Why a Design-Only Website Redesign Gets No Results: 5 Layers That Move Conversions, and How to Measure Visual Changes with A/B and Before/After Tests (2026)

"We redesigned the website, but enquiries didn't go up." It is one of the most common things we hear when a business asks about a redesign. The new site looks polished and colleagues like it. Yet the numbers barely move.

Most of the time, the cause is that only the visuals changed. A visitor finds you in search, opens the page, reaches what they were looking for, is convinced and gets in touch. That sequence depends less on colours and photos than on how information is organised, what the content says, how fast the page is, how the path to contact works, and whether any of it is measured.

This article uses published research and official guidance to show which of these five non-visual layers move results. The second half explains how to check whether a visual change actually worked, using A/B tests and before/after comparisons, with sample size calculations.

Main sources for this article

• web.dev speed case studies (Rakuten 24, The Economic Times, Vodafone, Deloitte's "Milliseconds Make Millions") and the Core Web Vitals thresholds
• Google Search Central guidance on A/B testing, page experience and the SEO Starter Guide
• Nielsen Norman Group articles on tree testing, card sorting, the aesthetic-usability effect and redesign decisions
• Baymard Institute checkout research (number of form fields, reasons for cart abandonment)
• Microsoft's 2009 paper on online experiments and Kohavi and Thomke's 2017 Harvard Business Review article

What this article covers	Short answer
Why design alone doesn't deliver	Results depend on five non-visual layers (section 1)
Where to start	Work back from the number you want to move (section 2)
What to do in each layer	IA, content, speed, flow and forms, measurement (sections 3–7)
How to measure visual changes	A/B tests and before/after comparisons (sections 8–9)
Cost and time	Effort ranges per layer (section 10)

1. Why a design-only redesign doesn't move the numbers

This is not an argument that visual refreshes are pointless. Visuals have a job to do. The problem is expecting too much of them while leaving every other layer exactly as it was on the old site.

Looks are judged in an instant and affect trust

In 2006, Lindgaard and colleagues at Carleton University in Canada reported that ratings of a web page's visual appeal after 50 milliseconds correlated strongly with ratings after 500 milliseconds. Visual impressions form almost instantly.

In a 2002 Stanford study led by B.J. Fogg with 2,684 participants, 46.1% of comments about a site's credibility mentioned its look (layout, typography, colour and so on). The share was 46.2% for e-commerce sites and 54.6% for finance.

Study	Finding	Caveat
Lindgaard et al. (2006)	Visual impressions form in 50 ms	Did not measure staying or behaviour
Fogg et al. (2002)	46.1% of credibility comments mention look	Share of comments, not people
Kurosu & Kashimura (1995)	Attractive designs feel easier to use	Not the same as actual ease

Looks are the entry point to first impressions and trust. But none of these studies shows that changing the look increases enquiries.

Attractive design can hide usability problems

Nielsen Norman Group (NN/g) describes the aesthetic-usability effect. Masaaki Kurosu and Kaori Kashimura of the Hitachi Design Center studied it on ATM screens in 1995: when an interface looks attractive, people perceive it as easier to use than it really is.

In a redesign this becomes a trap. When colleagues review the new design and say it is "easier to use", that may simply be the visual impression talking. NN/g gives an example of a test participant who struggled with navigation yet praised the site's colours and photos. You need to watch what people do, not what they say.

Results depend on a chain: found, wait, find, convinced, contact

An enquiry or a purchase happens only after several steps. If one step is blocked, no amount of visual polish further down will raise the numbers. Figure 1 shows which layer drives each step.

Visuals take the lead in only part of one step out of five: being convinced. At every other step, a new look leaves the old bottlenecks in place.

2. The big picture: five layers that drive results, and the order to fix them

Before setting the scope of a redesign, match the number you want to move to a layer. The table below lists the five layers plus visuals, with what changes, which numbers they tend to move and how to check them.

