Product experimentation
Product experimentation for teams whose A/B tests happen sporadically and whose learnings don’t carry into the next quarter. We install the practice with your team: hypothesis work, prioritisation the team owns, a cadence, and a register that makes the learning compound.
The problem this solves
The tooling is usually already there, but the process around it is missing: tests get run when someone champions one, results live in slides nobody reopens, and next quarter starts from zero. Prioritisation runs through a single person, which caps how fast the team can test. And because most public experimentation advice is written for sites with enormous traffic, teams with normal B2B volumes quietly conclude that testing isn’t for them.
Product experimentation is a different discipline from marketing-site conversion optimisation. The tests live in the product, the metrics are activation and retention rather than campaign conversion, and the constraint is rarely traffic. In our experience the constraint is how prioritisation happens and whether learnings have anywhere to accumulate.
What we actually do
- Set up the practice. A shared hypothesis template, a prioritisation method the team scores itself, and a testing cadence with clear ownership. At Mentimeter the rule that made it work was simple: only the team scores the backlog, consultants included.
- Run experiments together on data you can trust. We work through the first rounds of tests with your team, checking the experiment tool’s numbers against your own data so a result means what it says.
- Make the learnings carry. Results, including the inconclusive ones, go into a hypothesis register so each quarter starts from what the last one learned. Reading a null result honestly is part of the practice.
What you leave with
- A prioritisation method the team runs and scores itself, sprint after sprint.
- A testing cadence that runs without a consultant in the room.
- A register that carries learnings across quarters instead of starting over.
Who it’s for, format, and duration
This is for a CPO, Head of Product, or growth lead whose team has the tools and the will but no practice: tests don’t compound, velocity is capped by one person, or experimentation simply isn’t part of how the team builds. A working prioritisation method, and metrics close enough to the change to move, matter more than traffic volume.
If you don’t yet know which growth mechanism to test against, our Growth Loops Design Sprint is the one-time sprint that finds it and leaves a loop-mapped backlog. The work on this page is the ongoing practice that runs and compounds the tests afterwards.
There are several ways in: a lightning talk to make the case internally, a focused workshop (Kognic’s was 2.5 hours), coaching for the person who will own the practice, or an embedded engagement over several months like Mentimeter’s. Most teams that want the practice to stick land in the embedded format.
One honest note on scope: we install the practice with your team, and keeping the cadence and the register alive after we leave is your team’s job. That handover is the point.
Book a call
If your team runs tests but the learnings evaporate between quarters, email hello@scilla.studio and we’ll book 20 minutes to see whether your needs and our skills are a match.
Talk to Joni about your situation
Take 20 minutes. We'll talk through what's going on and whether this is the right match. No sales pitch.
Book 20 min with Joni →More client proof
- Mentimeter Once the team owned prioritisation, 43 experiments ran; a shorter enterprise form lifted sales contacts by half. Read the case →
- Kognic Seven engineering teams had the tool but no process; a workshop left each with a template and a named next step. Read the case →
- Apotek Hjärtat A hypothesis register carries each quarterly cycle's learnings into the next one. Read the case →
- Tradera Four-week sprints with one experiment goal per sprint made testing a standing rhythm. Read the case →
- H&M Eleven teams trained on experiment culture, grounded in a 1,400-experiment track record and its 31 percent win rate. Read the case →