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

# Coach

> What your finished conversations show, and the improvements Coach proposes from them.

Coach reads conversations that have ended, grades them, and looks for the same weakness happening repeatedly. When a pattern is solid enough to act on, it proposes a specific change — with quotes from your own leads as the argument.

Open it from a campaign: **Campaigns → your campaign → Coach**.

## The three modes

Set the mode per campaign. It applies to that campaign only.

<AccordionGroup>
  <Accordion title="Off — the default">
    Conversations are still graded and weaknesses are still tracked. Nothing is proposed and nothing changes.

    This is worth leaving on: measuring costs you nothing, and switching Coach on later starts with everything it has already learned rather than from zero.
  </Accordion>

  <Accordion title="Suggest changes">
    Proposals are generated and queued for you. Nothing changes until you press **Try it** on one.
  </Accordion>

  <Accordion title="Apply automatically">
    Proposals apply themselves, as experiments against your original wording.

    This requires the **Operate** [permission level](/autopilot/trust-levels), which has to be earned. Coach cannot be put on automatic by a new account.
  </Accordion>
</AccordionGroup>

## What Coach shows you

At the top of the page: how many conversations were graded, what share of your questions got answered, how many funnels ran to completion, and the average number of messages a conversation took.

Below that, **the questions your leads skip most** — the single most useful number Coach produces, because a question nobody answers is usually a question worded badly rather than a lead who was not interested.

## Kinds of proposal

Coach proposes at different levels of seriousness, and it tells you which kind you are looking at:

| Kind                 | What it changes                                                            |
| -------------------- | -------------------------------------------------------------------------- |
| **Rewording**        | The wording of one step. The most common, and the safest.                  |
| **Added detail**     | A fact added to your context, drawn from a sentence in your own documents. |
| **What you collect** | Whether a question is required to continue, or merely asked.               |

Every proposal shows **Yours** and **Suggested** side by side, along with the number of conversations behind it and quotes from real exchanges.

### Added detail always shows its source

When Coach proposes adding a fact, it also shows the sentence in your own documents that the fact came from.

This is the reason you can accept a factual addition without re-verifying it: you can see it is your own sentence. Coach never invents a fact about your business.

### Structural proposals explain themselves

When Coach suggests changing *what* you collect rather than how you word it, it first tells you that rewording was already tried and did not help.

Without that, the suggestion reads as Coach simply deciding you should ask for less. With it, it is an argument: we tried the cheap fix, here is what happened, here is the next option.

<Note>
  Coach never edits your context documents. A change to a document cannot be tested — both versions would read the same to a lead — so for those proposals you make the edit yourself and press **I've added this**.
</Note>

## How a change is tested

Applying a proposal does not replace your wording. It starts an experiment:

<Steps>
  <Step title="Both versions go live">
    Leads see one or the other at random.
  </Step>

  <Step title="Coach compares them">
    On the same measurements it used to find the weakness.
  </Step>

  <Step title="The better one is kept">
    If the new wording wins it is promoted. If it does worse, your original comes back. If the result is unclear, nothing changes.
  </Step>
</Steps>

At most **3 experiments run per campaign per week**. That limit exists so you still recognise your own campaign — a rewrite you never read is much harder to spot when five of them are running at once.

## When Coach has nothing to say

The page distinguishes between several quiet states rather than showing one blank screen, because they mean different things:

* **Switched off for this campaign** — measuring, not proposing.
* **No conversations have finished yet** — Coach only learns from conversations that ended.
* **Watching, but hasn't seen enough** — a weakness is being tracked but has not recurred often enough to act on.
* **Nothing worth changing right now** — enough conversations graded, no pattern found. This is a good result, not an error.

## Dismissing a proposal

Press **No thanks** and tell Coach why. It will not suggest that change again. The reason is used to avoid similar proposals, so a sentence is worth more than a blank dismissal.