Layer	What you change	Numbers it moves	How to check
Information architecture	Page grouping, order, labels	Reaching target pages	Tree testing
Content	Copy, cases, pricing, FAQ	Search traffic, reading	Search Console
Speed	Images, scripts, server	Bounce rate, CVR	Core Web Vitals
Flow and forms	Button placement, fields, steps	Enquiry completion	Funnel analysis
Measurement	Recording key actions	Decision accuracy	Test submissions
Visuals	Colour, type, photos	First impression, trust	A/B testing

Each non-visual layer has its own measuring tool. Having a tool means you can compare before and after.

Small fixes or a full rebuild?

In a 2015 article, NN/g's Hoa Loranger noted that many redesigns start for reasons such as the site "looks old" or competitors seem to be doing better, and often fail to justify their cost. Her advice is to try the smallest change that solves the problem.

A full rebuild is still the right call in some situations. The decision table below follows the conditions listed in the same article.

Situation	Suited approach	Reason
High drop-off on specific pages	Targeted fixes	The cause is narrow
Numbers are fine, design looks dated	Targeted fixes + test visuals	Switching costs are high
Small fixes have stopped paying off	Consider a rebuild	Structural limits
CMS or tech past end of support	Rebuild	Cannot run it safely
Very low enquiry rate across the site	Rebuild + research	Core design problem

For help deciding whether you need a redesign at all, see 10 signs your website needs a redesign. This article focuses on what to change once you have decided to redesign.

Work back from the number you want to move

When choosing layers, narrow the target to one or two numbers. The UK government's GOV.UK Service Manual also recommends narrowing to 3 or 4 key performance indicators once a service matures, and keeping the old figures as a baseline.

Number to move	Look at first	Look at next
Enquiries	Flow and forms	Content (cases, pricing)
Search traffic	Content	IA, speed
E-commerce conversion	Speed, forms	IA (product findability)
Job applications	Content (the work itself)	Flow
Phone calls	Flow (number placement)	Measurement (tap tracking)

Once the number is set, visual debates can be settled by asking whether a change helps that number.

3. Information architecture: a structure people can navigate

Information architecture (IA) is how you divide pages, order them and name them. If the menu groups and labels stay the same, visitors get lost in the same places, whatever the new design looks like.

Signs of an IA problem

Look at IA if you see the following. All of them can be found in Google Analytics 4 (GA4), on-site search logs or the text of enquiries.

• People type words into site search that already appear in the menu
• Visitors to service pages bounce back and forth between the home page and listings
• You receive enquiries asking "where can I find your prices?"
• Internal department names or product codes are used as menu labels

The last one is especially common. Terms that make sense inside the company are often not the words visitors use when they search.

Validate the structure first with card sorting and tree testing

NN/g recommends card sorting and tree testing for checking IA. In card sorting, participants group cards labelled with page names, revealing how visitors mentally organise your content. In tree testing, participants look for things in a text-only menu hierarchy ("Where would you find X?"), measuring whether they can find them.

Method	What it shows	Participants (NN/g)	When
Open card sort	Visitors' groups and labels	15+ qualitative, 30–50 quantitative	Before drafting structure
Tree test	Whether the new structure works	Few qualitative, more for quantitative	Before fixing structure
Tree test of old site	Current baseline	Same tasks as new	At kick-off
Usability test	Real confusion on screens	About 5, repeated	Once a prototype exists

The sequence NN/g suggests is usability test of the current site, then card sort, then tree test. Because tree tests need no visual design and the structure can be edited in a spreadsheet, you can run them as often as needed before investing in visuals.

Tree tests measure the share of people who reach the correct location (success rate) and the share who go straight there without backtracking (directness). If you run the same tasks on the old site, you can show how many points the new structure improved.

A caution when changing IA

Changing the structure often changes URLs too. Missing redirects from old to new URLs cause search traffic to drop. We cover the steps in Redesign without losing rankings: 301 redirects.

