Conversion rate optimization

A conversion rate optimization agency that finds the leak before it tests a fix

We study how visitors use your site, write down what we think is stopping them, and test the change only where your traffic can give a readable answer.

Get your conversion review
Trusted by the world's fastest-growing brands

Portfolio · selected work

The work, not the deck.

Every site below is live and owned by the client. Each window is a capture of the real page, not a drawn mockup.

Categories we work across
HospitalityFinancial servicesManufacturing and exportConsulting and researchRecruitmentD2C
urjalifes.inUrjalife's Pravaah resort website: rooms, activities and dining across a photographic homepage.
UrjalifeHospitality
adventswealth.comAdvents Wealth website: a financial planning homepage covering wealth building, tax and retirement.
Advents WealthFinancial services
syounaa.inSyounaa website: a wellness apparel storefront built around smart-textile product ranges.
SyounaaWellness apparel
infokool.comInfokool website: a precision cooling manufacturer, with an outdoor cabinet product grid and capability sections.
InfokoolManufacturing
howardjohnsonhinton.caHoward Johnson Hinton website: a mountain-lodge hero over a room-booking bar, followed by renovated room types.
Howard Johnson HintonHotel
sycamoreconsultant.comSycamore Consultancy website: a recruitment agency, with separate paths for employers and candidates.
Sycamore ConsultancyRecruitment
stutiexports.comStuti Exports website: a technical fibers and specialty yarns exporter, with core markets and capability sections.
Stuti ExportsTechnical textiles
duhomespg.inDuhomes website: student housing, with walking distances to nearby colleges and a room rate table.
DuhomesStudent housing
vaspro-global.comVasPro Global website: a technical advisory and project consulting firm, with expertise and mission sections.
VasPro GlobalConsulting
pretor.aePretor website: a marketing research and promotion consultancy, opening on research, strategy and growth planning.
PretorMarketing research

What CRO covers

Three questions that come before any test

Conversion rate optimization is the work of getting more of the people who already visit to take the step you want. A conversion rate optimization agency earns its fee by answering three questions in order, because a test without them is a guess with a stopwatch.

Test brief · estimate pageCan be tested01Where do visitors stall1,056 starts336 submitted720 left the form02Why do we think soHypothesisForm: 11 fields → 6, photo optional14 of 30 recordings stop at the upload03Can traffic answer itabout 9 weeks

Where do visitors stall

We read your analytics, watch session recordings and look at form analytics, field by field. That shows which step loses people and how, so the first change is chosen from evidence and not from taste.

Why do we think so

Each change gets a written hypothesis: what we will alter, what we saw that suggests it, and which number should move. A hypothesis that can be wrong is what makes the result worth reading.

Can your traffic answer it

A test needs enough visitors to tell a real difference from chance. We work out the number first. Where the page is too quiet, we fix the clear problems and measure before and after instead of running a test that cannot finish.

Two ways to run it

Testing that starts from evidence and traffic

Two ways to run a testing program. Both change pages and count results; they differ in where the first idea comes from and when a result is called.

Ideas firstA list of tests to runResearch firstEvidence, then a test
Where ideas come fromA library of tests that worked on other sitesAnalytics, recordings, form analytics and a few customer conversations
Before a test startsA variant built and a split switched onA written hypothesis and a sample size worked out from your own rate
When a result is calledWhen the tool shows a high confidence figureAt the sample size set in advance, and not when the chart first looks good
What counts as a winMore clicks on the button that changedMore qualified inquiries, booked consults or revenue
A quiet siteRun the test anyway and waitFix the clear problems and read before and after, with the limits stated
Why we size the test first

A test that cannot reach a readable answer in a sensible window costs weeks of attention and returns a shrug. Working out the number first tells us whether to test, to fix and watch, or to leave the page alone.

The evidence

Six facts that shape how we test

What the testing literature and the tools say about when a result can be trusted, with the source under each.

  1. PeekingEvan Miller

    Stopping a test when it first looks significant produces false winners.

    Evan Miller writes that the problem appears whenever you peek at the data and stop an experiment that seems to be giving a significant result. He gives the example that peeking ten times can turn what looks like 1% significance into about 5%. His advice is to decide the sample size in advance.

  2. Sample sizeEvan Miller

    The sample you need depends on your rate and the smallest change worth detecting.

    The calculator asks for a baseline conversion rate, a minimum detectable effect, statistical power and a significance level. A smaller change on a lower rate needs far more visitors than a large change on a higher one.

  3. ResearchNN/g

    Tests run without research make variations that are essentially guesses.

    Jen Cardello of Nielsen Norman Group writes that if A/B tests are used in lieu of research, the variations are essentially guesses. She recommends brief surveys, usability testing and tree tests first, and notes that five users can uncover about 85% of the massive flaws.

  4. FormsBaymard

    The average checkout asks for 11.3 form fields; Baymard says most need only 8.

    Baymard Institute reports that the average checkout flow in 2024 had 11.3 form fields, down from 11.8 in 2021 and 12.7 in 2019, and says most sites need only 8 in total. It is a benchmark for checkout; a quote form or booking form has its own reasons for each field.

  5. Key eventsGoogle

    In GA4 any event you collect can become a key event.

