# Work the recommendations

> The living to-do list from your assistant's conversations, split by who can act - with a contact per group and an email the assistant writes for you.

Source: CBX documentation, version 4.0 preview (unreleased). Canonical page: https://docs.configbox.at/docs/4.0-preview/admin-guide/ai-assistant/send-recommendations-to-your-team. Last updated 2026-08-11.

---
> **Audience:** store owners and shop managers · **Scope:** the Recommendations screen — the
> backlog, its statuses, the contacts and the email · **Last reviewed:** 2026-08-11

**AI Assistant → Recommendations** is the living to-do list your assistant's conversations
produce, split by **who can act on it**. It maintains itself: as conversations are analysed, new
evidence either joins a recommendation the list already has or opens a new one. Nothing to
generate, nothing to re-read — what you see is current, and the bar above the tabs says so.

Every recommendation is backed by the conversations that raised it: the visitor's own words,
quoted, with a link into the [journal](https://docs.configbox.at/docs/4.0-preview/admin-guide/ai-assistant/read-the-conversation-journal) while the conversation
exists. A recommendation without evidence cannot exist.

![The Recommendations screen: groups, evidence and payoff](./img/recommendations.jpeg)

Each entry carries the visitor quotes that raised it, how many conversations hit the same wall, and
a **payoff** — how much effect it buys for how much effort.

## The five groups

| Group | Gets | Typically |
|---|---|---|
| **Catalog & marketing** | Product explanation gaps | Whoever writes product content |
| **Product owner** | Requests visitors made that the shop could not fulfil | Whoever decides what the shop does next |
| **Developers** | Likely bugs, each with a ready-made brief | Your developer, agency or in-house team |
| **Software vendor** | Feature proposals — abilities the assistant lacked | Your ConfigBox contact |
| **The whole team** | Frustrated visitors | Usually the owner or customer experience |

Each tab counts what still **needs action**. A group at zero is good news, not an empty screen.

## Working the list

Each recommendation can be **marked done** (say what was done — the note stays on the record) or
**dismissed** (say why — dismissed items stay dismissed; the analysis will not re-file them).
Resolved items collapse into **Done & dismissed** below the list, from where they can be reopened.

Two statuses the system sets on its own:

- **sent to the contact** — the item was covered by an email from this screen.
- **regressed** — the item was marked done, and then *new* conversations hit the same problem.
  The strongest signal on the screen: a fix that did not hold.

## The contacts and the email

Each group has a **name and email address** in Settings → AI Configurator Assistant. Leave a group
empty and nothing breaks — its list stays readable, only the send button is disabled.

1. Press **Write an email about these**. The assistant writes a short email from the group's open
   items — what to do first and why, quoting real visitors.
2. **Read it.** Subject and message are editable; change anything.

   ![The email the assistant wrote, ready to read and edit](./img/recommendation-email.jpeg)

3. Press **Send it**. Exactly what you see goes out, the covered items are marked **sent**, and
   the email itself is kept on record — what was sent to whom, when, covering which items.