Google's SEO Starter Guide says organising a site logically helps search engines and users understand how pages relate, but also that search engines can understand your pages as they are now, so there is no need to reorganise in a hurry. Restructure so that visitors can find things, not for search engines.

4. Content: rewrite pages to answer real questions

If the body copy is carried over from the old site, a new look changes very little in either search performance or how convinced readers are. Redesigns move numbers mostly when the content itself is rewritten.

Whether you are found depends on content

Google states plainly that Search shows the most relevant content even if its page experience is sub-par. Good looks and speed only make a difference when many similar pages compete.

Google's SEO Starter Guide also notes that changes can take a few hours to several months to show in search results, and suggests waiting a few weeks before judging their effect. Because content work takes time to pay off, it is worth starting early.

Decide the order of rewrites

You do not need to rewrite every page. Start with the pages with the biggest numbers. The audit method is covered in How to run a content audit; here is just the order of priority.

Priority	Pages	Number to check	What to rewrite
1	Pages often read just before an enquiry	Contribution to enquiries	Pricing, cases, process
2	High impressions, few clicks	CTR in Search Console	Title, description, intro
3	Ranking 11–30	Average position	Missing answers
4	Outdated information	Last updated	Figures, rules, images
5	Overlapping topics	Impressions for same queries	Merge and redirect

The logic of this order is that pages closer to an enquiry tend to move numbers with less rewriting.

Content business sites often lack

People considering an enquiry usually want to know "how much will it cost", "how does it work" and "does it apply to me". Pre-redesign sites often leave these out.

Visitor's question	What happens if missing	What to add
How much?	They leave before asking	Price ranges and drivers
How does it work?	They hesitate to get in touch	Steps and timeline
Does it fit me?	They go back to comparison sites	Cases by sector and size
Who does the work?	Low trust	People and region
Which language?	European clients get lost	State supported languages

The last row matters especially for companies with customers in both Japan and Europe. If the Japanese and English versions say different things, visitors in one language will not be convinced.

5. Speed: pass Core Web Vitals and check the link to results

Speed is the layer most likely to get worse in a visual redesign. Large photos, video backgrounds, animation and third-party tags pile up, and it is not unusual for the new site to be slower than the old one.

The pass marks are set by three metrics

web.dev defines "good" Core Web Vitals thresholds at the 75th percentile of real users as follows. We recommend writing these numbers directly into your redesign requirements and acceptance criteria.

Metric	What it measures	Good threshold	Good on mobile in 2024
LCP	Time until main content shows	2.5 s or less	59%
INP	Responsiveness to input	200 ms or less	74%
CLS	Layout shift while loading	0.1 or less	79%
All three	–	–	43% (desktop 54%)

The right-hand column comes from HTTP Archive's Web Almanac 2024. Fewer than half of sites pass all three on mobile. Speed is still a layer with room to improve on most sites.

Published cases where speed changes moved results

web.dev publishes case studies on the link between speed and business results. The most valuable are A/B tests that changed only speed, leaving visuals and features identical, because they isolate the effect of speed from the effect of looks.

Case	What changed	How measured	Result
Rakuten 24	Removed render-blocking, lazy-loaded images, etc.	1-month A/B test (50:50)	CVR +33.13%, revenue per visitor +53.37%
Vodafone (Italy)	Server-side rendering, image optimisation	A/B test (50:50)	31% better LCP, sales +8%
The Economic Times	INP from about 1,000 ms to 257 ms	Before/after comparison	Bounce rate −50%, pageviews +43%
Deloitte et al. (37 brands)	0.1 s faster mobile speed	Data from Europe and the US	Retail CVR +8.4%, travel +10.1%

Don't apply the case-study numbers directly

These are cases from large, high-traffic companies. There is no guarantee the same percentages will appear on a small business site. The Economic Times result is a before/after comparison, so other changes are mixed in. Even in the Rakuten 24 case, the figure that a good LCP was linked to up to 61.13% higher conversion on the home page is explicitly described as correlation, not causation.