    Google defines a key event as an event that measures an action that is particularly important to the success of your business, and says any event you collect can become one. So a form submission, a call tap or a booked slot can each be counted, once the event exists.

  6. ToolsGoogle

    Google Optimize ended on September 30, 2023.

    Google says Optimize and Optimize 360 are no longer available as of September 30, 2023, and that experiments still active on that date ended. It says it is investing in third-party A/B testing integrations for Google Analytics. A site still waiting on Optimize is not running tests.

The scope

What each part of the program produces

Six parts. Each ends in something your team can open, question and keep.

Tag AssistantCheckedEventMediumStatusform_startcpcOnceform_submitcpcFired twiceform_submitcpcFixed · oncecall_taporganicOnceKey events: form_submit, call_tap

Measurement audit

We check that the form, the call tap and the booking are counted once each, with the traffic source attached, before anything is judged against them. Broken or doubled events are fixed in your container. You receive: a measurement audit listing each event, how it fires and what we changed.

GA4Key eventsTag Manager
Session recordings30 reviewedRoof estimateNamePhonePhotoSendStarted, did not submit14 of 30stop at the photo upload0:52left at 1:07StartedPhoto field

Behavior research

We watch session recordings of visitors who started and left, read form analytics field by field, and look at where taps and scrolls land on a phone. You receive: a research report that ranks the problems by how many sessions show each one, with clips and screenshots.

RecordingsForm analyticsHeatmaps
Hypothesis H1Awaiting approvalWe will changeCut the form to 6 fields; photo optionalBecause we saw14 of 30 recordings stop at the uploadIt should move7.0% → 8.5%SubmissionsQualified inquiriesRank 1 of 2CommentApprove plan

Hypotheses and priorities

Each idea is written as a change, the evidence behind it and the number it should move, then ranked by the likely size of the effect, the effort and the traffic on that page. You approve the list before work starts. You receive: a test plan with one written hypothesis per line.

HypothesesRanking
Sample-size note80% power · 5%Shorter form · submissionsA/B test7.0% → 8.5% · 10,000 sessions · 9 weeksSticky call bar · callsFix list3.0% → 3.8% · 16,000 phone · 5 months01 mo2 mo3 mo4 mo5 mo

Test design and sample size

For each test we work out how many visitors are needed to read a change of the size that would matter, and how many weeks that takes at your traffic. If the answer is too long, the item moves to the fix list. You receive: a sample-size note per test, with the inputs shown.

Sample sizeSplit test
Variant checkReady for approvalOriginalShorter formSpeed checkedDeviceLayoutSubmitsNo flickeriPhone 15Pixel 8Mac · SafariWindows · Edge

Build and quality check

We build the variant in your site or testing tool, check it on real phones and in the main browsers, and confirm that the page does not slow down or flicker. You approve the variant before it goes live. You receive: the variant, the device check and the speed readout.

VariantsDevicesSpeed
Result: shorter formShip5,190 sessions per version95% rangeSubmissions7.0% → 8.2%+1.2 pts+0.2 to +2.3Qualified3.1% → 3.5%+0.4 ptsinside the noise−10+1+2PointsShip; keep watching the qualified share

Results and decisions

When a test reaches its planned size we read it on leads or revenue, show the range of the likely effect and not only the average, and record the decision: ship, drop or run again. You receive: a result note per test and a decision log you keep.

ResultsDecision log

How it runs

Measure, hypothesize, test, decide and log

You approve the plan before a test starts and the variant before it goes live, and each stage leaves a document. The numbers below come from one sample site and agree throughout.

research-findings.csv

Research findings, estimate page

Sample · Sample Copperleaf Roofing
FindingEvidenceSessions affectedSeen in
Form asks for 11 fieldsForm analytics: 1,056 starts, 336 submitted720 left the formAnalytics
Photo upload is required14 of 30 reviewed recordings stop at the upload fieldMany mobile startsRecordings
Phone error clears the form9 of 30 reviewed recordings re-enter details, then leaveMobile onlyRecordings
Call button sits under the form144 calls from 4,800 sessions, 3.0%All mobile sessionsAnalytics
Thirty days: 4,800 sessions, 1,056 form starts, 336 submissions (7.0% of sessions), 144 calls. That is 480 inquiries, forms and calls together, of which 211 (44%) were qualified: roof work inside the service area. The test below counts form submissions only, so its qualified figures are qualified submissions. The 30 recordings are those of visitors who started the form and did not submit.

Our stack

The tools behind the research and the tests

The tools that count visitors, record sessions and build variants, each with its job in a test. The accounts sit in your name.

GA4
Why GA4

Google Analytics 4 counts visits and events on a site or app, and lets any event be marked as a key event.

How we use it

We set up events for the form, the call tap and the booking, and read them by traffic source, so a result is judged on the action that matters.

AttributionMulti-touchNo-cost tier
GA4Yes
HubSpotCRM
DreamdataB2B
SpreadsheetManual

Questions

Conversion rate optimization, answered

Start here

Get a review of where your visitors leave

Send us the page and tell us what a visitor should do there. We’ll walk the path as a first-time visitor, say what we’d measure first, and tell you whether your traffic can support testing.