How do I use AI to find labour inefficiencies across my restaurant sites?
In this article
Labour is the largest controllable cost in any restaurant group. But for most multi-site operators, understanding where it’s leaking – and why – means pulling reports from multiple systems, cross-referencing spreadsheets, and hoping the answer is somewhere in the middle. By the time you’ve found it, the week has moved on.
Tenzo’s MCP (Model Context Protocol) changes that. It connects your operational data – sales, labour, reviews, reservations, inventory, and GM logs – directly to whichever AI model you already use: Claude, ChatGPT, Copilot, or Gemini. No exports, no custom dashboards, no waiting for a developer. You ask a question in plain English and get an answer grounded in your actual data.
Here’s what that looks like in practice.
What to ask your LLM to see labour inefficiencies
Tenzo co-founder and CEO Christian Mouysset recently ran a live demo on a Tenzo account (see below). The prompt he used was straightforward:
“Can you do a labour productivity analysis, taking sales, labour, and reviews into consideration? Let’s look at the last ninety days, make it visual, and make some recommendations.”
No CSV upload. No pre-built report. Just a question, asked the way an ops director would ask it in a Monday morning meeting.
What came back?
Within seconds, the MCP had pulled sales per labour hour, labour cost ratios, and review scores across every site – and cross-referenced them.
The headline finding: the two highest-revenue sites were among the least productive on labour. Group-wide, the business was turning £32 of sales per labour hour at a 34% labour ratio. But the spread told the real story. Shoreditch was running at £44 per labour hour. Kensington at £27. Manchester was spending £81k in labour to produce slightly less revenue than Soho, which was spending £55k.
The analysis didn’t stop at the numbers. It identified the type of problem at each site. Manchester and Birmingham were flagged as execution problems, not demand problems – low productivity and low review scores despite sufficient covers. The recommendation: pull labour back towards the Soho and Shoreditch shape. For Kensington, the output was different – it earns its review scores, so cut conservatively.
It also surfaced something less obvious: the group’s leanest sites were carrying its best review scores. Labour isn’t just a cost line. It’s directly connected to guest experience.
Drilling down into a single site
The analysis didn’t stop at group level. Christian asked it to dive deeper into Manchester specifically.
What came back was a breakdown by role, day part, and day of week. The core finding: textbook overstaffing. Front of house hours were flat at £27 per labour hour across every day part – breakfast, lunch, and dinner – and almost every day of the week. A well-run site flexes labour up on busy periods and back on quiet ones. Manchester wasn’t doing that.
Sunday was the worst offender. It was pulling the second highest hours of the week against the joint lowest sales. The recommendation was specific: rebuild the Sunday rota, flex front of house at dinner peak, and set a productivity floor.
This is the kind of analysis that would typically take an area manager half a day to produce manually. The MCP produced it in the time it takes to read this paragraph.
Why this is different from a standard BI report?
Most labour reports tell you what happened. They show you the number, not the reason, and they certainly don’t tell you what to do next.
What the Tenzo MCP adds is context. GM logs – the end-of-shift reports your managers are already filling in – sit alongside the quantitative data. If the dishwasher broke down, if a section had to close, if two staff called in sick and agency cover was called – that context is in the room when the AI forms its answer. “No other hospitality BI tool can join qualitative GM notes to quantitative labour data in a single query”. It’s a cross-domain join that simply isn’t possible anywhere else.
See how our customers feel about the Tenzo MCP.
Want prompt ideas? Or to understand the benefits more?
What else can you ask it
Labour productivity is one use case. The same connector works across every data source Tenzo holds. Operators are already using it to ask questions like:
- Which of my collab menus actually drove more spend per guest, accounting for how busy we were?
- Is there a pattern between inventory variance and food quality complaints in my reviews?
- On my highest labour cost percentage days, what did my GMs log?
- Which location has the best revenue but the worst review scores and what’s causing it?
These are questions that previously couldn’t be asked because the data lived in separate systems. The MCP makes the join. The AI does the analysis. The operator gets an answer they can act on before the next service.
How does Tenzo work?
Tenzo aggregates data from over 120 integrations – POS, labour, inventory, reviews, reservations – normalises it, and stores it in a unified data warehouse with hospitality context already built in. The MCP sits on top of that warehouse and connects it directly to your AI model of choice.
The key word is context. The MCP doesn’t just pass raw numbers to the AI. It passes hospitality definitions (SPLH means net sales divided by labour hours), your business context, and your GM logs. That’s what allows it to answer questions accurately at hourly granularity across multiple data sources – something that remains impossible with a DIY approach, even if every tool in your stack had its own MCP.
Want to learn more?
If you’re already a Tenzo customer, you can connect it today. If not, get in touch!
Works with your chosen LLM.
