HOME/FEATURES/EXPERIMENTS
CAPABILITY · EXPERIMENTS

Ship experiments. Measure what matters.

Visual A/B builder, split-URL, multivariate, multi-armed bandit, and NBA per-visitor personalization — all driven by Bayesian stat-sig math with automatic winner promotion at 95% confidence.

Watch 2-min demo ▸
BAYESIAN + FREQUENTISTSEQUENTIAL TESTINGNBA ARCHETYPES
TEST · HERO CTA● LIVE · 7 DAYS
CONTROL
Start free trial
2.40%
VARIANT
Start free trial →
3.24%+35%
BAYESIAN CONFIDENCE62% / 95%
▸ AUTO-PROMOTES AT 95% · EST 2.3 DAYS
THE PROBLEM

Every test type lives in a different tool. Your stack is the problem.

A/B in Optimizely, MAB in a different SaaS, multivariate in a third — and none of them know about each other. Flight Deck puts every experiment type under one roof with a shared math engine, a shared audience layer, and a shared analytics backbone.

◆ CAPABILITIES

Every experiment type, one math engine.

VISUAL A/B

No-code editor. CSS-selector precision.

Point-and-click to change text, styles, images, and layout. For anything trickier, paste HTML. Preview across device, traffic source, and locale before launch — the variant config is edge-cached so visitors see it in under 50ms with zero flicker.

  • Point-and-click edits to text, style, image, structure
  • HTML paste for complex variants
  • Device + locale preview before launch
  • Sub-50ms flicker-free delivery via edge cache
PRODUCT VISUAL — PLACEHOLDER
SPLIT-URL + MULTIVARIATE

Full landing rewrites. Interaction effects.

Split-URL (Growth+) routes traffic between full page variants — great for landing page rewrites where the whole layout changes. Multivariate (Performance+) tests combinations of elements simultaneously so you can see interaction effects, not just main effects.

  • Split-URL tests for landing page rewrites (Growth+)
  • Multivariate for interaction effects (Performance+)
  • Factorial + fractional factorial designs
  • Sample size calculator shows you what you need
PRODUCT VISUAL — PLACEHOLDER
MULTI-ARMED BANDIT

Adaptive allocation when time is money.

When waiting for stat-sig is more expensive than exploring, MAB shifts traffic to better-performing variants as evidence accrues. Great for flash-sale banners, promo headlines, and anywhere lost opportunity cost exceeds learning value.

  • Adaptive traffic allocation based on live performance
  • Thompson sampling — Bayesian under the hood
  • Minimum-traffic floor per variant (avoid premature convergence)
  • Auto-switch from MAB to A/B when the winner is clear
PRODUCT VISUAL — PLACEHOLDER
NBA PERSONALIZATION

Per-visitor archetypes, not per-cohort tests.

Next Best Action publishing (Growth+) clusters your audience by Recency × Frequency × Monetary behavior, assigns an archetype per visitor, and serves the variant that works for that archetype. No data team required. Recomputes nightly at 3am UTC.

  • RFM behavioral clustering — fully automated
  • Per-archetype variant assignment, updated daily
  • Graceful fallback for first-visit / anonymous traffic
  • Auditable in analytics — see which archetype is winning where
PRODUCT VISUAL — PLACEHOLDER
STATS + GUARDRAILS

Stop underpowered tests before they ship.

Bayesian credible intervals by default, sequential testing with Pocock / O'Brien-Fleming bounds when you need early-stop math, and guardrail metrics that catch harmful variants before they roll out. Power analysis baked in so you never launch an underpowered test.

  • Bayesian (default) + frequentist side-by-side
  • Sequential testing with early-stop bounds
  • Guardrail metrics on every experiment
  • Built-in sample size + time-to-completion calculator
  • Auto-start, auto-stop, auto-promote at 95% CI
PRODUCT VISUAL — PLACEHOLDER
◆ TECHNICAL SPECIFICATIONS

Under the hood.

EXPERIMENT TYPES
Classical A/B, split-URL (Growth+), multivariate (Performance+), MAB, NBA (Growth+)
MATH ENGINE
Bayesian (primary) + frequentist. Sequential testing with Pocock / O'Brien-Fleming bounds.
TARGETING
Geo, device, traffic source, URL regex, custom attributes, exclusion groups
ALLOCATION
1–100% traffic, canary-friendly. MAB via Thompson sampling with min-traffic floor.
GUARDRAILS
Configurable harmful-variant thresholds on MRR, refund rate, latency, error rate
SDK
< 9KB gzipped JS, loads async, median decision < 50ms
CONCURRENT LIMIT
1 / 2 / 5 / 10 / unlimited (Free / Launch / Growth / Performance / Enterprise)
ANALYTICS
Matomo-backed; real-time visitor feed with variant overlay
DEPLOYMENT
WordPress plugin for one-click deploys with automatic rollback
◉ READY FOR DEPARTURE

Ship your first variant this afternoon.

One-line SDK. Bayesian math out of the box. First winner typically in 2–4 weeks depending on traffic.

Book a demo