VariFlask — AI A/B test ideas for your site
Sign In Get Started

Why VariFlask?

Ideas grounded in real intelligence. Tests built to run. Results that tell you why.

GA4 + Search Console → idea generation Always-on Continuous Loops Agent gateway: MCP · A2A · REST API ~5KB snippet From $9/mo
Core Differentiator

Ideas Built on Five Layers of Intelligence

Most CRO tools give you a blank canvas and leave you to figure out what to test. VariFlask starts by doing the research for you — studying your specific context before generating a single suggestion.

Before recommending what to test, our AI analyses five layers of context:

Your Own Data Unique to VariFlask

Connect GA4 and Google Search Console (more integrations coming) and your real traffic patterns, conversion data, bounce rates, and keyword opportunities flow directly into idea generation. We surface the specific pages leaking conversions and the queries bringing visitors who aren't converting — so every test suggestion is grounded in your actual numbers, not assumptions.

Your Site

Page structure, content hierarchy, conversion flows, existing CTAs, and user journey gaps — analysed in detail.

Your Industry

Benchmarks, typical conversion patterns, and what high-performing sites in your vertical consistently do right.

Your Competitors

Publicly available positioning, messaging strategies, and UX patterns your competitors are using to convert visitors.

CRO Best Practices

Hundreds of proven testing frameworks, hypothesis patterns, and conversion principles — applied to your specific context.

The result: a prioritised experiment backlog where every idea has a clear rationale — grounded in your real data, not generic templates. See the A/B test ideas generator with real before/after examples. No other A/B testing tool feeds your GA4 and Search Console data directly into idea generation. Each hypothesis is specific to your site, your audience, and your actual numbers.

Core Differentiator

True Viewport Impression Tracking

Most A/B testing tools — Optimizely, VWO, AB Tasty — count an “impression” the moment a page loads, even if the visitor never scrolled down to see your change. This dilutes your data and slows down your ability to reach meaningful results.

VariFlask uses the browser’s IntersectionObserver API to only count an impression when your change actually enters the visitor’s viewport. This follows IAB/MRC viewable impression standards — the same standards used in digital advertising.

The result: no diluted results, faster statistical significance, and more accurate data.

Traditional Tools
Page Load = Impression

Counts visitors who never saw your change

VariFlask
Viewport Entry = Impression

Only counts visitors who actually saw it

Engagement Intelligence Per Variant

Other tools have heatmaps (Hotjar, Crazy Egg) or A/B testing (Optimizely, VWO) — but never both tied together. VariFlask tracks per-element engagement metrics directly tied to each variant:

Viewport Dwell Time

How long visitors linger on changed elements

Hover Tracking

Mouse attention signals on each element

Scroll Depth Correlation

How far visitors scroll in each variant

Element-Level Click Tracking

Per-element clicks tied to each variant

This tells you not just if a variant won, but why — did visitors spend more time reading the new headline? Did they hover over the CTA more? No other SMB A/B testing tool does this.

AI-Powered Test Ideas with Smart Goals

AI analyzes your live site and generates specific, actionable A/B test hypotheses. Each idea comes with pre-configured goals — the AI identifies what to measure (signup clicks, form submissions, page visits) with the exact CSS selectors and URL patterns.

One click on “Test This” takes you from idea to running experiment. Ideas are delivered on a schedule you choose — daily, weekly, or on demand.

Example AI-Generated Idea

“Replace the hero headline with a benefit-driven version emphasizing free trial. Hypothesis: visitors respond more to value propositions than feature descriptions.”

Goal: CTA click on .hero-signup-btn Goal: Page visit /signup
Core Differentiator

Continuous Loops: Always-On Optimization With Defensible Results

Pick a page, pick the element types you want optimised, and launch one Continuous Loop. The loop runs forever as a champion-challenger: the current champion competes against an AI-authored challenger, the round resolves with a defensible Bayesian winner, the winner becomes the next champion, the AI authors a fresh challenger, and the loop continues — round after round — until you pause it.

Every round is deliberately conservative. A challenger only gets promoted when it clears all of these:

7-Day Minimum Per Round

Wipes out day-of-week effects and the novelty lift that fades in 1–3 weeks.

95% Probability-to-Win

Bayesian threshold plus a credible-interval lower bound that must exceed zero — the effect has to be positive, not just probable.

Every 8th Round = Regression Check

Re-runs the previous champion against the current one. False promotions from novelty or seasonal blips get caught and reverted.

Auto-Pause After 3 Inconclusive Rounds

If three rounds in a row fail to find a winner, the loop pauses itself and emails you — so credits aren’t silently burned chasing a signal that isn’t there.

