Skip to content

How do you know your last digital journey change worked?

Most teams don't. They launch and hope. We show you how to try changes on real customers before you commit, then teach your team to do it without us.

Check your numbers (free · instant · no email needed) Or ask for a free review

We have been the client: ten years in house, five agency side · Nine markets across Europe and North America

01 · For you
02 · With you
03 · By you
04 · Gone

Stop deciding what to change by whoever pushes hardest.

The redesign went live and sales dipped.

Was it the design, the season, or the price rise that went out the same week? Nobody can say, so nobody changes anything.

The homepage changes when the director does.

Each new senior hire brings a view of what the site should say. Each view is reasonable. None of them gets checked.

The new feature "performed well".

Compared with what? A report that only covers the weeks after launch cannot tell you what would have happened without it.

Forty ideas and one meeting to choose.

Everyone has a list. The list gets settled by whoever is most senior or most persuasive, not by what customers actually do.

Nobody here is doing anything wrong. This is how decisions get made when there is no way to check them.

There is a way to check

Try the change on half your customers first.

Half your visitors see what is currently there. Half see the change. After a few weeks you keep whichever did better, and you know it was the change that did it, not the weather or the price rise.

This is called experimentation, or A/B testing. Large online businesses do it every week. It works for smaller ones too, with one honest condition: you need enough customers to see a difference.

Everything a good test needs, in one place.

A test is only as good as the research behind it, the design in it, the build and the numbers that judge it. We cover all four, bringing in specialists when the work needs them, and teach your team each one as we go.

User research

Talk to customers and watch them use your site, so you test what is actually getting in their way.

UX and UI design

Design the changes worth testing, and finish the ones worth keeping so they are ready to ship.

Analytics and insights

Set up tracking you can trust, then turn the numbers into decisions rather than another report.

Development

Build the tests ourselves, so they do not sit in a queue behind your product roadmap.

Need one of these on its own? We do that too, on the same terms: your team learns it, then we step back.

Five questions to answer before you start.

Get these right at the beginning and testing becomes part of how your team decides. Get them wrong and it becomes another tool nobody uses.

  1. 01Do we have enough customers?

    What matters is how many people buy, book or sign up, not how many visit. With fewer customers you can only spot big differences, which is fine if you test big ideas. The check below tells you in ten seconds.

  2. 02Do we need special software?

    Eventually, but not first. Testing rarely fails because of the software. It fails because nobody owns it. Decide who will run it, then choose a tool that suits them.

  3. 03Should we hire a specialist?

    Usually not yet. Give it to someone who already understands the business, and protect two days a week of their time in writing.

  4. 04What would we test?

    The things your leadership already argues about, not what is easiest to change. Your first list is sitting in last quarter's meeting notes.

  5. 05Will it pay for itself?

    Sometimes not. Work out what it would cost to run against what it could return before you spend anything else. If the numbers do not add up, the right answer is to wait.

Free · instant · no email needed

How long would a test take on your site?

Two numbers from your analytics. We will tell you how long you would wait to know whether a change worked, and whether testing is worth it for you yet.

That is a 2.0% conversion rate. Use whatever counts as a sale for you.

Smallest change you could spot in six weeks

11%

A 10% change would take 7 weeks to spot. We never run a test longer than six weeks, so the change has to be bigger.

Test bolder changes: ones you expect to move results by 11% or more.

Half your visitors see each version, with 95% confidence and an 80% chance of spotting a real change. We cap tests at six weeks: run longer and returning visitors start to blur the result. A guide, not a promise.

If your numbers work

A free digital review

Tell us your site and roughly how many orders, bookings or sign ups you get a month. The person who would do the work looks at both, then talks you through what they found in a 30 minute call.

  • Whether you have enough customers for testing to work yet
  • Three changes on your site we would test first, and why
  • Roughly how long a test would take at your numbers
  • An honest recommendation, even if it is "not yet"
What the answer can sound like
At your traffic you would wait five weeks to detect anything smaller than a ten percent change. Testing is not your next move.

We would rather tell you that now than sell you something that cannot pay back.

When the answer is yes

From your first test to running it without us.

Four stages, always in this order, with no months attached. We move on when your team is ready for the next one, and we leave when they pass the standard below.

  1. For you

    We set everything up and run your first tests while your team watches. The aim is early proof, not teaching.

  2. With you

    Named people in your team take pieces of the work. We review everything in the open.

