Blog › Marketing strategy
Run a growth experiment every week (automatically)
Treat marketing like a product. KPI trees, hypotheses, and a fresh test every seven days.
· 4 min read
Marketing that doesn’t experiment is just guessing louder. The fix is to treat it like a product: hypotheses, tests, and honest measurement.
One small experiment a week is enough. It fits in the time you already spend on marketing, and after a quarter you have twelve answers where you used to have twelve opinions.
A fresh test every seven days
Build a KPI tree from your goal, form one hypothesis, ship a variant, and measure lift. Kill the losers, scale the winners, and log what actually moved the needle.
One disciplined experiment a week is fifty-two compounding bets a year.
The discipline is in the word "one". Change five things at once and you learn nothing about any of them.
Step 1: build a KPI tree
A KPI tree breaks your goal into the numbers that produce it, until you reach something you can change this week.
More customers
├── More enquiries
│ ├── More profile visits
│ │ └── Better first lines on posts
│ └── More clicks to the site
│ └── A clearer offer in the post
└── More enquiries that become customers
└── Faster first reply
The top is the outcome. The bottom is where experiments live: a first line, an offer, a reply time. If you cannot draw the path from a test to the goal, the test is not worth running.
Step 2: write the hypothesis first
Before you publish anything, write one sentence:
We believe that [change] will [effect] because [reason]. We will know by [measure] within [time].
- "We believe that opening posts with a customer's problem will earn more replies, because readers recognise themselves. We will know by comparing replies on two posts within a week."
Writing it first is what keeps the test honest. Afterwards it is too easy to decide that whatever happened was what you expected.
Step 3: change one thing
Make two versions that differ in exactly one respect.
| Test | Version A | Version B |
|---|---|---|
| First line | States the result | Names the problem |
| Format | Text only | Text with a picture |
| Length | 80 words | 200 words |
| Call to action | A question | A link |
| Time | Morning | Evening |
Keep the topic, the channel and the audience the same. Publish them close together in comparable conditions.
Step 4: measure honestly
- Decide the measure in advance. Replies, clicks, enquiries. Not whichever looks best afterwards.
- Give it time. A few days for most social posts.
- Compare like with like.
- Be modest about small numbers. Twelve replies against nine is not a law of nature. Run it again before you rely on it.
Most tests will show no clear difference. That is a result too: it tells you that variable does not matter much for your audience, and you can stop arguing about it.
Step 5: keep a log
A test you do not write down, you will run again in six months.
| Date | Hypothesis | Result | Decision |
|---|---|---|---|
| 3 Nov | Problem-first lines earn more replies | 14 vs 6 replies | Adopt; retest in a month |
| 10 Nov | A picture increases clicks | No clear difference | Drop; not worth the effort |
After a quarter, this page is your playbook: what works for your audience, in your words, with evidence.
What to test first
Start near the bottom of the tree, where changes are cheap.
- First lines.
- The offer, or the next step you ask for.
- Format: text, picture, short video.
- Topic, between your content pillars.
- Timing.
Leave pricing and positioning until the small tests have taught you how your audience behaves.
Let an agent run the cadence
Let an agent run the cadence and the backlog of “we should test that” finally gets tested.
The reason experiments stop is not that people doubt them. It is that nobody has time to think of the next one. That part can be scheduled.
- Ideas, weekly. FinalPost AI's Growth Metrics agent delivers a north-star KPI tree and three experiments ranked by impact on each run. Put it on a weekly schedule and a fresh shortlist arrives every seven days. The CVR Optimization agent adds two ready-to-run tests for your pages.
- Variants, drafted. Agents draft the versions for your approval.
- A place to keep them. A/B tests group posts from your queue as labelled variants under a hypothesis you write first.
- The log. Declare a winner and it is recorded with the experiment.
Be clear about what is automatic and what is not. The ideas and the drafts arrive on a schedule. Publishing still waits for your approval by default, and judging the winner is deliberately manual: each variant links to the live post, you compare the results there, and you decide. A winner is declared by a person who looked, not by a number that might be measuring something else.
Review the result in your weekly marketing review, and choose next week's test.
Frequently asked questions
What is a growth experiment?
A small, deliberate test of one change, such as a different first line or format, with a prediction written beforehand and a measure decided in advance.
How long should a marketing experiment run?
Long enough to gather a fair response. For social posts that is usually a few days. For changes to a web page, often two weeks or more.
What if most of my experiments fail?
That is normal and useful. A test that shows no difference tells you where not to spend effort. A few clear wins a quarter is a good outcome.