Google also says that while Core Web Vitals are used by its ranking systems, good scores do not guarantee top rankings. It is safest to see speed as a layer that directly moves business numbers, not a magic lever for search positions. We explain how to monitor it in Automating Core Web Vitals monitoring.

6. Flow and forms: remove friction on the way to contact

Flow and forms are the layer closest to the number of enquiries. Whatever the new design, if the contact button sits only at the bottom of the page and the form has 15 fields, people leave just before getting in touch.

What Baymard's research shows about input effort

Baymard Institute in the US continuously studies the checkouts of major e-commerce sites. Its 2024 research found that checkouts have 11.3 form fields on average, while most sites need only 8.

The reasons for cart abandonment (updated September 2025, excluding people who were "just browsing") are also dominated by flow and form problems.

Reason for abandonment	Share	Equivalent on a business site
Extra costs too high	40%	Total price unclear until the end
Didn't trust site with card details	19%	Operator or location not visible
Account creation required	18%	Sign-up needed to download a brochure
Checkout too long or complicated	17%	Too many form fields

Baymard puts the documented average cart abandonment rate at 70.22% (average of 50 studies), and estimates that large e-commerce sites could gain an average 35.26% increase in conversion by fixing checkout usability. B2B enquiry forms share the same pattern: input effort and uncertainty are the main reasons people drop off.

What you can remove from an enquiry form

Business enquiry forms often grow fields for sales convenience. The table below sorts common fields by whether they are really needed for a first contact.

Field	Needed for first contact?	Alternative
Name, email	Yes	–
Message (free text)	Yes	Add an example
Company name	Usually	Make optional
Phone number	Depends	Make optional
Address, postcode	Rarely	Ask when quoting
Budget, headcount, sector	Can confuse	Optional, multiple choice
Confirmation screen	Often not	Send a copy by email

Before removing fields, agree with sales on what they need for the first reply. Fewer fields are pointless if reply quality drops.

Check the flow on your top pages first

Check the flow on the pages with the most enquiries and the most traffic first. The table lists what to look at.

Check	Where	Common problem
Is the contact entry visible?	First mobile screen, mid-content	Only at the bottom
Is the next step stated?	End of each service page	No links to related pages
Can the phone number be tapped?	Mobile	Number inside an image
What happens after sending?	Confirmation page	No reply timeframe given
Are input errors clear?	Form	All errors shown after submit

How to communicate input errors and how large targets should be are also covered by specific criteria in W3C's accessibility standard WCAG 2.2. For European sites there are legal implications, so see also Redesigning for the European Accessibility Act.

7. Measurement: you cannot improve what you cannot measure

Of the five layers, measurement is the one to fix first. If tracking is broken, you cannot tell which layer worked or whether the redesign succeeded at all.

Numbers to keep before the redesign

Once you launch, the old site's numbers cannot be collected again. Export and save at least the following from 8 to 12 weeks before launch. GA4 retention settings can be short for some reports, so export while the data is still visible.

Number to keep	Source	Granularity
Completed enquiries and rate	GA4 (key events)	Weekly
Traffic and contribution by page	GA4	Top 50 pages
Search impressions, clicks, position	Search Console	By query and page
Core Web Vitals	Search Console, PageSpeed Insights	By URL group
Phone and email enquiries	Internal records	Weekly
Enquiry quality	Sales records	Number of qualified leads

Enquiry quality does not appear in GA4. If enquiries increase but none of them turn into sales conversations, that is not a result.

Define key events and test them

In GA4, the actions you count as results are marked as key events (formerly "conversions"). According to Google's help centre, standard properties can mark up to 30 key events, counting starts from the moment you mark them, and historical data is not changed.

