A 90-day growth sprint is one quarter spent testing a small number of falsifiable hypotheses against a single named growth constraint, on a fixed weekly decision rhythm, ending with that rhythm handed to your team.
That definition is doing a lot of work, so it is worth separating from what it is not. A sprint is not a quarterly marketing plan with a Gantt chart. It is not a channel retainer with a three-month minimum. It is not a strategy deck delivered in week two and quietly ignored by week six. The difference in every case is the same: a sprint names one constraint before any money is spent, keeps the number of live experiments small enough to read, and ends with an internal owner rather than a renewal conversation.
Below is the structure, week by week. It is the same shape We Scale Startups runs as a paid engagement, written out so you can run it yourself. If you would rather not, the paid version is here — but the method is not the part worth charging for.
Why 90 days is the unit
Ninety days is roughly the shortest window in which a B2B acquisition channel produces a signal you can trust.
Thirty days is the tempting alternative and it is usually too short. Paid campaigns are still in learning. Content has not been indexed, let alone ranked. Outbound sequences have not completed a full cycle, so reply rates are measuring the first two touches of a five-touch play. Teams that run 30-day tests spend most of their time killing things that were about to work and scaling things that got lucky.
A year fails in the opposite direction. It is long enough for a wrong bet to become an organisational fact — a hire made, a tool bought, a narrative defended in three board meetings — before anyone is willing to say it did not work.
A quarter also happens to be the unit your board already thinks in. That is not a trivial point. The output of the sprint lands in the meeting where decisions actually get made, in the format that meeting already uses.
Week 1: name the constraint, then agree the measure
Nothing else in the sprint works if this week is skipped, and it is the week most teams skip because it produces no visible activity.
The job is to replace “growth is slow” with one specific, falsifiable sentence. Something closer to: pipeline is inconsistent because our ICP is too broad and every channel carries a different promise. That sentence has to come from evidence, not from the loudest opinion in the room.
Three sources are usually enough to get there:
- Funnel data. Export your GA4 events and your CRM stage-to-stage conversion rates. You are looking for the step where the drop is disproportionate, not the step with the biggest absolute loss.
- Sales calls. Listen to five recent ones. Not the wins — the ones that stalled. You are listening for the objection that keeps recurring, which is almost never the objection your website answers.
- Customer evidence. What did your last ten closed-won customers have in common that your last ten stalled deals did not?
Then agree the single measure that would prove the constraint moved. One measure. If the team cannot agree on one, the constraint is not named tightly enough yet, and you should stay in week 1 rather than proceed with a compromise.
Write down the baseline for that measure before you touch anything. In three months you will want it and nobody will remember what it was.
Weeks 2 to 4: design and launch
Pick three to five experiments. Not eight.
The number matters more than it looks. Fewer than three and a single dud wastes the quarter. More than five and two failures compound: nobody has enough attention to build each test properly, and when several changes hit the same funnel simultaneously the results become genuinely unattributable. You will finish the quarter with movement and no idea what caused it.
Score the candidates on three things — expected impact on the constraint, your confidence based on data you already have, and ease of execution with the team you actually have rather than the team you are hiring. Rank them, take the top three to five, and put the rest in a backlog you will genuinely revisit at day 90.
Every experiment needs four things written down before it launches:
- The hypothesis, stated as a belief that could turn out to be wrong.
- The threshold that would make you scale it, decided now, before you are emotionally invested in the result.
- The owner — one name, not a team.
- The kill condition. What result would make you stop? If there isn’t one, this is not an experiment.
Then build and launch. Landing page variants carrying genuinely different value propositions, not different button colours. Creative sets that test a message rather than a format. Email sequences with behavioural triggers rather than a calendar.
The weekly decision meeting starts this week and keeps its slot for the rest of the quarter. Protect it more carefully than you think you need to.
Month 2: scale the signal, repair the leaks
By now the tests are producing readable numbers, and this is where sprints most often quietly die.
They die because the weekly meeting drifts into a status update. Someone reports what happened; nobody decides anything; the meeting shortens; within three weeks it is a Slack thread. The discipline that prevents this is simple to state and hard to hold: every experiment gets an explicit scale, stop, or fix call, every week, out loud.
The rule of thumb worth starting from is to scale what is clearly returning above its threshold, kill what is clearly below it, and fix — rather than kill — the cases where the channel is working but a step after the click is leaking. That last category is the one teams get wrong most often. A channel that drives qualified traffic into a broken activation flow looks identical in a CPL report to a channel that drives junk. The distinction only shows up if you are reading the funnel past the first conversion.
Two things to track alongside cost per lead: the conversion rate from lead to qualified opportunity, and pipeline velocity. Cost per lead on its own will reliably talk you into scaling your worst channel.
Month 3: transfer ownership
The sprint is only worth doing if something survives it. Month 3 exists to make that true, and it needs to be scheduled rather than hoped for.
Three things move to the team:
- The cadence. Documented — who attends, what gets reviewed, what decisions are in scope, what happens when the numbers disagree.
- The dashboard. One dashboard that everyone trusts, owned by a named internal person. Not four dashboards that each tell a partial truth.
- The hypothesis log. What was tested, what the threshold was, what happened, and what you now believe. This is the asset that stops the next quarter re-running experiments you already have answers to.
Then the person who has been driving the sprint steps out of the loop deliberately, while there is still time to notice if it collapses. A handover in the final week is not a handover; it is a hope.
Day 90: the review
Three questions, answered honestly:
What moved, and can you attribute it? What did not move, and do you now know why? What does the next quarter test as a result?
The third question is the point. A sprint that ends with a list of achievements has not really finished. A sprint that ends with a sharper constraint than the one you started with — even if the numbers were mediocre — has done its job, because you are now betting on better information than you had in week 1.
The failure modes worth naming
Four patterns account for most sprints that produce nothing:
Too many experiments. Covered above, and it remains the most common one. Ambition looks like eight tests. Learning looks like four.
No authority to stop things. The sprint owner needs to be able to kill a channel the founder personally championed. If they cannot, the weekly decision is decorative.
Thresholds set after the result. Deciding what counts as success once you can see the number is not analysis. Write the threshold down first, in the same document, before launch.
Running one before product-market fit. A sprint optimises an acquisition system against a constraint. If the constraint is the product or the customer definition, no volume of channel testing will resolve it. Pre-PMF teams get more from customer research and a sharper ICP than from a quarter of experiments.
Doing this without us
The structure above is complete. Nothing has been held back for the paid version, because the structure is not the scarce thing — the willingness to name one constraint, keep the experiment list short, and hold the weekly decision when the answer is uncomfortable is the scarce thing, and no engagement can buy that on your behalf.
If you want to run it yourself, the 90-Day Growth Sprint Planner is the working template.
If you are not sure the constraint is named correctly — which is the failure that quietly wastes the other eleven weeks — the Growth Diagnosis at £2k–£4k answers that question on its own, in about a week, without committing you to a quarter.
And if you would rather have someone senior run the quarter with you, the 90-Day Growth Sprint is £8k–£12k over twelve weeks. What that buys is the experiment design, the judgement calls in month 2, and the handover artefacts — not execution hours.
Related reading: random acts of growth on why uncoordinated activity feels productive, systems vs activity retainers on what you are actually buying, and the growth signal loop on turning evidence into decisions.