> For the complete documentation index, see [llms.txt](https://docs.fullsession.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.fullsession.io/lift-ai.md).
# 14. Lift AI
**Lift AI** is FullSession's automated optimization layer. You point it at a funnel and a target metric, and it works continuously in the background — analyzing sessions, finding the issues most likely to be costing you conversions, recommending fixes, and then **measuring whether your fix actually worked**. Where the rest of FullSession helps you investigate, Lift AI helps you prioritize and prove impact.
> Lift AI is available from the **Pro** plan (\[Chapter 17, section 17.1]). On Free and Growth the Lift AI item is locked, with an **Upgrade plan** link to **Settings → Subscription**.
***
### 14.1 What Lift AI Is
Lift AI is built around three connected objects:
| Object | What it is |
| --------- | -------------------------------------------------------------------------------------------------------------------- |
| **Goal** | A monitoring target you create — a specific **funnel** and a **KPI** to improve (e.g. "improve checkout conversion") |
| **Run** | A scheduled analysis of a goal. Lift AI **runs automatically every 8 hours** for each active goal |
| **Issue** | A conversion-blocking problem a run detects — with a severity, a recommendation, and an estimated impact |
The flow is: you **create a goal** → Lift AI **runs** on it automatically → each run surfaces **issues** (recommendations) → you act on an issue → Lift AI **measures the real lift** over a validation window.
#### How it differs from the Funnels AI explanation
FullSession has two AI capabilities that are easy to confuse:
| | **Funnels AI** (\[Chapter 12, section 12.5]) | **Lift AI** (this chapter) |
| --------------- | -------------------------------------------- | ---------------------------------- |
| Trigger | You click "analyze" on a funnel step | Runs automatically on a schedule |
| Scope | One step, one drop-off, right now | A whole goal, tracked continuously |
| Persistence | Ephemeral explanation | Persistent goals, runs, and issues |
| Measures fixes? | No | Yes — predicted *and* actual lift |
In short: Funnels AI is an **on-demand explanation**; Lift AI is an **always-on optimization program**.
***
### 14.2 Creating a Goal
You create a goal in a short setup wizard. The goal ties together a funnel, an audience, and a KPI, plus some context that helps the AI tailor its analysis.
#### Step 1 — Context
| Field | What you provide |
| --------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Name** | A name for the goal (e.g. *"Checkout Conversion"*) |
| **Description** | Optional notes |
| **Industry** | Your industry, from a list (e.g. *Software / SaaS*, *E-commerce / DTC*, *Financial Services*) |
| **Role** | Your team's role(s) — Support, UX & Design, Engineering, Marketing, Customer Experience, Product Management, Data & Analysis, Account Management |
The industry and role help Lift AI frame its recommendations for your context.
#### Step 2 — Funnel selection
Choose the **funnel** the goal will monitor — pick an existing funnel (\[Chapter 12 — Funnels]) or define a new one on the spot.
> **A goal's funnel needs more than three steps.** Lift AI analyzes transitions between steps, so a meaningful funnel is required.
#### Step 3 — Verify (data readiness & KPI)
The final step checks that Lift AI has what it needs and sets your KPI:
* **Installation & data-readiness check** — confirms the tracker is installed and there's enough data for reliable analysis.
* **Baseline segment** — the audience to measure against (defaults to **Everyone**, or pick a segment).
* **KPI** — Conversion Rate or Revenue Per Visitor (section 14.3).
#### What you set vs. what Lift AI computes
| Set by you | Computed automatically |
| ----------------------------------- | ----------------------------------------------- |
| Name, description, industry, role | **Baseline value** (your current KPI) |
| Funnel and baseline segment | **Prediction confidence** (low / medium / high) |
| Validation window (days) | |
| KPI (and RPV config, if applicable) | |
The **baseline** is read from your funnel's current performance, and the **prediction confidence** reflects how much data is available — neither is hand-entered. If there isn't enough data or the required signals are missing, Lift AI won't let you create the goal until that's resolved.
***
### 14.3 KPIs: Conversion Rate & Revenue Per Visitor
A goal targets one of two KPIs:
#### Conversion Rate
The share of visitors who complete the funnel. This is the default and needs no extra configuration — Lift AI reads the baseline from your funnel.
#### Revenue Per Visitor (RPV)
Revenue divided by visitors — useful when you care about money, not just completion. RPV combines your funnel's conversion rate with an **Average Order Value (AOV)**, which you supply one of two ways:
| AOV source | How it works |
| ----------------------- | ------------------------------------------------------------------------------------------------------------------------------ |
| **Enter directly** | Type your average order value |
| **From a custom event** | Point Lift AI at a numeric value on one of your `FUS.event(...)` events (\[Chapter 7, section 7.5]) and it averages it for you |
> **The KPI is fixed once the goal is created.** You can edit most of a goal later, but the KPI and baseline can't be changed — to switch KPI, create a new goal.
***
### 14.4 How Analysis Runs
Once a goal is **active**, you don't trigger anything — Lift AI does the work on a schedule.
#### Scheduled, automatic runs
Lift AI **runs every 8 hours** for each activated goal. Each run:
* Analyzes recent sessions against the goal's funnel.
* Produces a **predicted conversion rate** (with a confidence range) representing where the KPI could land if the detected issues were addressed.
* Calculates a **predicted improvement** — the lift on the table.