The result: an always-on optimization that doesn’t trade away defensibility. Every round is a clean 50/50 between exactly two variants, so you get a per-round “X beat Y by Z%, P > 0.95” claim you can put in a report — not the shifting traffic allocation of a multi-armed bandit. Online false-discovery-rate control keeps the loop’s lifetime FDR at or below 10%, even after dozens of rounds. No other SMB A/B testing tool ships always-on AI-authored loops like this.

See how Continuous Loops work →
Core Differentiator

Run A/B Tests From Any AI Agent: MCP, A2A & REST API

VariFlask is the only A/B testing platform built to be driven by AI agents. Our Agent Gateway exposes 8 agent tools over both MCP (Model Context Protocol) and A2A (Agent-to-Agent Protocol), plus a full 17-endpoint REST API — so assistants like Claude, Cursor, and automation platforms like n8n can run the entire CRO workflow without a human clicking through a dashboard.

Connect once with OAuth 2.1 — clients just need the server URL and a one-time browser sign-in, with no secret stored in config — or use a static sk_live_ API key. Then an agent can:

Register sites & pull intelligence

Add a website, get the tracking snippet, and read AI-analyzed industry, competitor, and keyword intelligence.

Generate ideas & variant code

Author variant code from a plain-English prompt — e.g. “change the hero price from $99 to $499 and update the Stripe link” — no pre-existing idea required.

Launch experiments & loops

Create and launch experiments, then start always-on Continuous Loops — entirely programmatically.

Pay-as-you-go credits

Per-call billing with credit packs (Starter $9/100, Growth $39/500, Scale $129/2,000) — unmetered on Enterprise.

Works with Claude Desktop Claude Code Cursor n8n Any MCP / A2A client

Optimizely, VWO, and AB Tasty give you a dashboard and an SDK. None of them expose a native MCP or A2A agent gateway. VariFlask lets your AI stack do the CRO work end to end.

MCP + A2A quickstart → Agent tool reference → REST API quickstart →

Bayesian Statistics in Plain English

No p-values, confidence intervals, or statistical jargon. VariFlask translates complex Bayesian statistics into language anyone can understand.

How results look in VariFlask:

“Variant B is converting 15.2% better. 89% chance it’s the real winner. Still early — need more visitors to be sure.”

Automatic winner detection with configurable confidence thresholds. Designed for marketers and product managers, not data scientists.

Lightweight & Fast

~5KB gzipped

10× smaller than VWO/Optimizely — synchronous load, zero content flicker

Async engagement tracking

Loads after page paint — zero performance impact

Cloudflare edge network

Sub-5ms latency globally

No CWV impact

No impact on Core Web Vitals scores

How VariFlask Compares

Feature VariFlask Optimizely VWO AB Tasty
Site + competitor + industry analysis ✓ ✗ ✗ ✗
One-click test launch (no-code) ✓ ✗ Requires setup Requires setup
Viewport-gated impressions ✓ ✗ ✗ ✗
Per-variant engagement metrics ✓ ✗ ✗ ✗
AI-generated test ideas ✓ ✗ ✗ Personalisation only
Smart goals from AI ✓ ✗ ✗ ✗
Continuous always-on AI-authored loops ✓ Bandit only Bandit only Allocation only
Native MCP + A2A agent gateway ✓ ✗ ✗ ✗
REST API access ✓ ✓ ✓ ✓
Generate variant code from a prompt ✓ ✗ ✗ ✗
Plain-English results ✓ Partial Partial Partial
Snippet size ~5KB ~80KB ~50KB ~40KB
Free trial ✓ ✗ 30-day trial ✗
Bayesian statistics ✓ ✓ ✓ ✓
Pricing Public, from $9/mo $50K+/yr Demo only* Custom quote

* VWO and AB Tasty merged in 2026 under Everstone. VWO no longer publishes prices — its pricing page is now “Schedule a Demo,” and AB Tasty has always been custom-quote. VariFlask keeps pricing public and self-serve. Competitor figures are list/estimate ranges from public procurement data and change frequently; verify on each vendor’s site.

