Marketing & growth
Conversion rate optimisation
More of the traffic you already pay for turning into enquiries or orders, worked through a ranked hypothesis backlog with every test sized for significance before it goes live.
The short version
Conversion work goes wrong in two directions. Either it never gets past changing a button colour, or a test runs on 400 sessions, reports a 22% lift, and the lift quietly disappears the following month. Your traffic decides which methods are even available to you, because statistical testing has a minimum sample size and no amount of enthusiasm shortens it. We will tell you on the first call which situation you are in and what can honestly be measured.
What is included
- A measurement check before any testing starts: funnel steps tracked, each event firing once, and form abandonment captured field by field, so the baseline is real.
- A friction audit combining session recordings, scroll and click data, form analytics and a manual walkthrough of the whole flow on a real mid-range phone.
- Qualitative input alongside the numbers — an on-site exit question, or an unmoderated test with five participants, because recordings show what happened and never why.
- A hypothesis backlog scored on expected effect, confidence and effort, each item written in the form: because we observed X, changing Y will move Z.
- A written plan per test naming the primary metric, the guardrail metrics, the minimum detectable effect, the sample size required and the stop date, all agreed before launch.
- A running results log holding wins, losses and inconclusive tests with their raw numbers, because the inconclusive ones are what stop you retesting the same idea next year.
How we work
Our approach
The stages this work runs through, in order. Every one of them ends in something you can see or sign off.
-
Step 01
Check the baseline is real
Conversion problems and measurement problems look identical from a dashboard, so the first pass validates the funnel before anything else. Does each step fire once, is the confirmation page counting a conversion every time somebody refreshes it, is bot traffic inflating the denominator. We then set a baseline period long enough to cover your weekly and monthly cycles. Testing against a broken number produces confident decisions in the wrong direction, which is worse than doing nothing at all.
-
Step 02
Find friction, not opinions
Recordings, click maps and field-level form analytics show where people stall. A manual walkthrough on a mid-range Android over an ordinary connection usually shows why, because most friction is a slow step, a form asking for something people will not give, or an obvious question left unanswered. We add one qualitative source so we are not inventing motives. Everything found enters the backlog as an observation, and observations get ranked by the money at stake.
-
Step 03
Size the test before running
Every test gets its arithmetic done first. We agree the primary metric, the smallest effect worth detecting and the confidence level, and from those the sample size and the run length. Tests then run for at least two full weeks so weekday and weekend behaviour are both included, and nobody stops one early on a promising day, because peeking is how a coin flip becomes a case study. If your traffic cannot reach the sample, we say so and change method.
-
Step 04
Ship winners, keep the log
A winning variant does not stay inside the testing tool. It gets built into your site properly, because client-side test code slows pages down and eventually breaks against a redesign. After implementation we re-measure to confirm the gain survived the move into production. Losing and inconclusive tests get written up with their numbers, since a documented dead end has real value. The backlog is then re-ranked, because what one test teaches usually changes what is worth testing next.
Before you ask
Questions we get asked first
How long does a test take to run?
Your traffic decides that, not us. As a rule a test runs at least two full weeks to cover both weekly cycles, and long enough to reach the sample the arithmetic demands. Detecting a 10% relative improvement on a 2% conversion rate at 95% confidence needs somewhere near 25,000 sessions per variant, which is out of reach for most sites. Below that threshold we work differently: bigger changes, sequential before-and-after measurement, and honest caveats about what it proves.
How is it priced?
Two shapes, both fixed in pounds. A one-off audit that hands you a scored, ready-to-run hypothesis backlog your own developer can work through. Or a monthly retainer covering a defined number of research cycles and tests per month. Testing tool licences and any development outside the agreed scope are separate and itemised where they arise. We do not price on a share of the uplift, because that makes the person measuring the result the person paid by it.
What do you need from us?
Analytics and heatmap access, plus either a route to deploy code or a developer we can hand implementation notes to. Your commercial figures: what an average order is worth, the margin on it, and the value of a lead once it closes, so tests get ranked by money rather than by percentage. A decision-maker who accepts that a fair share of tests will lose. And agreement that a losing variant stays switched off without the hypothesis being relitigated afterwards.
How do we separate a real win from a fluke?
Per test, the primary metric clearing the confidence level we agreed, with guardrail metrics showing nothing broke elsewhere on the site. Across the programme, the annualised effect of the changes actually shipped, calculated from your figures rather than the testing tool's. For lead generation we look past form fills to whether the extra enquiries close, since a change that fills your pipeline with poor leads is a loss. Inconclusive results get reported as inconclusive.
Want this scoped for your situation?
Tell us the goal and the deadline. You get an honest read on whether this is the right service, a fixed price if it is, and a straight no if it isn't.
Or email [email protected] directly. We reply within one working day.