Experimentation & growth

Test fast without flying blind.

Experimentation is not a collection of tricks. It is a discipline: a growth model that shows where to look, instrumentation that lets you read the result, and a protocol that makes outcomes easier to interpret when the available volume is sufficient.

Request a diagnostic

Direct answer

What does a serious experimentation practice actually involve?

Breaking revenue down into actionable variables, instrumenting measurement at each step, writing explicit hypotheses, then testing them one at a time with a duration and stopping rule fixed before launch. What holds becomes a documented standard; what does not is dropped and logged.

Important considerationNo aggressive scraping, no mass unsolicited outreach, no dark patterns, no bypassing platform rules. And no guaranteed gain: a test can very well produce no readable effect.

The frame

What we do not do.

No aggressive scraping, no mass unsolicited outreach, no dark patterns, no bypassing platform rules. These practices can create short-lived gains while increasing brand and compliance risk.

What remains is more demanding and more solid: formulating hypotheses, testing them properly, keeping what holds, and industrialising it.

Approach

From growth model to industrialisation.

Experiment passport

Growth model: Break revenue down into actionable variables: volume, conversion by step, value, retention. We test where the lever actually exists.
Instrumentation: Without reliable measurement at each step, a test proves nothing. Reading precedes experimentation.
Backlog: Hypotheses stated explicitly: what changes, for whom, and the expected effect on which variable.
Prioritisation: Estimated impact, confidence in the hypothesis, implementation effort. Costly, low-confidence tests wait.
Protocol: One variable at a time, defined population, duration fixed before launch, written stopping rule.
Reading: An honest outcome: conclusive, inconclusive, or unreadable for lack of volume. All three are useful.
Industrialisation: What works becomes a documented standard, built into processes and permanent pages.

Example framework used during the engagement. It is adapted to your context, data and decisions.

Reading condition

Without volume, a test is not wrong: it says nothing.

The first decision is not what to test, but whether the effect sought can actually be read over the planned duration. Below that threshold, an apparently positive result is noise.

When reading is not possible, we say so and redirect the effort: customer research, measurement, or fixing the offer and the site.

Before testing: Reliable measurement at each step and the capacity to ship variants.
Instead: Customer research and structural fixes when volume is missing.

Deliverables

What the engagement produces.

Growth model built from your own data.
Prioritised, maintained hypothesis backlog.
Test sheets: hypothesis, protocol, duration, success criterion.
Results log, including inconclusive tests.
Standards derived from validated tests, documented for your teams.

Fit

When experimentation pays off.

Relevant
A steady flow of visitors or prospects, workable measurement, and the ability to ship tested variants quickly.
Better postponed
Volume too low to read an effect, unreliable measurement, or structural issues — positioning, offer, site — no test can offset.

Frequently asked questions

How is this different from strategy?

Strategy decides the direction and the structuring trade-offs. Experimentation optimises execution inside that direction. Testing intensively without a direction just means accelerating without knowing where you are going.

How many tests per month?

It depends on traffic, shipping capacity and the volume needed to reach a conclusion. We prefer a few properly run tests to a stated pace the data cannot actually support.

From what threshold does a result count?

We fix the duration, the population and the minimum meaningful difference before launch. A test stopped as soon as it turns favourable produces a misleading conclusion.

What resources do we need to commit?

A decision-maker as point of contact and access to shipping — development or a testing tool. Without the ability to deploy, the backlog just accumulates.

What happens to tests that fail?

A test can produce no readable effect. It still has value if it rules out a hypothesis and the decision is documented.

How do you scale a validated result?

A validated result is turned into a standard: built into templates, campaigns or processes, then monitored to check it holds over time.

Next step

Let's check whether your volumes allow for useful testing.

A scoping call identifies the growth variables where a test can actually be read.

Scope your testing programme contact@sevengoldagency.com