AI Assessment · Restaurant Groups

Where do your restaurants stand with AI?

12 questions to find out whether your AI executes, understands or acts. And what to do on Monday to reach the next level.

Take the test
2 minto take the test
12questions, 6 dimensions
3action plans
Question 1 / 12Answer for the whole group

The framework

What you give AI decides what it can do

From one level to the next, the AI doesn't change. What changes is what you let it see.

Level 1 · Executes

One instruction. One deliverable.

You paste a review, it drafts a reply.

What you give it
An instruction (the “prompt”).
Example
Pasting a Google review into ChatGPT to get a reply.
Its limit
The volume you can copy by hand. And no guardrails on sensitive reviews.
Level 2 · Understands

Your data. Your rules.

You upload an export, it analyzes it.

What you give it
Your exports (reviews, sales) and your business rules.
Example
Uploading 3 months of reviews and asking for the pain points at each location.
Its limit
The time spent exporting and cleaning. A frozen snapshot, out of date by tomorrow.
Level 3 · Acts

Connected. Proactive.

It fetches the data on its own and tells you what to prioritize.

What you give it
Ongoing access to your tools (connectors, MCP, APIs).
Example
Every Monday, a report telling you which location needs you, and why.
Its limit
The quality of your data and your guardrails.
Why this assessment

Everyone uses AI. Very few get results from it.

The gap isn't the tool: ChatGPT and Claude are already very good. It's what the AI can see when it answers you.

68.8%of restaurant and hotel professionals use generative AI at work
9%see a significant impact on their business
38%cite “no one in-house knows how” as a barrier, up from 17% a year earlier.

Sources: Food Hotel Tech Barometer 2025 & 2026 (781 professionals, France) · QU, 2026 State of Digital & Beyond.

Whatever your level

The 3 questions we always get

Is it risky to put company data into AI?+
  • Use business accounts (Team / Enterprise): by default, your conversations aren't used to train the models.
  • No personal data about guests or employees without a legal basis under data privacy rules: anonymize it.
  • Write one simple rule: what can be shared, and what never is.
It makes up numbers. How do you handle hallucinations?+
  • AI mostly hallucinates when it doesn't have the data: give it the source instead of asking it to guess.
  • Ask it to cite where each number comes from, and to say “I don't know”.
  • Check any number behind a decision before the meeting.
An agent that replies on its own, or still copy-paste?+
  • Both exist: copy-paste is level 1; the agent is level 3.
  • To reach a tool, it needs a connector (native, MCP or API): it's a question of access, not intelligence.
  • Automate the routine first (5-star reviews), never the sensitive.