So if you forget to set key events on launch day, that data cannot be recovered later. Before launch, submit test enquiries on staging and check that they are recorded under the same names as on the old site. In Europe, cookie consent changes the share of visitors who are measured, so include that setup in the migration. We cover it in Can you use GA4 under GDPR?.

Log the dates of changes

One thing often overlooked is a record of what changed and when: launch date, form changes, campaign starts, price changes. Without it, you cannot pinpoint the cause when numbers move. Use GA4 annotations or a shared spreadsheet to log the date and content of each change.

8. Testing whether a visual change worked: A/B tests

This section covers how to check whether a visual change really worked. The most reliable method is an A/B test that shows the old and new designs over the same period. Because the period is the same, seasonal and advertising effects hit both versions equally, leaving only the difference.

Most ideas do not improve the numbers

In a 2009 paper, Microsoft's experimentation team reported that only about one third of the ideas it tested improved the metric they were designed to improve. The rest made no difference or made things worse.

On the other hand, in a 2017 Harvard Business Review article, Ron Kohavi and Stefan Thomke described a small change to how Bing displayed ad headlines that raised revenue by 12%. Small changes can have big effects, and big changes can have none. That is why you should never judge without measuring.

What experiments showed	Source	Lesson for redesigns
About a third of ideas improved metrics	Microsoft (2009)	"It must be better" is unreliable
Small headline change, revenue +12%	HBR (2017)	Big changes aren't always needed
Speed-only change, CVR +33%	web.dev Rakuten 24	Non-visual layers can matter more

Calculate the traffic you need first

The weakness of A/B testing is that it needs a lot of traffic. The lower the enquiry rate and the smaller the difference you want to detect, the more visitors you need. Using the common settings of a 5% two-sided significance level and 80% power, the visitors needed per variant are as follows.

Baseline enquiry rate	Detect 10% lift	Detect 20% lift	Detect 30% lift
1%	~163,000	~43,000	~20,000
2%	~81,000	~21,000	~9,800
5%	~31,000	~8,200	~3,800

Because you compare two variants, the total traffic needed is twice this. Detecting a 20% lift on a page with a 2% enquiry rate takes about 42,000 visitors in total. Figure 3 shows how many months the test would take depending on the page's monthly traffic.

On business sites with only a few thousand visitors a month, an A/B test using completed enquiries as the metric is often unrealistic. In that case, use a more frequent earlier action, such as button clicks or form starts, as the metric, or switch to the before/after comparison in the next section.

Google's four rules for testing without hurting search

To avoid damaging search performance with A/B tests, Google Search Central gives the following four points.

Rule	What to do	What to avoid
No cloaking	Show Googlebot what users see	Pinning Googlebot to the original
Use rel=canonical	Point test URLs to the original	Setting test pages to noindex
Redirect with 302	Use temporary redirects (JS is fine)	Using 301 (permanent) redirects
Limit duration	Remove the test once you have a result	Running it longer than needed

Google also explains that Googlebot generally does not support cookies, so it sees the same version as users whose browsers reject cookies.

Tools and how to run a test

Google Optimize, Google's free A/B testing tool, was discontinued on 30 September 2023. Today the common options are third-party tools that integrate with GA4, or server-side splitting. Tool costs vary widely with features and traffic, so check each vendor's pricing page.

Run a test like this:

1. Write the hypothesis in one sentence ("Placing case studies above the pricing will increase form starts")
2. Choose one metric and calculate the visitors and time needed in advance
3. Don't change the duration midway (don't stop on a good day)
4. When finished, switch to the winner and remove the test setup
5. Record the result and hypothesis, including losing tests

Point 3 is especially important. If you keep checking results and stop when the numbers look good, you are much more likely to declare a "win" when there is no real difference.

9. When you can't A/B test: measuring before and after

Not enough traffic, every page replaced at once, no way to split visitors. In these cases you compare before launch with after launch. But simply putting the two side by side cannot separate the redesign from seasonal, advertising or economic effects.

