> ## Documentation Index
> Fetch the complete documentation index at: https://docs.atllasx.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Lift Tests

> Hold back a random share of cancelling members from all contact, then read the lift in dollars against the members you called.

Every recovery report answers "how much came back". A **Lift test** answers the harder question underneath it: **how much of that would have happened anyway?**

A lift test holds back a small, random share of your cancelling members from every call and text. Everyone else is called as usual. When enough members have been through, the **Lift report** compares the two groups. It measures both the same way, straight from your billing records, and states the difference in dollars.

The two groups are chosen after every eligibility check, so the only difference between them is that we called one of them.

<Note>
  This page is about a test you start yourself. hyzl also holds back a small
  share of cancelling members, about 10%, continuously in the background, to
  keep its recovery fee fair. See [What it
  costs](/docs/revenue-recovery/overview#what-it-costs). That standing holdout
  is separate from the tests here.
</Note>

## Where lift tests live

Lift tests have two homes, and the split is deliberate.

* **Reports** is the feature itself. It holds the offer, a running test's progress, and the finished report you download. Open it from **Revenue Recovery, Reports** in the sidebar.
* **Analytics** holds one **Lift** card in the row of Money won back cards. It is always there, and it states the current answer in one line. Before your first test it reads a dash. While a test runs it reads **In progress**. Once a test finishes it shows the multiple, such as **2.5×**. The card links to Reports in every state.

## Start a test

The offer sits on the Revenue Recovery **Reports** page. Click **Start a lift test** to open the dialog. The same dialog opens from **Run a lift test** in the page header, and from the **Lift** card on Analytics.

1. **Choose how much to hold back**: 10%, 25% (the default), or 50%. Holding back more members reaches a reliable answer sooner, because both groups grow at the same time. Holding back less means fewer uncontacted members, but a longer test.
2. **Read the duration line.** It shows its inputs, not just its conclusion: how many members this share needs, and your average weekly cancellations over the last 8 weeks. Divide one by the other and you get the estimate ("about 6 weeks").
3. **Read the terms.** They state how many members go uncontacted, that this is permanent, how the test ends by itself, and what ending it early costs you.
4. Click **Start test**. It begins immediately.

A lift test costs nothing extra. Held-back members are never dialed or texted, so they use no call credits. The members you do call are billed exactly as they are today.

<Note>
  If a share would likely run past the 90-day limit before it reaches the
  members it needs, the dialog warns you. It then points you at 50%. A test that
  runs out of time still shows you everything it measured, but it cannot produce
  a verdict.
</Note>

<Note>
  One test runs at a time, and it covers every recovery workflow on your
  account, on both billing rails. There is no cooldown between tests. The next
  one can start only once the last report is final. That needs enrollment to
  stop, and the 21 days of counting after it to finish.
</Note>

## While the test runs

The Reports page replaces the offer with the test's progress card. It shows four things:

* how many members have been enrolled, out of the target
* the two conditions that stop enrollment, on one line: that target, or a date
* how many members are being held back
* an early look at each of the two groups

That date is the 90-day limit, counted from the day the test started. It is a limit, not a prediction. Enrollment stops earlier when the test reaches its member target first.

The **Lift** card on Analytics reads **In progress**. Its caption carries the same enrollment count and the same date.

Two things to know about the members being held back:

* **They are never contacted, ever.** Not during the test, and not after it ends. A member held back today does not get called next month. This is what makes the comparison honest, and it is the real cost of running a test.
* **Inbound is unaffected.** If a held-back member calls or texts you, they are answered exactly as they would be today. That member is then left out of both groups, because they are no longer an untouched comparison.

Early figures inside the progress card are badged **Early, not yet conclusive**. They move a lot while the groups are small.

## How a test ends

A test stops enrolling new members at whichever comes first:

* it reaches its member target, or
* it hits the 90-day limit.

You can also end a test early at any time, from the **End test early** button inside the progress card on Reports. The confirmation says what you keep and what you give up:

* everything measured so far is kept
* held-back members stay uncontacted
* calling resumes immediately
* there is no verdict

Once enrollment stops, the test spends 21 more days counting outcomes. Each member is counted for 21 days from their own cancellation, not over one shared calendar window. The last member enrolled therefore needs their own 21 days. The card says **Enrollment complete, counting final outcomes** during this phase, and the numbers can only rise.

<Note>
  You can always stop a test, and you can never stop one and claim proof. A
  verdict needs the full member target the test was sized for. This rule is what
  stops a test from being ended on a good week and read as proof.
</Note>

## Read the lift report

When the last member's 21 days close, the report freezes and appears on Reports. It is counted once and never recomputed. It also stays pinned to its own test period. No date-window selector re-scopes it.

The report leads with dollars:

* **The headline** is the extra money the called group brought in, above what the same number of untouched members brought in on their own.
* **The multiple** ("2.4× more members came back") compares the two groups' return rates.
* **Two group bars**, each labeled with how many members came back and what they paid.
* **A per-100 line**: per 100 cancelling members called, how many more came back and how much more came in, against 100 members left alone.
* **A monthly projection**, on its own labeled line. This one is a projection, not measured money: it applies the measured per-member difference to your current cancellation rate. It is never added into any total on the report.
* **An honesty footer**: the exact cancellation dates measured, the 21-day per-member counting, and the date the figures were final. It also confirms that both groups passed identical eligibility checks.

**Download report** saves the same figures as a CSV.

### Proven, or direction only

A finished test carries one of two outcomes:

* **Proven.** The test reached its full member target. The report shows the verdict and the multiple.
* **Direction only.** The test ended early, or it reached the 90-day limit before its target. The card says where it stopped, and where a verdict would have started ("you ended this test at 140 of the 300 members needed"). It shows every figure it measured. These numbers are a direction to act on, not proof.

## After the report

A finished report is permanent. It stays on the Reports page, and there is no way to put it away. The **Lift** card on Analytics carries the verdict. That card leads with the multiple, such as **2.5×**. Under it sits a **Proven** chip, the gap in percentage points, and the test dates. The dollars stay on the full report: a measured total is one test period's figure, and the multiple holds true as your volume moves.

While a later test enrolls, both surfaces switch to it. The **Lift** card counts the new test's members, and the Reports page shows its progress.

A finished test whose calling did not beat the held-back group is stated as plainly as a win. The card reads **None**, with the chip **Calling did not beat the holdout**.

A **Direction only** result also stays. The card keeps every figure the test measured, and only a newer result replaces it.

**Run another test** opens the same start dialog. A newer test takes the card on Reports, but only once it finishes. Your earlier report moves down into **Past tests**.

### Past tests

**Past tests** lists every test you have finished, newest first. The test on the card above is not repeated there.

Each row carries the enrollment dates, the share you held back, and the result: the measured dollars, **No lift**, or **No verdict**.

* **View report** opens that test's frozen report, the same one the card showed.
* **Download** saves its figures as a CSV.

hyzl never recomputes a report. A row you open a year from now reads exactly as it read the week the test finished.