Talking to your restaurant: gimmick or new interface?

Talking to your restaurant is progress only if the conversation replaces real research work and cites its sources. Asking a chatbot for the weather changes nothing. Asking "why has this issue kept coming back at the second location for three weeks?" can change a decision, as long as the answer is short, based on dated facts and honest about what is missing.

Tuesday, 7:40 a.m., in the car

The owner of four restaurants is waiting for the light to turn green. Three questions keep running through his head: did opening go well at the third location, has the manager responded to Friday's incident, is there anything urgent before his lunch visit. To get answers today, he will have to call two people, open two apps and reread messages. He will probably do it at 11, after the point when he could have acted.

What this scene calls for is not one more screen. It is an answer in a sentence or two, with enough behind it to check.

Why so many "conversations" are pointless

The question had nothing to look up

If the answer is already on your phone, the conversation is a detour. An assistant that recites a number visible on a screen saves a click, not time.

The answer has no footing

A well-turned sentence with no origin is dangerous in operations. You cannot tell whether "everything went fine" means "nothing was reported" or "the checks were done and passed". The second is a fact, the first is an absence.

The tool makes things up to fill gaps

An assistant that fills in the blanks feels useful until the day you act on data that never existed. For a restaurant operator, "I don't have that information" beats a plausible answer.

It talks, but knows nothing about your business

Without a memory of your standards, your roles and your history, the conversation stays generic. It answers as it would for anyone, which means not for you.

Gimmick or interface: a sorting table

Criterion	Gimmick	Real interface
What it saves	A click	A search or a consolidation across several sources
Shape of the answer	Long, fluent	Short, with the facts behind it
Missing data	Filled in	Flagged as missing
Provenance	Invisible	Measured, reported, proven, observed, inferred
Requested action	Executed or ignored	Prepared, summarized, confirmed by a person
Business context	Generic	Standards, roles, the group's history

A fifteen-minute test

1. Write down your five recurring questions over a week, on your phone, at the moment you ask yourself.
2. For each one, note where you go for the answer today (a call, an app, a binder) and how many people you interrupt.
3. Rephrase them in plain language, as you would to a deputy. For example: "What have my managers not dealt with?"
4. Ask them to the tool and check three things: the length of the answer, whether it shows the facts behind it, and how it handles missing data.
5. Try to trip it up. Ask for something you know does not exist. If it invents an answer, stop the test.
6. Finish with an action. Ask it to prepare a task. It should show the expected effect and wait for your confirmation.

Common mistakes

• Judging the tool on eloquence. A clear but wrong answer costs more than a hesitant but correct one.
• Asking questions out of curiosity. The test should cover questions that come before a decision.
• Forgetting to record things. A conversation cannot draw facts from a restaurant that records none: if the checks are not done, the answer will say so, and that is useful.
• Asking the tool to judge a person. It connects facts; it does not rank or score. See why AI should not judge your employees.

What to measure

• The number of calls and chase-up messages you make before 11 a.m., over one week, before and after.
• The share of answers that show their evidence, out of ten questions asked.
• The delay between a problem appearing and the moment you see it, tracked on three real cases.

Where Tsuno comes in

Tsuno is built for this kind of interface. You ask in plain language, from Tsuno or from your own AI (ChatGPT, Claude, Mistral). For example: "What deserves my attention today?" or "How did the week go at location X?" The answer is short and based on the group's memory (standards, observed facts, actions taken, all dated), with the facts behind it. When data is missing, Tsuno says so. If it proposes an action, it summarizes the effect, you confirm, and it is recorded. It never judges a person. The details are on the features page.

Going further

Monday morning is a good testing ground: see Monday at 8 a.m.: what a multi-site operator really needs to see and how to know which location needs you tomorrow. For managers, read managers do not need another dashboard. And for the bigger picture, the pillar article do restaurants really need another software tool?.

Key takeaway

A conversation is an interface when it saves a search, cites its sources and admits its gaps. Test it with your own questions, including the ones it should not be able to answer.