---
title: "AI assistant connected to the salesperson's data · Client case · Quentin Andrieux"
description: "An AI assistant that knows each salesperson's plan, targets and portfolio, and only comments on what it can actually see. Anonymised client case."
url: https://quentinandrieux.com/en/client-cases/ai-assistant-sales-plan/
lang: en
author: Quentin Andrieux
updated: 2026-11-11
---

> Version texte de https://quentinandrieux.com/en/client-cases/ai-assistant-sales-plan/ (the HTML page is the reference). Author : Quentin Andrieux, https://quentinandrieux.com/

[Home](https://quentinandrieux.com/en/) [Client cases](https://quentinandrieux.com/en/client-cases/) AI assistant

Events and media group · international sales team

# An AI assistant that knows each salesperson's plan.

Inside an events group's sales cockpit, an assistant connected to the person's own data: their plan, targets, portfolio, and what they have already signed.

The flow

1.  **Question** The salesperson types in their plan's panel.
2.  **Context** Plan, targets, portfolio and signed deals, injected with every question.
3.  **Answer** Based on that data only, with its limits stated.
4.  **Adjusted plan** The salesperson edits the block and reruns the projection.

What the assistant sees, and what it does not.

## The context

The project first aimed at a broad assistant: follow-ups, meeting preparation, voice entry of activity, a shared knowledge base. The CEO refocused the first step on the salespeople's season plan. The assistant followed that refocus: it serves the plan, not the whole activity.

## What was built

- A chat panel inside the cockpit, private to the salesperson, with one thread per season.
- Context injected with every question: the current plan, targets by brand, the account portfolio and what is already signed. The model is not trained on the client's data; it receives it at the moment it answers.
- Typical questions: "is this plan realistic given last season?", "which brand am I furthest from?".
- Stated limits: the assistant reads neither quotes, nor emails, nor meeting notes.
- The same logic in each plan block's projection: a structured output with a confidence level, a rationale and a point to watch.

Assistant · season plan *Fictional example*

Is this plan realistic given last season?

On **Brand A**, you have already signed 72% of the target: the rest holds with two blocks. On **Brand C**, your plan aims at 3 times last season's signed amount with half as many meetings: that is the gap to look at.

Which brand am I furthest from?

**Brand C**: 31% signed. I cannot see your quotes or emails, only your plan and portfolio.

Fictional exchange.

## What was hardened in testing

A first version compared the target with the planned amount without counting what was already signed, and announced a gap to someone who had already passed their target. The calculation was fixed before go-live: the assistant always starts from the signed figures coming up from the ERP.

## What changed

The salespeople named the AI comments as the most useful part of the tool. Another piece of feedback sets the current limit: as long as the assistant does not know the full customer history and event feedback, its recommendations stay at common-sense level. That is the natural next step: connecting emails and history, with the client's agreement on how that data is processed.

## Further reading

- [The full sales cockpit](https://quentinandrieux.com/en/client-cases/sales-cockpit-erp/)
- [From the sales plan to an enriched prospect list](https://quentinandrieux.com/en/client-cases/from-sales-plan-to-prospect-list/)