A written bet
You know what you were testing before the traffic started.
11Services · Growth
A/B Split Testing.
Two layouts in parallel. A honest slice of traffic sees the challenger. Winners move to the default; losers leave the same day — no three-change “experiments”.
01The usual mess
Most tests change three things and call it science.
The “winner” is declared on Tuesday. Sample sizes were a guess. The testing script made the page slow. Nobody can take the variant down without a fire drill.
We write the bet first. A small share of traffic sees one challenger layout. Winners become the default. Losers leave the same day.
02What we do
Two layouts. A honest slice. A way back.
Hypothesis and the number that would make us stop — written before anyone builds a variant. Only a small percentage of traffic hits the challenger until we know.
Promote winners into the main path properly. Drop losers — no flag graveyard. You get a short readout a marketing lead can actually use.
03How the job runs
How an experiment actually runs.
- 01
Write the bet.
What we think, what we will measure, when we would stop early.
- 02
Size the slice.
A small traffic share that can still support a decision.
- 03
Ship one variant.
Without making the control slow, and without leaking into it.
- 04
Read the number you named.
Ignore the vanity charts that showed up for free. Winner into the main path — or a clean rollback the same day.
04Who it is for
Who this is for.
Marketing and product teams who want experiments an engineer can live with — not a tag manager full of forgotten variants.
Landing and funnel iteration
Pricing and CTA tests
Teams burned by messy test tools
Sites we already instrument
05What changes
What you walk away with.
A careful slice
Most visitors stay on the control until you promote.
A way back
Losing variants leave the codebase, not a graveyard of flags.
06Work
Work that used this service.
Cases from the archive that used this line of work.
07Pricing
What this costs.
A single honest experiment, a monthly programme, or a readout when the test already ran elsewhere.
01Most chosen
One experiment.
From £995
per test
Hypothesis, small traffic share, promote or drop — one change at a time.
Run one testGet this from as little as £25 per month with Flexi-Pay over 24 months.
Based on a 40% deposit, then the balance across 24 months. Indicative, excluding VAT. Flexi-Pay terms. Terms and conditions apply.
02Most chosen
Experiment programme.
From £1,750
per month
A cadence of tests on landing and funnel work, with engineering in the loop.
Talk programme03Most chosen
Experiment readout.
From £450
per engagement
You ran the test — we interpret it and say promote, drop or re-run.
Book a readoutGet this from as little as £11 per month with Flexi-Pay over 24 months.
Based on a 40% deposit, then the balance across 24 months. Indicative, excluding VAT. Flexi-Pay terms. Terms and conditions apply.
Compare
Ticks are in the fee. Crosses are not. A higher plan keeps every tick from the plan to its left, then adds the rows below.
| Included | 01One experimentFrom £995 | 02Experiment programmeFrom £1,750 | 03Experiment readoutFrom £450 |
|---|---|---|---|
| Test design | |||
| Written hypothesis and primary metric | ✓ | ✓ | ✓ |
| Small traffic share on the variant | ✓ | ✓ | × |
| Outcome | |||
| Promote winners / drop losers | ✓ | ✓ | ✓ |
| Short readout for marketing | ✓ | ✓ | ✓ |
| Engineering merge notes when promoting | ✓ | ✓ | ✓ |
| Cadence | |||
| Retainer backlog of careful tests | × | ✓ | × |
| Not in the fee | |||
| Experiment tool licences (at cost) | × | × | × |
Retained Service
It’s cheaper to partner up.
One-off engagements are fair on their own — and they add up. A monthly retainer unlocks named capacity across Build, Growth and Care for less than stacking those services as separate projects.
Single tests and readouts are fixed once the hypothesis or data is agreed. Programmes are rolling monthly, cancel with 30 days’ notice. Experiment tool licences are passed through at cost. Excludes VAT.
Next step
If you cannot roll it back, it was not an experiment. It was a deploy with extra steps.
What do you think is losing? One change. We will say if the traffic can support it.