Steps for a fair comparison

The steps below reduce errors in a before/after comparison.

1. Set a baseline period: keep weekly numbers for 8 to 12 weeks before launch
2. Freeze the change date: don't change ads, prices or forms in the two weeks around launch
3. Prepare comparisons: the same period last year, and pages you are not changing
4. Set an evaluation period: 4 to 8 weeks after launch; assume search numbers take several weeks or more
5. Segment: compare by traffic source (search, ads, direct) and device (mobile, desktop)
6. Use difference-in-differences: subtract the change on untouched pages from the change on redesigned pages

Core Web Vitals field data in Search Console and CrUX is aggregated over 28 days. After a speed fix, the numbers can take up to about four weeks to reflect it.

Remove seasonal effects with difference-in-differences

Step 6, difference-in-differences, looks at the gap between the growth of changed pages and the growth of unchanged pages. Figure 4 illustrates the idea with fictional numbers.

In this example, a simple before/after view tempts you to report "a 0.6-point improvement". But untouched pages also grew 0.4 points, so the redesign's own effect is more reasonably about 0.2 points.

Common misreadings

The table summarises errors that often occur in before/after comparisons.

Misreading	What is really happening	Prevention
Comparing with a peak season	Seasonal movement only	Add same period last year
More ads at launch	Ad effect mixed in	Freeze ads around launch
Tracking setup changed	Only the counting changed	Keep the same event names
Cookie consent setup changed	Measured share changed	Record consent rates too
Judging on the first few days	Novelty and search fluctuation	Wait 4 to 8 weeks
Judging on one big deal	Small numbers, big chance	Separate counts and qualified leads

The fourth row is especially common on European sites. Rebuilding the cookie banner changes the consent rate, which changes the number of people GA4 can count.

Fill in the "why" with small user tests

Numbers tell you what happened, not why. NN/g's Jakob Nielsen found that testing with about five users reveals around 85% of usability problems, and recommended running many small tests rather than one big one.

The less traffic a site has, and the harder A/B testing is, the more valuable small user tests become. If people try the same tasks on the old and new sites, you can see where they stopped getting lost even with small numbers.

10. Cost and time ranges

The effort to improve the five layers varies widely with the size and state of the site. The table below assumes a business site for a company of a few dozen employees (around 30 to 100 pages). Hourly rates are the Dutch agency average (EUR 106, Simplicate benchmark Q4 2025) and a typical freelance web developer rate (EUR 73, Knab). Yen figures use the European Central Bank reference rate on 30 October 2026 (EUR 1 = JPY 175.68).

Task	Effort	Cost range (EUR)	Approx. in JPY
Audit of current tracking and data	8–24 h	EUR 580–2,540	~100k–450k
Card sorting and tree testing	16–40 h	EUR 1,170–4,240	~210k–750k
Rewriting top 10 pages	20–60 h	EUR 1,460–6,360	~260k–1.12m
Speed improvements	16–80 h	EUR 1,170–8,480	~210k–1.49m
Flow and form changes	8–32 h	EUR 580–3,390	~100k–600k
Measurement design and setup	8–24 h	EUR 580–2,540	~100k–450k
One A/B test (design, analysis)	8–24 h	EUR 580–2,540	~100k–450k

Effort figures are estimates based on DevNJ's experience and exclude participant incentives, testing tool fees, photography and translation. You do not need to do everything at once; starting with the one or two layers that drive your target number brings costs down considerably.

For full redesign price levels and how to read a quote, see Website redesign costs and how to read a quote, which compares survey data from Japan and the Netherlands.

Timeline guide

Also build the time needed to verify the effect into your plan.

Stage	Typical length	Why
Recording the old site's baseline	8–12 weeks	To see weekly fluctuation
IA research and prototyping	2–6 weeks	Repeated tests and fixes
Build and pre-launch testing	4–12 weeks	Depends on size
Post-launch evaluation	4–8 weeks	Waiting for search and speed data

The production process itself is broken into seven phases in The website redesign process and schedule.

