum Gracie of Otto Media gives the shortest usable rule in the whole set: restructure high traffic pages around the visitor’s first question rather than your preferred introduction. Direct answer first, supporting proof second, next action third. He compares engagement time, onward paths and completed enquiries, and specifically warns against leaning on bounce rate alone. It is the same principle behind ordering your pages around what visitors are looking for.
Remove any opening paragraph that delays the answer visitors came to find.
Callum Gracie, Otto Media
If it is a product, measure the first five minutes
Three of the twelve run software, and all three independently stopped looking at traffic.
Runbo Li built Magic Hour, an AI video tool, as a two person team. Early users landed on an open prompt box and froze.
Engagement cratered. People don’t want infinite possibility, they want a starting point.
Runbo Li, Magic Hour AI
Templates replaced the blank canvas. He measured it with funnel analytics plus cohort retention, tracking how many day one signups exported a video in their first session and then watching the 7 and 30 day curves, rather than celebrating a single number.
Watch five users try your product for the first time without any guidance. Don’t explain anything. Just watch where they hesitate, where they pause, where they squint… The answer is almost never ‘add more options.’ It’s almost always ‘make the first win faster.’
Kruno Sulić shortened the path from signup to first result on Cliprise and found the problem was not interest.
They were dropping off because the product asked them to make too many decisions too early.
Kruno Sulić, Cliprise
His metrics are worth copying wholesale: completion of the first key action, time to first successful output, the percentage of new users who reach a second session, and the volume of support questions tied to onboarding confusion. That last one is Eric Turney’s point again, arriving from a completely different industry.
Ian Lawson at Slickplan broke up onboarding entirely, replacing the front loaded tour with short prompts that appear at the moment each feature is needed, with feedback and bug reporting kept close by so people can say what confused them while it is still fresh.
Good onboarding should demonstrate the next action, not explain the entire product.
Ian Lawson, Slickplan
Finding the thing worth fixing
Every person above found their problem by watching rather than guessing, and none of them found it in a report.
Niclas Schlopsna, managing partner at spectup, reviews a lot of early stage companies and sees the same failure constantly: a polished homepage with the actual message buried under jargon. His fix is unglamorous, which is rather the point.
Watch someone unfamiliar with your business use your website without helping them. Notice where they hesitate, what they ask, and what they miss. Those moments often reveal more than another round of internal design discussions.
Niclas Schlopsna, spectup
Jason Bland found his nine field problem in session recordings. Nassira Sennoune found hers in the enquiries themselves, by noticing that nearly every message asked the same question. RHILLANE Ayoub found his on sales calls.
Pick the objection your sales calls handle over and over and answer it on the page instead. If a question comes up in every single call, the page is not doing its job, and the call is doing work the page should have done.
RHILLANE Ayoub, RHILLANE Marketing Digital
Four different inputs, one method. Go and look at where people are already telling you what is wrong, then put the answer on the page. It is cheaper than a redesign and it tends to be more accurate, because it comes from people who had no idea they were giving feedback.
What I would take from all twelve
If you are about to change something on your site, three things.
Decide in advance what you expect to go down. A change that removes the wrong enquiries will look like a bad month before it looks like a good quarter, and almost nobody holds their nerve through that without having written the expectation down first.
Pick the metric for that page, not for the site. Time on page means the opposite thing on a specification page and a sales page, and one dashboard setting will confidently mislead you about one of them.
Then watch for what stops happening. Fewer confused calls, fewer repeat questions, fewer people abandoning the form halfway. Those are not the metrics anyone puts on a monthly report, and they are usually the first sign that something is genuinely working.
The advice I would add to the pile is the one thing I had to learn the hard way. If you are doing this for a client, agree in advance that fewer enquiries is an acceptable result. That belongs in the conversation you have before the work starts, not after it. If you do not have that conversation first, the first quiet week undoes the whole thing.