Growth experiments are changes you ship to move one agreed number, each with a written prediction, a judgement date and a final verdict. Ours end in one of two words: kept, if the change moved the number, or rolled back, if it did not. Without that verdict, a growth experiment is just a change nobody checked.
TL;DR: Sign a baseline before you test anything. Log every change with five fields, including the date you will judge it. Set that date by how long the change takes to show up, not by how impatient you are. Then give every change a verdict, undo the rollbacks, and keep the log. The log becomes the most valuable growth asset you own.
Walk into almost any company and ask: of the changes your team made to grow last quarter, which ones worked? You will usually get a confident answer about one or two, a shrug about the rest and no written record of any of them. New landing page. New pricing table. Dozens of product descriptions rewritten. A navigation change. All shipped. Almost none checked.
That is not growth. That is activity. And activity without verdicts has a nasty side effect: the bad changes stay in place, quietly costing money, next to the good ones.
What should a baseline include?
A verdict needs something to measure against, so every engagement starts with a baseline both sides sign in week one. It contains:
- One number that pays the bills, such as revenue from organic search, qualified leads or repeat orders.
- Where that number is today, the date, and the source it came from.
- What is already moving it: seasons, launches, a recent price change.
Then both sides sign it. We mean that literally. By month three, memories of the starting point drift, usually in the direction that suits whoever is talking. A signed baseline ends that argument before it starts.
Pull the baseline from the source, never from memory or a spreadsheet someone typed. At Desify, the global commerce company where our engine was built, a monthly ad cost was entered as an assumption instead of read from the source. It understated a month's spend by 22% and made that month look far healthier than it was. If it was not pulled from the source, it is not a number.
What is a north star metric?
A north star metric is the one number that best captures the value a company delivers and the revenue that follows. Product teams use it to align everyone on a single goal.
For growth experiments, the idea is useful with one adjustment. Pick a number that moves within weeks, that you actually control and that connects to revenue. "Brand awareness" fails all three tests. "Revenue from organic search" passes. That number is the one your baseline records and every experiment is judged against.
How do you measure a growth experiment?
Each change is logged with five fields before it ships:
- What changed. Specific enough that someone else could undo it.
- Why. The reason we believed it would help.
- The date. When it went live.
- The expected effect. Which number should move, and in which direction.
- The judgement date. When we will decide.
The fifth field is the one most teams skip, and it is why most teams never decide anything.
Measurement then comes down to one question on the judgement date: did the baseline number move, in the expected direction, by then? Not "traffic looks healthier". Not "engagement is up". The number both sides signed.
How long should a growth experiment run?
Long enough for the change to show up, and no longer. Judging too early kills good work. Judging too late keeps bad work alive.
For search, Google has given everyone a reference. At Search Central Live Deep Dive Europe in Barcelona on 2 October 2026, Google's Gary Illyes shared typical timings, as recapped by John Campbell of ROAST and reported by Search Engine Journal. He also warned that the stages are connected, so a delay in one carries into the next.
| Change | Typical time for Google to reflect it (Illyes, 2026) | When we judge it |
|---|---|---|
| New page published | Discovered in about 20 hours; end-to-end indexing about 1.5 hours | 2 to 4 weeks, once it can earn impressions |
| Sitemap submitted | Processed in about 24 hours | With the pages it lists |
| Title or snippet rewrite | 1 to 2 days | Within one to two weeks |
| Canonical clean-up | 1 to 3 weeks | After about a month |
| Site or domain move | 1 to 3 months | After a quarter |
The right-hand column is our practice, not Google's. We write the judgement date into the log on the day the change ships, so nobody panics early and nobody waits forever.
When should you roll back a change?
Roll back when the change has run its clock and the number did not move in the expected direction. Then actually undo it.
Not "let's give it another month" without a reason. If there is a genuine reason to extend, such as a seasonal dip or a tracking fault, write it in the log with a new judgement date, and that extension gets judged too.
Founders sometimes flinch at rollbacks, as if each one were a mistake on the invoice. The opposite is true. A rollback is a change that was tested, measured and removed before it could cost more. A team that never rolls anything back is not a team that never makes mistakes. It is a team that never checks.
