Noé Geste
ABLab v2.0 · internal tool · Lenus design team

The A/B-testing workspace I built for my team.

ABLab turns a VWO export into a Bayesian verdict, then saves the result as something the team can find and reuse.

Free for the team Google sign-in Approved by Legal & Security
The ABLab Running tab: sidebar with Running, Patterns, Ideas, Pipeline and Tasks, smart alerts and live experiment cards
60,000
Monte Carlo draws per probability
3
funnel stages scored on every test: Visitor → Lead → Client
12
built-in behavioural hypotheses, plus custom ones
90%
probability required to call a winner
The before

One result meant three tools.

Visitor counts, coach IDs and conversions lived in different places. I stitched them together in a spreadsheet, then shared a screenshot.

Hard to read

A spreadsheet screenshot did not tell a key account manager what to implement or why.

  • No history in one place
  • Numbers without a decision

Slow to repeat

The same export work returned every week, across designers and markets.

  • Too many tabs
  • Tracking depended on memory

Easy to forget

Answering "have we tried this in Finland?" meant digging through old files.

  • No shared taxonomy
  • Results lost their hypothesis
How it works

From export to verdict in minutes.

It works with the exports our pages already produce. No script to install.

step 01

Create

Name the coach and test, choose a hypothesis, then tag the markets and design changes.

Inger Marie · Hero Social Proof Split
social proof · NO · DK · SE · FI
step 02

Upload

Drop the CSV, add visitor counts and generate the dashboard. New exports become weekly snapshots.

lead-experiments.csv
Data → Experiment ID, State
2514-a, 1 · 2514-b, 5 … 12,163 rows
step 03

Decide

Review three funnel stages. At 90%, finalize the verdict and save the learning to the Results Library.

Implement variant
Two columns are enough. -a is control, -b is variant, state 5 is a won client.
Features

The numbers lead to a verdict.

ABLab runs 60,000 Monte Carlo draws per funnel stage. It calls a verdict at 90%.

A probability I can explain.

For each stage, ABLab draws control and variant samples from Beta distributions and counts how often the variant wins. The result stays separate for each stage.

  • No separate sample-size calculator
  • Nudges start at 85%, a verdict lands at 90%
  • Every stage scored on its own
Visitor → Lead96.4% Winner
Lead → Client68.8% Inconclusive
Visitor → Client92.0% Winner

The hypothesis stays attached.

Every test records the belief behind it. Finished results build a history of wins, losses and inconclusive calls.

  • Twelve behavioural hypotheses out of the box, plus your own
  • One searchable history of what held up
The Patterns tab: a hypotheses leaderboard with win-rate bars, radial rings and pattern cards

Three stages, one funnel.

Visitor to lead, lead to client and visitor to client stay separate. A sign-up win that hurts close rate cannot hide inside an average.

  • Sign-up, win and overall rate reported separately
  • Trade-offs visible on the first screen
Visitor → Lead · 12.40 → 15.90%+28.2%
Lead → Client · 7.76 → 7.13%−8.1%
Visitor → Client · 0.96 → 1.13%+17.8%

Alerts that read the numbers.

The Running tab flags tests ready to finalize, falling variants and patterns worth trying elsewhere.

  • Consider reviewing · Mandyfit · 87%
  • Winning pattern to replicate · Christina S. · +18%
  • Missing learnings · CTA Color Test
AI verdict card: Implement Variant, with a three-sentence summary and a one-line reason
Built-in strategist

A verdict people can use.

The built-in strategist explains the numbers in plain language and suggests three next tests. A key account manager can forward the conclusion as written.

AI verdict card: Implement Variant, with a plain-language summary
Suggested next experiments with priority pills, hypothesis text, tags and a Save as Idea button

A suggestion can become a pre-filled test in one click.

Adoption

I built it alone.
The team kept using it.

There was no budget or engineering team. It caught on because it made a weekly job easier.

150+

tests run across four Nordic markets.

4

designers using it in daily practice, with one shared language for hypotheses.

1

person to design, build and ship it, with Claude as the development environment.

One result became its own case study: a food-first redesign tested across 14,304 visitors. The variant produced 11 clients from 72 leads, against 6 from 89. ABLab made that funnel trade-off visible.

Read that experiment
Integrations

It fits the tools we already use.

No migration: existing exports in, existing accounts for sign-in.

VWOthe CSV export it reads
Google Sheetswhere results used to live
Firebaseauth and Firestore
Vercelwhere it is deployed
Google sign-inauthorised accounts only
Metabasecoach IDs, linked once
Anthropicthe built-in strategist
Strategy

One shared focus each month.

Strategy reviews finished tests, summarizes what held up and assigns one hypothesis across designers, coaches and markets.

Proven patterns

Repeatable wins, with the tests behind them.

Disproven assumptions

What we expected, what happened and what we learned.

Market insights

How Norway, Denmark, Sweden and Finland respond differently.

Focus Test

One coordinated assignment per designer and coach.

Strategy report cards: Proven Patterns, Disproven Assumptions and a Hypothesis Quality Assessment
Four Focus Test assignments, one per designer, each with its coach and market

What changed for us.

ABLab replaced my weekly spreadsheet ritual with a shared record of what we tested, what won and what to try next.

ABLab

An internal experiment tracker for the Lenus growth and design team. Designed, coded and shipped solo, then approved by Legal & Security.

The product

Elsewhere

© 2026 ABLab · internal experiment tracker v2.0 Made with 60,000 Monte-Carlo draws and a lot of exports