// dashboard: jonas_alves — status: RUNNING

Opinion is input.
Data decides.

I'm Jonas Alves — founder & Chief Innovation Officer of ABsmartly. I've spent 15+ years making A/B testing the default: first scaling Booking.com's experimentation platform to over a thousand simultaneous experiments, now bringing that capability to every serious product team.

Read the experiment log LinkedIn ↗
years_in_experimentation 0+ ▲ still compounding
concurrent_experiments_at_booking 0+ ▲ full-factorial, monitored
confidence_in_this_career_choice p < .001 ▲ significant

01 Experiment log

A career, written the way I think about everything: hypothesis first, then results.

EXP-2008-001 2008 → 2012 · Booking.com SIGNIFICANT ▲

Full-stack developer → Product Owner, Experimentation Platform

hypothesis: testing infrastructure can grow from a basic tool into the company's central decision system.

result: transformed Booking.com's testing infrastructure into a sophisticated platform handling 1,000+ simultaneous experiments, with real-time monitoring and interaction detection across full-factorial designs. Watched a culture form where nobody shipped a change without thinking about how to test it — opinion became input, data made the decisions.

EXP-2012-002 post-Booking · Founder & CEO, ABsmartly SIGNIFICANT ▲

Founding ABsmartly

hypothesis: Booking-grade experimentation shouldn't be reserved for companies that can afford to build it themselves.

result: built an experimentation platform for sophisticated product teams — real-time results, data ownership, and guardrails against toxic wins baked in. The founding conviction: "to do experimentation at scale you might need to change everything." It's a culture change, not a tool purchase.

EXP-NOW-003 today · Chief Innovation Officer, ABsmartly ● RUNNING

Chief Innovation Officer

hypothesis: as AI writes software faster than ever, knowing what actually works becomes the bottleneck — and that's an experimentation problem.

status: leading the platform's evolution at the intersection of AI agents and experimentation. Data still being collected.

02 Guardrail metrics

Six principles I monitor on every team I work with. When one of these dips, the "wins" stop meaning anything.

M1

Make A/B testing the default

Test every customer-facing change, regardless of where it came from. Shipping untested should be a conscious decision — not the norm.

M2

Decouple deployment from release

Deploy dozens of times a day; control visibility with feature flags. Shipping speed and release safety are different problems.

M3

Democratize experimentation

Remove every barrier between an employee with a hypothesis and a running test. Centers of excellence should enable — never own.

M4

Invest in real-time monitoring

If you can't see an experiment hurting you as it happens, you'll only find out after it has. Infrastructure is the culture's foundation.

M5

Guard against toxic wins

A primary metric going up means nothing if cancellations, support contacts, or performance quietly go down. Define guardrails before the test starts.

M6

Small changes, compounding impact

1% here, 0.5% there — across tens of thousands of experiments a year, incremental learning beats big bets.

03 Speaking & writing

TALK

Scaling Experimentation with Confidence — Lessons from Booking.com

How experimentation culture is built phase by phase: first a small team learning to test, then platform + method, and only then experimentation as the default. Why most companies fail when they try to skip straight to phase three.

PODCASTS

Pioneer conversations on experimentation

Featured across industry podcasts on experimentation methodology, platform design, and what it takes to change how organizations make decisions.

WRITING

Industry commentary & best practices

"10 A/B Testing Resolutions for 2026," analysis of industry moves like the VWO/AB Tasty merger, and regular writing on experimentation done right.

04 Current focus: AI × experimentation

AI is collapsing the cost of building software. Agents write the code, ship the feature, open the PR. What AI doesn't tell you is whether any of it worked — whether real users, with real money and real patience, are better off.

That's the bottleneck moving: from can we build it? to did it work? Which is — and always was — an experimentation problem. My current work at ABsmartly lives exactly at that intersection: making experimentation as native to AI-driven development as version control is to code.

05 Get in touch

Based in Lisbon, Portugal. Always happy to talk experimentation, platform design, or why your biggest win last quarter might have been a toxic one.

LinkedIn↗ ABsmartly↗