* Detects and updates the goal's **issues** (section 14.5).
Paused or archived goals are skipped.
> **There's no "run now" button.** Analysis is driven entirely by the schedule — you can't force an immediate run. New goals will show their first results after the next scheduled run.
***
### 14.5 Reading Recommendations (Issues)
Each goal has a **recommendations** view listing the issues Lift AI has found, prioritized so you can work top-down.
#### Summary stat cards
At the top, counters summarize issues by **severity** — **Severe, High, Medium, Low** — with trends, so you can see your overall problem load at a glance.
#### The recommendations table
| Column | Shows |
| ------------------ | ------------------------------------------------------------------------------------------ |
| **Recommendation** | The issue title (e.g. *"Rage clicks at the shipping step"*) |
| **Status** | Where the issue is in its lifecycle (section 14.6) |
| **Page** | The URL where it occurs |
| **Issue type** | Rage click, dead click, error click, page-load error, uncaught exception, or network error |
| **Severity** | Severe / High / Medium / Low |
| **Funnel steps** | The affected transition (e.g. *2 → 3*) |
| **Baseline** | Current conversion at those steps |
| **Predicted lift** | The expected improvement if fixed (a range, e.g. *15.2% → 16.9%*) |
| **Actual lift** | The measured improvement after a fix (appears once validated) |
| **Last seen** | The most recent run that detected it |
#### Issue detail
Open an issue to see its full detail:
* **Rationale** — the AI's explanation of the **root cause**.
* **Recommendation** — a suggested **fix**.
* **Impact** — the estimated **conversion loss** and the affected-vs-unaffected session breakdown.
* **Sample sessions** — example recordings, so you can **watch the issue happen** (\[Chapter 6 — The Session Player]).
> **Tip** — work the **Severe** and **High** issues with the largest **predicted lift** first. The sample sessions let you confirm the AI's read before you invest in a fix.
***
### 14.6 The Issue Lifecycle & Measuring Real Lift
Lift AI's distinctive value is closing the loop: not just *finding* issues, but **proving whether your fix worked**. That happens through an issue's status lifecycle.
#### The status flow
| Status | Meaning |
| ----------------------- | ---------------------------------------------------- |
| **New** | Just detected |
| **Undecided** | Acknowledged, not yet triaged |
| **Dismissed** | You've rejected it as not worth acting on |
| **Planned** | Scheduled for a fix |
| **Change is live** | You've deployed a fix |
| **Validating** | Lift AI is monitoring the impact |
| **Validation complete** | The validation period has ended and the result is in |
You move an issue through these states as you work it.
#### Predicted lift vs. actual lift
* **Predicted lift** is Lift AI's *estimate* — what the KPI could gain if the issue were resolved. You see it as soon as the issue is detected.
* **Actual lift** is the *measured* result. When you mark an issue **Change is live** and then **Validating**, Lift AI captures the baseline and monitors performance for the goal's **validation window** (a number of days you set). When that window closes, the issue reaches **Validation complete** and the **Actual lift** column shows what really happened.
This before/after measurement is what turns Lift AI from a recommendation engine into an accountable optimization program — every fix gets a verdict.
> **Tip** — set the validation window long enough to gather representative traffic for the affected steps. Too short and the "actual lift" reading will be noisy.
***
### 14.7 Managing Goals & Issues
#### Managing goals
| Action | Notes |
| ----------- | --------------------------------------------------------------------------------------------------------------------------- |
| **Create** | Via the setup wizard (section 14.2) |
| **Edit** | Change name, description, funnel, baseline segment, industry, role, and validation window — **but not the KPI or baseline** |
| **Pause** | Stop scheduled runs without deleting the goal |
| **Archive** | Hide a goal from the active list |
| **Delete** | Remove the goal entirely |
Goals are scoped to the funnels you own.
#### Managing issues
* **Change status** — move issues through the lifecycle (section 14.6); status changes are tracked over time.
* **Filter** — by status, severity, issue type, funnel steps, and date.
* **Sort** — by impact, conversion loss, predicted lift, or detection date.
#### What Lift AI doesn't do
To set expectations clearly, these **don't exist**:
* **No alerts/notifications** when a new issue is found (no email or Slack on detection) — check the recommendations view, or use **Alerts** (\[Chapter 15]) for metric-based notifications.
* **No industry benchmark comparisons.**
* **No auto-generated fix code** — recommendations are written guidance, not patches.
* **No issue clustering**, **A/B-test integration**, **custom alert thresholds**, **leaderboards**, **export/reporting**, or **webhooks**.
* **No manual run trigger** — runs are scheduled only.
> **The big picture** — Lift AI turns a funnel and a KPI (**Conversion Rate** or **RPV**) into a continuously analyzed **goal**. It **runs every 8 hours**, surfaces ranked **issues** with a root-cause **rationale**, a **recommendation**, an estimated **impact**, and **sample sessions** to watch, then measures **actual lift** over a validation window as you move each issue from *new* to *validation complete*. It's automated (no prompts, no manual runs), accountable (predicted vs. actual lift), and distinct from the on-demand Funnels AI explanation.
***
> **Next up:** \[Chapter 15 — Alerts & Notifications] covers proactive, metric-based alerting — being told when a conversion rate drops or errors spike, complementing Lift AI's continuous issue detection.
