Brief
Brimlough, an invented app, helps independent cafés cost their recipes and track stock. Visitors start a free trial from a pricing page, and few trials become paid plans. The sign-up asks for a card and a long set of business details, trials end with no follow-up, and last year's winning pricing page test was never rebuilt in the site's code: the testing tool still serves it to everyone.
The growth team is a growth lead who owns analytics and decisions and a designer-developer with half their week for growth. Traffic supports only a few tests a quarter. The founder wants one strategy the two can run, and reads one page a month.
Constraints:
- Developer time is the scarcest resource; every test and release competes with product work.
- Prices and plan contents stay fixed; presentation, sign-up and the trial journey are open.
- Sales notes and support email are available but uncoded.
- Nothing may ship that the business could not defend as a claim.
The product and its evidence
Brimlough's trial journey, a typical week
| Step | People | Note |
|---|---|---|
| Pricing page visitors | 2,800 | New visitors and returning prospects |
| Trial sign-up started | 420 | |
| Business details completed | 240 | Eleven fields, including turnover, number of staff and VAT number |
| Card entered, trial started | 140 | Card asked before the product opens |
| First recipe costed within seven days | 49 | |
| Paid, from finished trial cohorts | 17 | About 12% of trials started |
From the last twelve finished weekly cohorts: of trials that costed a first recipe in their first week, about 29% paid; of those that did not, about 3%. No email is sent during or at the end of the trial.
Brimlough's pricing page, sign-up and records
| Element | What the team found |
|---|---|
| Plan cards | "Single café" at £29 a month and "Café group" at £59; both figures are annual billing only, which the page does not say |
| Recommendation | "Most popular" badge on Café group; billing records show most paying customers have one site |
| Consent | One box, ticked by default: "I agree to the terms and to receive offers" |
| Last year's test | Plan cards led by café type beat the old cards on trial starts per pricing visitor; the report lists no guardrail and does not say whether the planned sample was reached |
| Sales and support | About 40 pre-sales emails a month; repeated questions about cancelling and whether the card is charged during the trial |
| Candidate test | Primary metric | Visitors per arm the team's calculation needs | Visitors a week reaching the test |
|---|---|---|---|
| Pricing page plan cards | Trial starts per pricing visitor | 6,000 | 2,800 |
| Card at sign-up or not | Paying customers per sign-up start | 9,000 | 420 |
| Trial follow-up emails, against a holdout | Paid per trial started | 700 | 140 |
Task
Write the strategy as one document, in this order, using the templates and checklists named.
- Read the trial rate properly. Confirm the trial-to-paid figure is read on finished start cohorts, as the diagnosis for trial users not converting to paid requires, and name the milestone that separates paying from non-paying trials. State the Trial-to-Paid Conversion RateShare of free trials that became paying subscriptions. you will report.
- Plan the research. Write a sprint on the sign-up and trial journey with the research brief template: one question, one owner, an end date, a tracking check with the tracking verification checklist, the located step, and where recordings, a poll, a small usability round with café owners and a coded read of sales and support email will point, as the picture framer research sprint and the research sprint checkpoint practised. Every later change traces to a finding from at least two methods.
- List what ships without a test. Sort the evidence into bugs, false or incomplete claims, consent problems and cheap reversible changes, as the no-split-test checkpoint rehearsed. Give each a check date and a log entry.
- Choose the tests the traffic can finish. Work out the weeks each candidate needs, choose a few, sequence them so they do not share visitors at once, and write each on the experiment hypothesis template with its primary metric, every Guardrail MetricA metric an experiment must not worsen beyond an agreed bound, watched beside the primary metric that decides the test. and a reading horizon. Say what happens to any candidate that cannot finish.
- Redesign the sign-up and the trial. Using the guide to sign-up and trial-start flows, cut the fields to what the first session needs, decide the card deliberately, separate consent from the terms, and make the first screen start costing a recipe. Judge changes by what new accounts do next.
- Rebuild the pricing page honestly. Using the guide to designing pricing pages, describe each plan by the café it suits, label both billing periods, choose the recommendation by fit, and answer cancellation and the trial's card terms on the page. Run every device through the three questions from the campervan hire audit.
- Stage the largest change. Name it, and release it to a random, sticky share of visitors against a concurrent comparison group, with gates, a written prediction, a reading date and rollback rules agreed in advance, as in the cookware relaunch exercise.
- Set the shipping and learning rules. Decide what happens to last year's winner by the steps in the tutorial on shipping a winning variant, from confirming it against the plan to removing the test code. Keep the learning log and its "implemented" field as the experimentation programme playbook does, with results on the experiment results report template.
- Write the calendar and the founder's page. A half-year calendar that fits two people, and a one-page summary in the what, so what, now what form: what the evidence shows, what it costs in paid customers, and what happens next.
The arithmetic to start from
| Candidate | Arithmetic | Weeks of entries |
|---|---|---|
| Pricing page plan cards | 6,000 × 2 ÷ 2,800 | about 4.3 |
| Card at sign-up or not | 9,000 × 2 ÷ 420 | about 43 |
| Trial follow-up emails | 700 × 2 ÷ 140 | 10, then the trial length before reading |
A half-year is about 26 weeks: the card test cannot finish; the other two can.
Evaluation criteria
| Criterion | What a strong answer shows | Common gap |
|---|---|---|
| Evidence before change | A research plan aimed at the located steps of the sign-up and trial journey, with every proposed change traced to a finding from at least two methods | Changes chosen from opinion or a competitor's site |
| The right tool for each change | Bugs and required fixes shipped, a few tests the traffic can support, and the largest change released in stages against a comparison group | Every idea queued as a split test the traffic cannot finish |
| Sign-up and trial design | Fields cut to what the next step needs, a deliberate card decision, consent kept separate and the first screen leading to a useful task, judged by what new accounts do next | Sign-ups counted as the result |
| Honest pricing page | Complete costs, plans described by who they suit, an intended choice that does not steer, and trial terms and cancellation answered on the page | Countdowns or pre-selected annual billing added for a lift |
| Guardrails that measure harm | Paid conversion, refunds, cancellations and support contacts read beside the primary metric for every test and release | Trial starts as the only metric |
| Shipping and learning | Wins rebuilt properly, checked in tracking, logged with an expected range, watched by new and returning visitors, a holdback where the change is large, and the learning log closed | A winning variant left running in the testing tool |
A strong submission notices what the evidence already says. Trials that cost a recipe early pay at many times the rate of the rest, so the sign-up and first screen are the constraint, not the end-of-trial offer. The pre-ticked consent and unlabelled annual prices ship as fixes. Last year's winner has no guardrail and no confirmed sample, so it is a lead to retest on the honest page, not a win to rebuild as it stands.
The card decision shows judgement. An A/B TestControlled comparison of two variants. of it cannot finish this half-year, so a strong answer either keeps the card with its terms stated on the pricing page and beside the field, recorded as judgement, or folds it into the staged sign-up release, read on paid conversion, refunds and cancellations from finished cohorts, and says why.
Submission checklist
Check your strategy against the evaluation table, and revisit the intermediate CRO level if the test arithmetic or guardrails felt shaky.
