From first idea to money in your bank

Prediction-based experimentation:
the method behind the guarantee

Nobody else has a database of 4.3 million A/B tests, so nobody else can do this: know an idea's odds before it costs you a single day of traffic. Here's the machine we built around that advantage.

See if your shop is a fitThe guarantee, in full

Testing makes money on three levers. Most programs fail all three.

Whether testing makes you money comes down to simple math:
Growth = Test quality × Test rate × Win rate

Test quality fails first. Your next test idea is almost never new, somebody already ran it on a shop like yours, you just don't know what happened. Brainstorms and best-practice lists mean you're betting four weeks of traffic on ideas nobody ever checked.

Test rate fails quietly. Most shops run one test at a time, afraid that parallel tests contaminate each other, a fear that actually changes a decision in roughly one in 50,000 cases (Microsoft checked, and Booking.com runs over 1,000 at once). Twelve tests a year is learning slower than the market moves.

Win rate fails last and hurts most. Guessing wins about 1 in 5 in the industry, and some of those "wins" are fake, called early at soft confidence, shipped, and quietly costing you money forever.

Multiply three weak numbers and you get what most shops have: a program that costs more than it returns, which is why most of them are dead by month four.

And multiplication cuts both ways. Push all three levers at once and the uplifts compound: small monthly gains stack into double-digit growth in six months, which no single hero test ever delivers (across 252 companies, Wharton found lone tests rarely move revenue even 2%).

Prediction-based experimentation is built to push all three at once. Here it is, lever by lever.

Phase 1 · moves Test Quality

Research: we find out why your customers buy

The first month is all research. No tests, no "quick wins", research. We read your reviews, surveys and support tickets, and your competitors' reviews too. Our research software runs ten analyses on all of it: buying motivations, psychological drivers, personality patterns, category entry points, brand perception, the emotional customer journey and more. We put 40+ hours into your session recordings and heatmaps to see where real shoppers get stuck or leave, and run revenue leak detection across your whole funnel: which slider image actually makes money, which filters have the worst conversion rate, which payment method quietly loses finished orders. Then our behavioural psychologists make sense of all of it.

You get a 20+ page research report: the psychological drivers behind your buyers, the moments that bring them into your category, and exactly where your funnel leaks revenue. Every test idea starts here, written as if / then / because:

Phase 2 · moves Win Rate

Prediction: every idea gets scored before it gets built

Every idea from the research goes into Apex and gets a weighted impact score across three fields: revenue exposure, how much money the pages it touches actually handle, its Likelihood to Win, and the effort to build it. The likelihood comes from the 4.3 million tests we track, the 4,000+ we ran ourselves, and everything the research just taught us about your shop.

The database isn't a recipe book, it's context. Scarcity timers lift revenue on impulse buys and hurt it on high-consideration products, so the exact tactic that wins on a fast-fashion shop backfires on a luxury one. Across 4,000+ of our own tests, not one concept has won every single time. What the score actually reads is which concept fits which shop, and that's the part you can't google.

The scores decide the order, and they don't sit still: every two weeks the ranking recalibrates on your results. Weak ideas die on the list, not on your traffic. What's left is a test plan where every idea earned its spot. That's your first month: research done, plan scored, and the first tests already live. Curious about the tool itself?

How we built the genome →
Explore Apex

Phase 3 · protects all three levers

Build and QA: tests that don't break your shop

Each test starts with a brief that carries the hypothesis, the shopper context and the design rationale, so you always see why, not just what. Then 3 to 5 designed variants for mobile and desktop that you approve before a line of code gets written. We build at code level, no visual editors: they slow pages down, cause flicker, and quietly bias results against your variants.

Before launch, a 10-person QA team tests everything on real devices via BrowserStack, tracking checks included. A broken test on a shop with a million visitors is expensive. Ours don't ship broken.

Phase 4 · moves Test Rate

Run: six to ten tests at once, statistically clean

Most agencies run one test at a time, because they're afraid parallel tests "contaminate" each other. The fear doesn't survive the math. Every visitor is assigned to each test independently, so every other test is split evenly across your control and your variant, and its influence cancels out on both sides. That's where the one-in-50,000 number comes from. And testing one by one doesn't even protect you, the winners all go live together in the end anyway, as a combination nobody ever tested. So we run six to ten at once, splits held at 50/50, sample-ratio mismatch monitored throughout. By month three, more experiments have run on your shop than most brands ship in a year, and the first winners are live and earning.

The statistics are pre-registered before launch: frequentist method, 80% confidence and 80% power, minimum detectable effect and duration fixed upfront, and no early stopping, ever. A fake win is worse than a loss. A loss you switch off, a fake win ships and costs you forever, which is why nobody here gets to peek. And we run 80% instead of the academic 95% on purpose: on a shop, waiting for 95 costs more in winners you never shipped than it saves in false alarms. These thresholds are set to make money, not to pass a journal review, and we publish them so anyone can audit our win rate.

