# Field Sales with AI: Shelf Photo Scoring

> The shelf photos reps take are read by the AI model the AI Assistant is set up with: a score, the facings counted, the listed products it cannot see - installed by itself alongside Field Sales and AI Assistant.

- **What it is:** The shelf photos reps take are read by the AI model the AI Assistant is set up with: a score, the facings counted, the listed products it cannot see - installed by itself alongside Field Sales and AI Assistant.
- **For:** Sales
- **Reports for managers:** Analyses: Shelf Photos.
- **Odoo versions:** 16.0, 17.0, 18.0, 19.0, 20.0 - Community and Enterprise
- **Runs on:** Odoo.sh or Odoo you host yourself - not Odoo Online, which accepts no third-party apps
- **Price:** 467 EUR, one-time, on the Odoo Apps Store
- **Technical name:** bambooforge_field_sales_ai
- **Not for:** Act on the score. It is written on the photo. Nothing else reads it, and nothing is changed on the order, the visit or the rep.; Remove the stamp. The stamp burnt into the photo goes with it: the shop's name, the moment, the rep's name and the check-in position.; Choose the photos. Every photo not scored is sent, of any kind. Switching the automatic job on also sends the photos taken before, oldest first.; Try again by itself. A photo marked Not Read waits for someone to press Score with AI.; Keep the model's cost in Odoo. Each photo sent is billed by your provider to your key.
- **Manual:** https://docs.bambooforge.dev/bambooforge_field_sales_ai/
- **Live demo:** https://demo20.bambooforge.dev/odoo
- **Store page:** https://apps.odoo.com/apps/modules/20.0/bambooforge_field_sales_ai
- **Manual last updated:** 2026-11-12

---
# Field Sales with AI: Shelf Photo Scoring

A rep takes a shelf photo at every visit; a supervisor has no time to look at hundreds a day. With
Field Sales and AI Assistant on Odoo 20, this module sends a photo to the model the AI Assistant is
set up with and writes back a shelf score out of 100, the facings it counted, the products it
cannot see, and a sentence on what to fix. It installs by itself when both are there.

This page is the complete manual: what is sent, to whom, what comes back, and who pays.

## Overview

**For whom.** A supervisor who reads shelf photos by the hundred. It needs Field Sales and
Distribution
and AI Assistant, on Odoo 20.

### What it does

- **A score per photo.** A shelf score out of 100, the facings counted, the listed products it
  cannot
  see, and a remark.
- **On demand.** **Score with AI** on a photo, or on a selection from the *Action* menu.
- **In the background.** With *Score Shelf Photos Automatically* on for a company, a job sends up to
  20
  photos of that company every 10 minutes.
- **Not read means not read.** A photo the provider refused, or an answer that is not the expected
  JSON, is marked *Not Read*. Nothing is guessed.
- **Shelf Photos.** *Field Sales > Analysis > Shelf Photos*: sort by score, filter *Score Under 50*,
  *Not
  Scored*, *Not Read*.

### What it does not do

- **Act on the score.** It is written on the photo. Nothing else reads it, and nothing is changed on
  the
  order, the visit or the rep.
- **Remove the stamp.** The stamp burnt into the photo goes with it: the shop's name, the moment,
  the
  rep's name and the check-in position.
- **Choose the photos.** Every photo not scored is sent, of any kind. Switching the automatic job on
  also sends the photos taken before, oldest first.
- **Try again by itself.** A photo marked *Not Read* waits for someone to press **Score with AI**.
- **Keep the model's cost in Odoo.** Each photo sent is billed by your provider to your key.

### What it needs

- **Field Sales and Distribution** and **AI Assistant** on Odoo 20, Community or Enterprise.
- A model that reads images, set in *Settings > General Settings > Integrations*: for OpenAI an API
  key
  and a vision model such as `gpt-4o`, or a local server with a model that reads images. A
  text-only model answers with an error, and the photo is marked *Not Read*.

## Concepts

**Score.** The model's mark of the shelf, out of 100.

**Facings.** How many fronts of the brand's products the model counted.

**Not Read.** The provider refused, or the answer could not be read.

## Install and first run

1. Install both apps: this module installs by itself.
   
   Expected: *Field Sales > Analysis > Shelf Photos* lists the photos as *Not Scored*.
2. In *Settings > General Settings > Integrations*, set the provider, the key and a model that reads
   images, and press **Test the connection**.
   
   Expected: the connection answers.
3. Open a photo and press **Score with AI**.
   
   Expected: the state is *Scored* with a score, or *Not Read* with the reason in *AI Remarks*.
4. To score in the background, tick *Score Shelf Photos Automatically* under *Field Sales >
   Configuration >
   Collections and Returns*.
   
   Expected: the photos of that company are scored by the job.

### Uninstalling

Uninstalling removes the scores from the photos. The photos stay.

## Flows

One thing you do with the add-on.