  3. By you

    Your team runs it. We sit in and say nothing unless something is wrong. Most of the learning happens here.

  4. Gone

    We check your team against the standard, write the handover, then we stop. If they are not ready yet, that is our problem to solve, not yours to pay for.

What real tests look like, written up.

What we expected, how many people saw it, what happened and what was decided. Most tests do not win, so we show the losses too. Sector and journey only. Never the client.

Eight real tests · scroll or swipe for more

Global automotive manufacturer · ConfiguratorWin

Shorter, plainer wording on the button to start configuring a car

We expected fewer, plainer words to get more people started, because the next step would look less complicated.

People in each half
29,300 vs 27,500
Ran for
4 weeks
Started configuring
+6.3%
Finished configuring
+6.7%

Decision. The authoring team rolled the new wording out across Europe.

Global QSR ecommerce business · MenuLoss, then a win

Showing the price on meal deals

Prices on the deal buttons meant fewer people clicked into deals. So we moved the price into the deal pictures instead, and orders went up.

First test, deal clicks
−1.12%
Second test, orders
+1.08%
Confidence it was real
99%
Ran for
4 weeks each

Decision. Keep the second version. The first test told us where the price belonged.

Multi brand UK pub and dining group · BookingNo difference

A "why book with us" banner on the homepage

Bookings were the same with and without it. That is still useful: without the test, someone would have credited the banner for the next good month.

People in each half
86,800
Ran for
11 days
Bookings
No change
Success agreed beforehand
Yes

Decision. Try a bolder change rather than run this one longer.

Multi brand UK pub and dining group · BookingWin

A two step prompt for people who look ready to book

We showed a short prompt only to visitors whose behaviour suggested they were planning a visit. More of them started a booking, and more finished one.

People in each half
158,000
Ran for
15 days
Bookings started
+2.7%
Bookings completed
+2.9%

Decision. Keep it, and refine the wording in the next round.

UK big ticket retailer · BasketLoss

Moving the delivery postcode box above the checkout button

We expected that seeing delivery details earlier would reassure people before they paid. Instead, far fewer of them went on to checkout.

People in each half
9,370
Ran for
2 weeks
Reached checkout
−14%
Checkout rate
45.4% to 39.0%

Decision. Stop the test and keep the postcode box below the checkout button.

UK big ticket retailer · Product listingsWin

Letting shoppers switch between monthly and total price

Big purchases are often paid monthly. We expected that letting people choose how prices are shown would get more of them to add an item to their basket.

People in each half
183,000
Ran for
2 weeks
Added to basket
+3.6%
Add to basket rate
6.35% to 6.58%

Decision. Keep the switch, and keep testing how the two prices are shown.

Multi brand UK pub and dining group · MenuLoss

Filtering the menu by dietary needs

It looked like an obvious improvement for customers. It cut bookings, and the result was clear enough to act on.

People in each half
108,000
Ran for
19 days
Bookings started
−4.4%
Bookings completed
−4.0%

Decision. Drop it. Helpful for customers is not always helpful for bookings.

Global automotive manufacturer · Test drive bookingLoss

Removing the comments box from the test drive form

Fewer fields usually means more completed forms. Here it did not: bookings dipped slightly without the box, so there was no case for removing it.

People in each half
6,810 vs 7,020
Test drives booked
−0.9%

Decision. Keep the comments box.

For teams already testing

Running tests, but nothing really changes?

That is rarely a testing problem. Score your programme against the seven bars we hand over against, or ask us for a programme diagnostic to find what is actually wrong. Usually it is ownership.

Ask about a diagnostic

Running an agency? Same product, delivered with your team under your brand. We work under your brand and we do not approach your clients, during or after.

Who you will work with

An independent experimentation consultancy. We build the capability inside your team, then we go.

We have done this from both sides of the table. Ten years in house, holding the budget and managing the agency. Five years agency side, doing the work.

When a project needs research, design, analytics or development, we bring in specialists we have worked with before, for as long as the work needs them and no longer. There is no bench to keep busy, which is why we can promise to leave.

In house at a motorcycle manufacturer, a builders’ merchant and a premium car brand.

Start here

Ask for a free review. We will tell you if testing is worth it yet.

The person who would do the work reads every message and replies within two working days, with a question rather than a calendar link.

  • A reply from the person who would do the work
  • An honest answer, including "not yet"
  • No sales call unless you ask for one

Prefer email? hello@learnedbehaviour.co.uk