11. Differences between Japan and Europe

For sites with customers in both Japan and Europe, the same layer calls for slightly different care.

Item	For Japan	For Europe
Measurement	Disclose external data transmission	Prior cookie consent lowers measured share
Law and accessibility	Reasonable accommodation duty (2024)	European Accessibility Act (from June 2025)
Forms	Many address and phonetic-name fields	Fewer fields, language choice
Showing prices	Often "quote on request"	Hourly rates or ranges more common
Language	Often Japanese only	Often English plus local language

Online purchasing is already mainstream in Europe. According to Eurostat, 77% of internet users in the EU bought goods or services online in 2024. Design your flows on the assumption that European visitors have also looked at plenty of competing sites.

12. Worked examples (fictional)

The following uses three fictional companies for illustration. They are not real businesses.

Case A: a manufacturer whose enquiries stayed flat after a new design

A parts manufacturer with about 3,000 visitors a month. After a complete visual refresh, enquiries stayed at around 10 a month. Investigation showed the contact button sat only at the bottom of each page and the form had 14 fields. With too little traffic for an A/B test, the company cut the form to 6 fields, added a contact entry mid-page on every product page, and compared before and after using difference-in-differences.

Case B: a service firm with plenty of search traffic that ends at reading

An accounting firm attracting 20,000 visitors a month to its blog. Articles had no links to service pages and no pricing. The firm added price ranges and the consultation process to the end of articles and A/B tested enquiry starts from articles. With enough traffic, it used "form opened", a more frequent action, as the metric and planned to reach a conclusion in about two months.

Case C: an online shop selling in both Japan and Europe

A food e-commerce site selling in Japanese and English. Photos in the new design were heavy and mobile LCP exceeded 4 seconds. Without changing the visuals, the company fixed image formats and loading order to improve speed alone first, then planned to test the visual changes separately.

Case	Blocked layer	Measurement chosen	First move
A Manufacturer	Flow and forms	Before/after (diff-in-diff)	Form cut to 6 fields
B Service firm	Content, flow	A/B test (earlier metric)	Pricing and process after articles
C Online shop	Speed	Speed-only change compared	Image optimisation

In all three cases, the first move was not visual. Visual changes came after the other layers were in order, in a form whose effect could be measured.

13. What not to do

These are common judgement errors in design-led redesigns. Each takes effort to undo later.

Don't	Why
Judge the new design by internal taste	Aesthetic-usability effect inflates ratings
Launch without saving the old numbers	You lose your baseline
Change looks, structure, content and speed together and claim "it worked"	You can't tell which change worked
Stop an A/B test early and declare a winner	Chance differences look like effects
Redirect test variants with 301	Search signals move away from the original
Judge success in the first week	Search and speed data lag behind
Use big-company uplift rates in forecasts	Scale and conditions differ

The third one is partly unavoidable in a full redesign. Even so, by logging which layers changed and following up with small layer-by-layer improvements after launch, you can gradually separate what worked. Typical failures and recovery steps are covered in 10 common website redesign mistakes.

14. Checklist

The points in this article, as a checklist from planning to post-launch evaluation.

Planning

• Narrowed the target to one or two numbers
• Chose the layers (IA, content, speed, flow, measurement) that drive them
• Decided whether targeted fixes are enough or a rebuild is needed
• Exported and saved 8 to 12 weeks of the old site's numbers

Design and build

• Ran tree tests on both the old and new structures
• Rewrote content starting with the top enquiry pages
• Wrote LCP 2.5 s, INP 200 ms and CLS 0.1 into the requirements
• Reviewed form fields by whether they are needed for first contact
• Tested that key events record under the same names as the old site

