Skip to content

AI that writes, tests and keeps improving your website.

Avenue Lift proposes new copy and layout for your pages, waits for your approval, tests the variants on real visitors and sends more traffic to the ones that convert. Then it starts the next test from what it learned.

  • One script tag to install
  • Free for 5,000 tracked visitors a month
  • AI changes wait for your approval by default
  • No IP addresses or personal data stored

Watch the traffic follow the evidence

This is a simulation, not a customer result. You decide how well each headline really converts; the allocation is computed by the same code that runs in the product. Drag a slider mid-test and see how long it takes to notice.

Simulation

Three headlines for one page

  • Control

    Book your free roof inspection

    Share of traffic33%

    0 visitors · 0 converted · 33% chance to be best

  • Variant B

    Know what your roof needs before winter

    Share of traffic33%

    0 visitors · 0 converted · 33% chance to be best

  • Variant C

    A 30-minute inspection, no sales pitch

    Share of traffic33%

    0 visitors · 0 converted · 33% chance to be best

Estimated conversion rate from the simulated visitors

  • Control

    50.0% 2.50% to 97.5%. Baseline.

  • Variant B

    50.0% 2.50% to 97.5%. Interval overlaps the baseline.

  • Variant C

    50.0% 2.50% to 97.5%. Interval overlaps the baseline.

0.00%Control 50.0%Bar: 95% credible interval

0 of 15,000 simulated visitors · 0 conversions

Press Run, then drag a slider mid-test and watch the allocation follow.

A simulation, not customer data. You set the true conversion rate of each headline; simulated visitors convert at those rates and the allocation is computed by the same Thompson-sampling code the product runs.

How it works

Three steps, and only the first one needs a developer.

  1. 1

    Install one script tag

    Paste a single tag into your site's head. It loads the tests for that page and nothing else. If it cannot load, or takes too long, visitors get your original page.

    html
    <script src="https://…/s/lift_pk_…"></script>
  2. 2

    Approve what the AI writes

    The AI reads your page and your brand notes, proposes what to test and writes the variants, each with its reasoning. By default nothing reaches a visitor until someone on your team approves it. You can also build variants by hand in the visual editor.

  3. 3

    Let the results steer the traffic

    Each visitor is assigned once and keeps their variant. As conversions come in, traffic shifts toward what works, the losers are retired, and what was learned feeds the next round of ideas.

The AI writes. A person decides.

Variants are written for your page and your brand, not pulled from a template. Each one arrives with the reason it might work, and it waits for your approval unless an admin has turned on auto-approve for that site.

Grounded in your page
The AI works from the content of the page being tested and the brand context you keep per site: voice, audience, the claims you actually make.
It learns from rejections
When a reviewer rejects a variant and says why, that reason is given to the AI the next time it writes for the site.
Numbers come from the stats engine
Result summaries are written by the AI, but every figure in them is checked against the computed statistics.
ExampleReview queue
  • AIHeadlineWaiting for review
    Now
    Book your free roof inspection
    Proposed
    Know what your roof needs before winter
    Why
    The current headline names the offer but not the reason to act now. This one leads with the outcome the page already promises further down.
  • AIButtonWaiting for review
    Now
    Submit
    Proposed
    Get my inspection time
    Why
    The form asks for a preferred date, so the button can say what the visitor gets for filling it in.

AI test ideas: included in every plan. AI-written variants: from the Starter plan.

A bandit, so fewer visitors see the losing version

A fixed split keeps sending half your visitors to the worse page until someone ends the test. Bandit allocation moves traffic while the test runs.

Thompson sampling
Each variant gets traffic in proportion to the probability that it is the best one, recomputed as conversions arrive.
Nothing is starved
Every live variant keeps at least 5% of traffic, so a slow starter can still prove itself and the control stays measurable.
Visitors stay put
Assignment is deterministic and sticky. Reweighting changes what new visitors see; it never flips someone who has already been assigned.
Results you can read
For every variant: conversion rate with a 95% credible interval, the probability it is best, and the expected loss of choosing it. Fixed-split tests are checked for sample-ratio mismatch.
Or keep a fixed split
Allocation is a per-experiment setting. Choose a classic fixed split whenever you want one.

Bandit allocation: from the Starter plan.

Edit the live page by pointing at it

Open your own site with the editor on top, click an element and change it. No staging copy, no deploy, no ticket for a developer.

Declarative changes only
A variant is a list of edits to elements: text, style, attributes, visibility, position. The snippet never runs arbitrary JavaScript from a variant.
Preview before anyone sees it
Open any variant on your real page with a preview link. Previews record no visitors and no conversions.
Fails safe
The page is briefly held back while changes apply, with a hard time limit. If anything goes wrong, the original page shows.
ExampleTextStyleHideInsertMove
main h1

Know what your roof needs before winter

A licensed inspector checks flashing, gutters and shingles and sends you a written report the same day.

Get my inspection time
What gets savedjson
[
  {
    "kind": "text",
    "selector": "main h1",
    "value": "Know what your roof needs before winter"
  },
  {
    "kind": "style",
    "selector": "main .cta",
    "styles": {
      "background-color": "#0b5d6b"
    }
  }
]
An invented page. The change list is the format the snippet applies: declarative edits to elements, never arbitrary script.

One workspace for every client site

Avenue Lift is built for the agency case first: many sites, several people, and clients who want to see results without learning a new tool.

Sites with client labels
Each site has its own goals, experiments, brand context and settings, and a label for the client it belongs to.
Read-only client reports
Share a results page by link. The client needs no account and can change nothing.
Four roles
Viewers read results. Members build and review experiments. Admins manage sites, people and API keys. The owner manages billing.
An audit trail
Who changed what, and when. From the Agency plan.
API and webhooks
Pull results into your own reporting and get signed notifications when a test finds a winner. From the Growth plan.

Measures conversions, not people

The snippet needs to know which variant a browser saw and whether it converted. It was designed to collect that and no more.

Sites that need consent first can switch to opt-in mode: nothing is stored or sent until your consent banner says so. The details are on the privacy and security pages.

What is stored

  • A random visitor ID in first-party storage (_lift_vid)
  • Which variant the visitor was shown
  • Conversions on the goals you define
  • The page URL, without its fragment
  • Device type and country
  • Referrer host and UTM source, medium and campaign

What is never collected

  • IP addresses: not read, stored or logged
  • Names, emails or form field values
  • Third-party cookies or cross-site identifiers
  • Keystrokes, mouse trails or session recordings
  • Anything at all from visitors sending Do Not Track or Global Privacy Control, unless you turn that off

Put your first test live this week.

The Free plan covers 1 site and 5,000 tracked visitors a month. No card needed to start.