Copy as markdown[View .md](https://docs.funnelfizz.com/concepts/reports "View the raw markdown for this page")[Open in Claude](https://claude.ai/new?q=Read%20https%3A%2F%2Fdocs.funnelfizz.com%2Fconcepts%2Freports.md%20and%20help%20me%20with%20this%20FunnelFizz%20topic%3A%20AI%20Reports "Open this page in Claude with context")[Open in ChatGPT](https://chat.openai.com/?q=Read%20https%3A%2F%2Fdocs.funnelfizz.com%2Fconcepts%2Freports.md%20and%20help%20me%20with%20this%20FunnelFizz%20topic%3A%20AI%20Reports "Open this page in ChatGPT with context")

# AI Reports

A FunnelFizz report is a one-click diagnostic of one funnel over a chosen period. You get a health score, an executive summary, what's working, where the gaps are, what to fix this week, and a few experiments to run. Same analysis your agent would do over [MCP](https://docs.funnelfizz.com/concepts/insights.md), captured as a saved artifact you can revisit, regenerate, and export.

## What's in a report[​](#whats-in-a-report "Direct link to What's in a report")

Every report has the same shape so you build a habit of where to look:

* **Headline + score (0–100).** Honest, no grade inflation. Score caps at **50** if Stripe isn't connected, and at **30** if tracking isn't installed, because we can only see so much without those pieces.
* **Executive summary.** Three sentences, what's the funnel doing right now.
* **What's working.** Two or three wins to protect, before we tell you what's broken.
* **Findings.** Each one tagged with a severity (`CRITICAL` / `WARNING` / `INSIGHT`) and a stage focus (Awareness / Consideration / Trial / Customer / Cross-stage / Setup). Includes a recommendation and expected impact.
* **Quick wins.** Checklist of "do this in minutes / hours / a day" actions.
* **Experiments.** Hypotheses to A/B test next, with a clear measure.

Reports render as Markdown so they read well in the dashboard and paste cleanly into Notion, Linear, Loom docs, or another model.

## How generation works[​](#how-generation-works "Direct link to How generation works")

1. Open `/reports` from the sidebar (or the **Generate report** button on a funnel).
2. Pick a funnel and a period: **Last 7 days**, **Last 30 days** (recommended), or **Last 90 days**.
3. Click **Generate**. A background worker reads your stages, splits, extensions, customers, and email/automation rollups, then prompts the model for a structured report.
4. Wait \~30–60 seconds. The list updates live; you don't have to keep the tab open.

When it's done, the report is saved to your workspace. You can copy the Markdown, download it as `.md`, or click **Regenerate** to run a fresh pass.

## Plan & quota[​](#plan--quota "Direct link to Plan & quota")

| Plan  | AI reports / month |
| ----- | ------------------ |
| FREE  | 0                  |
| HOBBY | 3                  |
| PRO   | 20                 |

Quota resets on the 1st of each month, UTC. The counter ticks at enqueue time so a click charges quota even if the worker hits a transient error, but transient failures auto-refund so you don't lose budget to flaky retries.

If you regenerate the same period and want to avoid burning quota, edit the original report's experiments instead, the score and findings rarely shift inside a 24-hour window.

## What we send to the model[​](#what-we-send-to-the-model "Direct link to What we send to the model")

Reports are aggregate-only. The context we send to Anthropic Claude is built deterministically from your funnel state:

* Stage counts, conversion rates, and split metrics (cached from the dashboard).
* Customer rollup (active MRR, trialing/active counts, churn rate, top product by volume).
* Email rollup (deliveries, opens, clicks, bounces, top campaign by open rate).
* Setup state (tracking verified? Stripe connected? sender domain verified?).
* The funnel's own name and description, sanitized and wrapped in an `<untrusted>` envelope.

We **don't** send raw profile PII (no emails, names, IPs, Stripe customer IDs), email body content, payment-method details, or per-event data. Anthropic does not train on FunnelFizz traffic; see [our privacy policy](https://funnelfizz.com/privacy) for the inference-only contract.

## Sharing & export[​](#sharing--export "Direct link to Sharing & export")

* **Copy** copies the full Markdown to clipboard.
* **Download** streams the same Markdown as `report.md`.
* **MCP** exposes `list_funnel_reports` and `get_funnel_report` so your agent can pull a saved report into another conversation.

Public read links (one URL you can forward to a co-founder without an account) are coming. Today, exporting Markdown and pasting into Notion / Linear is the easiest share path.

## Reports vs. MCP[​](#reports-vs-mcp "Direct link to Reports vs. MCP")

Use whichever fits the moment, they share the same underlying model.

|                | Reports                                               | MCP                                              |
| -------------- | ----------------------------------------------------- | ------------------------------------------------ |
| Where it lives | Dashboard, saved                                      | Your agent's chat                                |
| Best for       | Weekly diagnostics, artifacts you'll reread           | Ad-hoc questions, comparisons, one-off pulls     |
| Output         | Structured Markdown with score, findings, experiments | Whatever the agent decides to summarize          |
| Quota          | Per-month per-workspace                               | Per-API-key monthly cap (see Developer settings) |
| Shareable      | Yes, as Markdown                                      | No, ephemeral to the chat                        |

If you're already in an agent session, ask it to **"generate a 30-day report on funnel X"**, the `generate_funnel_report` tool runs the same pipeline and saves the result to your dashboard.

## When the score is low[​](#when-the-score-is-low "Direct link to When the score is low")

Reports are honest by design. If your score comes back below **30**, the most common cause is that tracking isn't installed yet, the model can only see Stripe and provider data and is rightly suspicious of the missing middle. Below **50** usually means Stripe isn't connected, so TRIAL/CUSTOMER counts are zero. Connect those and regenerate.

If your score is below 50 with both connected, the report's findings will tell you exactly which stage is leaking and where to look first.

***

**Related:** [Insights & MCP](https://docs.funnelfizz.com/concepts/insights.md) · [Stages](https://docs.funnelfizz.com/concepts/stages.md) · [Tutorials → Your first AI report](https://docs.funnelfizz.com/tutorials/your-first-ai-report.md)