### Score photos

1. *Field Sales > Analysis > Shelf Photos*: filter *Not Scored*.
2. Open a photo and press **Score with AI**, or select several and use the *Action* menu.
3. Sort by *Shelf Score*; the filter *Score Under 50* shows the worst shelves.
4. **Score with AI** on a photo scored before scores it again, and is billed again.

## How it works

### What is sent and what comes back

- **To whom.** The model the AI Assistant is set up with: OpenAI, or a local server. With OpenAI,
  the
  AI Assistant's setting *Allow provider-side storage* (off by default) decides whether OpenAI keeps
  it.
- **What leaves.** The photo as stamped, the names of the products of the outlet's assortment, and
  the
  names of the month's focus products. The customer's account, its balance and its prices are not
  sent.
- **What comes back.** A JSON answer. A score is held between 0 and 100. Anything else is *Not
  Read*.
- **The job.** *Field Sales: score shelf photos with AI* runs every 10 minutes, for the companies
  that
  ticked the box. Each photo is saved as it comes, so a slow provider loses nothing already done.

### Who reads and writes what

- *Shelf Photos* is on the supervisors' *Analysis* menu.
- Only a supervisor sends a photo to the model.
- The key is in the AI Assistant's settings, for administrators.

### How we make sure it works

Before every release, **8 automated checks** run against this app on Odoo 16, 17, 18, 19 and 20.
Among them:

- A provider error is kept on the photo.
- A shelf read and written back.

The complete list is in the source you receive.

## For managers

**Field Sales > Analysis > Shelf Photos** is the supervisor's list.

- **Shelf Photos** - the score, the facings, what is not seen and the remark, by outlet and by rep.
  *The supervisor decides which shelf to go and see.*

No document prints from this add-on.

## Reference

The complete reference of this app is generated from the code on every release and published with
this manual at https://docs.bambooforge.dev/bambooforge_field_sales_ai/ under *Reference*.

## Troubleshooting and FAQ

- *The model's answer could not be read.* The model did not answer in the expected form. Press
  **Score
  with AI** again, or use a stronger model.
- *not a shelf.* The model did not see a shelf in the photo.
- Any other text in *AI Remarks* on a photo *Not Read* is the provider's own error, for example a
  wrong
  key or a model that cannot read images.

### Questions people ask

**Does anybody check what the model says?** The score is shown, not used. A person reads it.

**Who pays?** You, to your provider, for each photo sent; a local server costs what it costs to run.

**Can I keep it off the internet?** Yes: choose a local server in the AI Assistant.

### Support

Write to [support@bambooforge.dev](mailto:support@bambooforge.dev) with your Odoo series and what the screen said.

## Changelog

### 20.0.1.0.1

- Every field has a help text; a text on the empty list.
- This manual.


## Reference

Generated from the module's own code (version 20.0.1.0.2, series 20.0) on 2026-11-12; every menu, field, setting, scheduled action, group, rule, report and endpoint below is read from the files, not written by hand.

### Menus

- `bambooforge_field_sales.bf_fs_analysis_menu`
  - **Shelf Photos**: opens Shelf Photos

### Models

#### res.company (extends Odoo's res.company)

| Field | Technical name | Type | Help | Flags |
|---|---|---|---|---|
| Score Shelf Photos Automatically | `bf_fs_ai_auto` | Boolean | Every few minutes, photos not yet scored are sent to the AI model. Unticked: only on demand. |   |

#### bf.fs.visit.photo (extends bambooforge\_field\_sales's bf.fs.visit.photo)

| Field | Technical name | Type | Help | Flags |
|---|---|---|---|---|
| AI | `ai_state` | Selection: Not Scored (todo), Scored (done), Not Read (failed) | Not Scored: waiting. Scored: the model answered in the expected form. Not Read: the provider failed or the answer could not be read; nothing is guessed. |   |
| Shelf Score | `ai_score` | Integer | The model's score of the shelf, out of 100. Nothing else in the app acts on it. |   |
| Facings | `ai_facings` | Integer | How many facings of the brand's products the model counted. |   |
| Not Seen | `ai_missing` | Char | The products of the outlet's assortment the model says it cannot see. |   |
| AI Remarks | `ai_remarks` | Char | The model's sentence on what to fix; when the photo could not be read, the reason. |   |
| Scored On | `ai_date` | Datetime | When the model last answered for this photo, or failed to. |   |
| Outlet | `partner_id` | Many2one |   | related to visit\_id.partner\_id |
| Rep | `user_id` | Many2one |   | related to visit\_id.user\_id |
| Date | `date` | Date |   | related to visit\_id.date |

### Scheduled actions

| Scheduled action | Interval | Model | Runs |
|---|---|---|---|
| Field Sales: score shelf photos with AI | every 10 minutes | bf.fs.visit.photo | `model._cron_bf_ai_score()` |

### Server actions and wizards

| Server action | Model | Where | Runs |
|---|---|---|---|
| Score with AI | bf.fs.visit.photo | Action menu of bf.fs.visit.photo | `records.action_bf_ai_rescore()` |