Verification

• Calculated the visitors and time needed for any A/B test in advance
• Set up A/B tests with canonical and 302, and remove them when done
• Logged launch and change dates
• Compared with the same period last year and with untouched pages
• Avoided major changes for 4 to 8 weeks after launch while evaluating
• Used small user tests to understand the "why"

Summary

• Looks drive first impressions and trust, but enquiries and sales are driven by information architecture, content, speed, flow and forms, and measurement
• Decide which number should move before the redesign, and fix the layer that drives it first. Targeted fixes are often enough
• An A/B test that changed only speed raised conversion by 33%, showing that non-visual layers can be measured
• Fix measurement first. Keep 8 to 12 weeks of the old site's numbers before launch
• Measure visual changes with A/B tests, but note that detecting a 20% lift at a 2% enquiry rate needs about 42,000 visitors in total
• Without enough traffic, combine a difference-in-differences before/after comparison with user tests of around five people

Sources

Visuals and usability

• Attention web designers: You have 50 milliseconds to make a good first impression! — Lindgaard et al. (Behaviour & Information Technology, 2006): visual impressions form in 50 ms
• How Do People Evaluate a Web Site's Credibility? — Consumer Reports WebWatch (Fogg et al., 2002): 46.1% of credibility comments mention design look
• The Aesthetic-Usability Effect — Nielsen Norman Group: the 1995 Hitachi study and cautions for user testing
• Radical Redesign or Incremental Change? — Nielsen Norman Group: reasons for redesigns and when a rebuild is justified

Information architecture and user testing

• Tree Testing — Nielsen Norman Group: success rate, directness and use in redesigns
• Card Sorting — Nielsen Norman Group: 15+ qualitative, 30–50 quantitative participants
• Why You Only Need to Test with 5 Users — Nielsen Norman Group: five users find about 85% of problems
• How to set performance metrics for your service — GOV.UK Service Manual: 3–4 KPIs and baselines

Search and measurement (Google)

• Website testing and Google Search — Google Search Central: four rules for A/B testing
• Understanding page experience — Google Search Central: relevance first; scores don't guarantee top rankings
• SEO Starter Guide — Google Search Central: site organisation and time for changes to show
• Mark events as key events — Google Analytics Help: key event limits; no effect on historical data
• Google Optimize sunset — Google Analytics Help: discontinued on 30 September 2023

Speed and results

• Web Vitals — web.dev: LCP, INP and CLS thresholds at the 75th percentile
• Rakuten 24 case study — web.dev: one-month A/B test, CVR +33.13%
• Vodafone case study — web.dev: 31% better LCP, sales +8%
• The Economic Times case study — web.dev: better INP, bounce rate −50%
• Milliseconds Make Millions — web.dev: 0.1 s faster, retail CVR +8.4%
• Web Almanac 2024 Performance — HTTP Archive: Core Web Vitals pass rates

Flow and forms

• Checkout Flow Average Form Fields — Baymard Institute: 11.3 fields on average, 8 are enough
• Cart Abandonment Rate — Baymard Institute: 70.22% abandonment, reasons, 35.26% potential gain
• WCAG 2.2 — W3C: error identification and target size

Experiments

• Online Experimentation at Microsoft — Kohavi, Crook and Longbotham (2009): about a third of ideas improved metrics
• The Surprising Power of Online Experiments — Harvard Business Review (2017): Bing headline change raised revenue 12%

Cost and Europe

• Benchmark report Agencies Q4 2025 — Simplicate: average Dutch agency hourly rate (EUR 106)
• Web developer freelance hourly rate — Knab: freelance hourly rate (EUR 73)
• Euro reference rate (JPY) — European Central Bank: EUR 1 = JPY 175.68 (30 October 2026)
• E-commerce statistics for individuals — Eurostat: 77% of EU internet users bought online in 2024
• European Accessibility Act — European Commission: products and services covered

All sources were checked on 30 October 2026. Case-study figures were obtained under each company's own conditions and the same percentages may not apply to another site. Check against your own numbers before making decisions.