VariFlask vs Optimizely → | VariFlask vs VWO → | VariFlask vs AB Tasty → | Best A/B Testing Tools 2026 →

Frequently Asked Questions

How is viewport tracking different from regular impression tracking?
Traditional A/B testing tools count an "impression" the moment a page loads — even if the visitor never scrolls down to see your change. VariFlask uses the browser's IntersectionObserver API to only count an impression when your modified element actually enters the visitor's visible viewport. This follows IAB/MRC viewable impression standards (the same standards used in digital advertising) and means your conversion rates reflect real human attention, not just page loads.
Will engagement tracking slow down my site?
No. The tracking snippet is ~5KB gzipped — about 10× smaller than VWO, Optimizely, or AB Tasty. The experiment-applying core loads synchronously to prevent content flicker; behavioral tracking (sessions, scroll depth, hover, dwell time) loads asynchronously after the initial page paint via requestIdleCallback, so it has zero impact on your Core Web Vitals or perceived load time. Everything is served from Cloudflare's edge network with sub-5ms latency globally.
How does the AI generate test ideas?
The AI analyzes your live website — its page structure, content hierarchy, calls to action, form layouts, and more — then generates specific, actionable A/B test hypotheses. Each idea includes pre-configured goals with the exact CSS selectors and URL patterns needed to measure success. You choose how often new ideas are generated, and you can launch any idea as a running experiment with one click.
What if my change is above the fold — does viewport tracking matter?
Yes, it still matters. Even above-the-fold elements aren't always visible — users may have different screen sizes, browser chrome heights, or may navigate away before the element renders. Viewport tracking ensures you're only counting visitors who actually had the opportunity to see and interact with your change, regardless of where it sits on the page.
How much traffic do I need for meaningful results?
Because VariFlask uses viewport-gated impressions (filtering out visitors who never saw your change), you typically reach statistical significance faster than with traditional tools. The exact traffic needed depends on your baseline conversion rate and the size of the effect you're testing for. Our Bayesian engine gives you a plain-English probability estimate that updates in real time, so you always know where you stand.
Can I use VariFlask with my existing analytics tools?
Absolutely. VariFlask uses a lightweight tracking snippet that runs alongside your existing analytics stack — Google Analytics, Mixpanel, Amplitude, Hotjar, or anything else. It doesn't interfere with other scripts and focuses specifically on A/B test execution and measurement.
What is a Continuous Loop and how is it different from a multi-armed bandit?
A Continuous Loop is an always-on champion-challenger optimization: AI authors a challenger, the round resolves with a defensible Bayesian winner, the winner becomes the next champion, and the loop authors a fresh challenger — round after round, until you pause it. Unlike a multi-armed bandit, which shifts traffic toward whatever looks like it is winning (sacrificing a clean causal claim), every loop round is a clean 50/50 split between exactly two variants. That gives you a defensible "X beat Y by Z%, P > 0.95, credible interval excludes zero" result you can put in a report. Guardrails include a 7-day floor per round, ≥500 samples per arm, a 95% probability-to-win threshold, regression checks every 8th round, online false-discovery-rate control (lifetime FDR ≤ 10%), and auto-pause after three inconclusive rounds. No other SMB A/B testing tool offers AI-authored always-on loops with this level of statistical defensibility.
Can AI agents like Claude, Cursor, or n8n run A/B tests through VariFlask?
Yes — this is unique to VariFlask. Our Agent Gateway exposes 8 agent tools over both MCP (Model Context Protocol) and A2A (Agent-to-Agent Protocol), plus a 17-endpoint REST API. An AI assistant can register a site, pull AI-analyzed site intelligence, generate test ideas, author variant code from a plain-English prompt ("change the hero price from $99 to $499 and update the Stripe link"), launch experiments, and start Continuous Loops — all programmatically. Connection is via OAuth 2.1 (clients like Claude Desktop, Claude Code, and Cursor just need the server URL and a one-time browser sign-in) or a static API key. Usage is pay-as-you-go with credit packs, and unlimited on the Enterprise plan.
How much does VariFlask cost?
Plans are Basic at $9/month (1 site, 10 ideas/month, pay-as-you-go Continuous Loops), Advanced at $29/month (10 sites, 50 ideas/month, competitor analysis, 1 Continuous Loop included), and Enterprise at $99/month (25 sites, unlimited ideas, team collaboration, REST API access, and unlimited Continuous Loops). That is a fraction of legacy enterprise tools, and our pricing is public and self-serve. Optimizely runs $50K+/year, while VWO and AB Tasty (which merged in 2026 under Everstone) are now sales-gated — VWO's pricing page is “Schedule a Demo” with no published figures, and AB Tasty is custom-quote only. Agent Gateway credit packs (for MCP/A2A/API usage) are sold separately: Starter $9 for 100 credits, Growth $39 for 500, Scale $129 for 2,000 — and are unmetered on Enterprise.

Start testing smarter, not harder

Join teams who are getting more accurate results with less traffic.

Start Free Trial