Every change that reaches its judgement date ends in one verdict: kept or rolled back
The second brain: what the log remembers
The log does a second job. Every decision, test and learning stays written down with its verdict. We call it the second brain. When people change, the memory stays. When a similar question comes up a year later, the answer starts from evidence, not opinion.
Some of the most useful entries are not from a client at all. They come from Desify, where our engine earned its first revenue before it touched anyone else's company. A few of them:
The catalogue that vanished. From 8 August to 8 September 2026, Desify's whole catalogue was hidden from Google Shopping. A warehouse setting plus an expired access token. About 1,500 impressions a day, lost, and nobody noticed for a month. The rule now: every session sweeps for anomalies, not just the question in front of us.
The setting that blocked whole carts. Desify holds no stock: vendors make fresh and the order ships. Yet an inventory setting that refused orders at zero stock blocked whole carts and pulled products from Google Shopping. It held one customer's order for two days. The fix was a sweep of 2,450 product variants in one pass, with zero left on the wrong setting. The rule: re-run the sweep after any bulk import.
The page that forgot what month it was. Festival pages stayed on one festival two weeks into the next season. The fix: evergreen titles, and every change verified by fetching the live page, not by trusting the edit screen.
The collection that grew from 9 to 212. A bulk change tagged 212 products into a festive collection meant to hold about nine. It was reverted four days later. The lesson: small collections are a deliberate choice, and rules need to be written down, not assumed.
Every one of these entries now shapes the checks we run for clients. That is the compounding part. Clients benefit from mistakes they never had to make.
How to run growth experiments in your company, starting Monday
You do not need us to start. You need a shared document and the discipline to use it.
- Write a baseline. One number, today's value, the date, the source. Get it signed by whoever owns the budget.
- Open a log with the five fields above.
- Before anything ships, fill in the expected effect and the judgement date.
- Once a month, walk the log. Every entry past its date gets a verdict: kept or rolled back.
- Undo the rollbacks. Actually undo them.
Do this for one quarter and you will know more about what grows your company than most teams learn in a year. If you have no growth team to run it, start with what a growth team does and how to grow without one.
Common questions
How do you measure a growth experiment?
Before it ships, log what changed, why, the date, the number it should move and the date you will judge it. On the judgement date, check one thing: did the signed baseline number move in the expected direction? If yes, keep the change. If not, roll it back. Proxy signals like engagement do not count unless they are the agreed number.
How long should a growth experiment run?
Long enough for the change to show up and no longer. For search, Google's Gary Illyes said in October 2026 that title changes typically show in 1 to 2 days, canonical changes in 1 to 3 weeks and site moves in 1 to 3 months. Set the judgement date from that clock on the day the change ships.
What is a north star metric?
A north star metric is the single number that best captures the value a company delivers and the revenue that follows. For growth experiments, choose a version that moves within weeks, that you control and that connects to revenue, such as revenue from organic search or qualified leads. Every experiment is then judged against it.
What should a baseline include?
A baseline should include one number that pays the bills, its value today, the date, the source it was pulled from, and anything already moving it such as seasons, launches or recent price changes. Both sides should sign it in week one, because by month three people remember the starting point differently.
When should you roll back a change?
Roll back a change when it has passed its judgement date and the agreed number did not move in the expected direction. Then actually undo it. Extend only for a written reason, such as a tracking fault or seasonal dip, with a new judgement date. A rollback is the system working, not a failure.
Where a Chief Growth Officer fits
This log is how CGO by ClappX runs every engagement, and it is the first thing a new client sees in the first 30 days of a Chief Growth Officer.
On Organic Growth we execute and own the number: search, the first answer a buyer sees, site structure, listings, catalogue, content, authority and tracking. On Intelligent Growth we advise and you decide: go-to-market, paid strategy, pricing and positioning, conversion and acquisition cost, retention, new markets.
Our commercial model depends on the log. One flat monthly fee, less than one senior hire, plus a share of organic growth only, agreed after month one against the baseline we both signed. Never a share of paid revenue. Month to month, no lock-in. If the numbers are not moving, stop paying us. That promise only works if both sides can see, in writing, what moved the number and what did not. The log is how we keep ourselves honest.
Want to see what a Chief Growth Officer would do in your first 30 days? Talk to a founder: human@clappx.com