Phase 5 · compounds all three

Learn: every result makes the next one smarter

We measure results in revenue per visitor, not clicks, and read them as a portfolio: 40 to 50% of tests come back negative or unclear, and we say so upfront. Every test lives in a four-stage system, research, backlog, pipeline, results, and stage four is the part nobody else has: every outcome gets filed by the buyer motivation it confirmed and the friction point it exposed, not by date. That's a memory you can query, not a results list.

You watch all of it on three dashboards: the research memory, the experiment pipeline, and an executive view with revenue per visitor, shipped winners and the next best bets. The loop is why the win rate keeps climbing, 27% in 2024, 55% last quarter. And month six is when the guarantee comes due: the lift gets measured, and everything we learned is documented and belongs to you, even if you ever leave. At OceansApart, month six looked like +€443,466 in monthly revenue.

We put odds on every test.
Here's how the odds did.

The Likelihood to Win goes into Apex before a test launches, so anyone can hold it against what actually happened. A few recent tests:

TestScore before launchWhat happened
1Discounted-price display · OceansApart71+€43,876 per month
2PDP subline · OceansApart66+€28,934 per month
3Size-guide rebuild74+3.1% revenue per visitor
4Trust-seal placement52Flat. Published as inconclusive, not spun into a win
5Free-shipping countdown58−2.1% revenue per visitor, switched off

40 to 50% of tests come back negative or unclear, and the odds don't make losses vanish. They change how often you hit. Every miss above went into the database and sharpened the next odds.

Your four options

You could build all of this yourself

Here's what each path really costs.

A CRO lead costs 55 to 90 thousand a year, a developer 60 to 90, a designer 45 to 75, an analyst 50 to 80, QA 45 to 70, plus about 30% employer costs.

That's 330 to 530 thousand euros a year, and 16 to 24 months before the machine runs.

Then your CRO lead quits, and the knowledge walks out the door. It can work, the best in-house programs do, backed by nine-figure revenue and a whole floor of analysts.

What money can't fix: you can hire the team, you can't hire the last eight years of data.

€500 to 2,000 a month. All the work stays with you. The tool can start tests, it can't tell you what to test, whether it'll win, or whether your numbers are honest. A machine with no memory, which is why most testing tools are dead by test three.

One or two tests a quarter, results delivered as "learnings" in a slide deck. A roadmap that freezes in month one, and when the contract ends, everything they learned leaves with the account manager.

The database, the platform, the team, and the guarantee, from day one.

Your first test idea gets scored before you've paid us a euro. Apex is included, no extra tool subscription. Most of our clients hand everything over to us, because the guarantee only stands when our people run the machine.

The gap doesn't stay the same size: with Apex versus testing without memory, month 1 to month 12

One agency. One exclusive tool.
Two ways to use us.

What you choose is how much of the work we do for you.

Done for youWe do the work

We do it all, you watch it earn.

Everything you just read, done for you. Our experts pick the ideas most likely to win, build them, QA them and read the results. You don't touch a test. Most of our clients choose this, because the guarantee stands when our people run the machine.

Full Apex access
Calls with our teamEvery two weeks
Research and insights on your visitors
Test ideas from our team
Likelihood to Win on every idea
Building test variants
QA on real devices
Launching tests on your shop
Performance reporting
Done with youYour team does the work

You drive. We navigate.

You get Apex and a ranked list of test ideas, and our team checks in every week on what to run next and what the results mean. Your hands on the wheel, but never blind: you see every idea's predicted win rate before you build it.

Full Apex access
Calls with our teamEvery week
Research and insights on your visitors
Test ideas from our team
Likelihood to Win on every idea
Building test variants
QA on real devices
Launching tests on your shop
Performance reporting
Apex backlog, ranked by win score

Apex only exists inside DRIP.

You can't buy access to it anywhere else.

15,6ROI
44,8%Winrate

"As a cooperative with multiple formats, we rely on data-driven decisions with real impact. DRIP delivers valuable insights and a strong A/B test win rate."

Timo Dunke
Digital Marketing Manager Personalisierung
DRIP and Apex

Ready to run this
engine on your shop?

If your shop does €100k+ a month, book a quick call with our team. Tell us where the shop stands and where you want it to go, and we'll tell you straight if it's a fit (and if it's not, we'll say that too).

See if your shop is a fit

30 minutes, and you'll know exactly where your shop stands.

The guarantee, in full

+10% more revenue in under six months, or month seven onward is free.

When you can predict what wins, you can promise the outcome. What counts as uplift, how it's measured, and what happens if we miss is written down in plain language.

We can promise it because we know our numbers: more than 50% of our A/B tests win, across 4,000+ documented experiments. That's when a guarantee stops being marketing and becomes underwriting.