What was the problem?
An organisation with six locations wanted to know every day how busy it had been and how busy it would get. That information decides how many staff are needed, and whether marketing should act. The figures lived in different systems. Collecting them and comparing them with the weather and holidays took time every morning, so it did not always happen.
How does the workflow work?
Every morning the workflow runs through the same steps, without anyone having to do anything:
- 01At 08:40 the workflow starts by itself.
- 02Visitor numbers, the weather forecast and school holidays are collected. Only counts, no names.
- 03The three sources are merged per location.
- 04Plain statistics produce a forecast for the next five days.
- 05An AI agent explains what stands out and why, based only on the calculated figures.
- 06Every number in the text is checked before anything is sent.
- 07The update is in the inbox before the day starts. A person decides what happens next.
Why does the AI not do the maths itself?
Language models are good at explaining, but not reliable at arithmetic. They can copy a number wrongly or invent one. That is why plain statistics make the forecast, and the AI only writes the explanation. The workflow then checks that every number in that explanation matches the calculation. If something does not match, the update does not go out.
That way you get the best of both: figures that are right, and an explanation in plain language everyone understands. For example, that Saturday will be busier than usual at one location because the autumn holiday starts and rain is expected.
What about privacy?
The workflow only works with counts per location, not with data about people. So no personal data goes to the AI. That keeps the set-up simple and safe. When a workflow does handle personal data, we limit what goes to an AI service and put the agreements in writing.
What does it deliver?
Collecting and comparing the figures no longer takes any time. Every location starts the day with the same picture, and decisions about staffing and campaigns are based on a forecast instead of a hunch. And because the workflow is monitored, it is noticed straight away if a source is unavailable.
In summary
- Every morning a workflow fetches visitor numbers, weather and holidays and makes a five-day forecast.
- Plain statistics do the maths, the AI only explains, and every number is checked before sending.
- Only counts pass through the workflow, no personal data; a person decides.
Frequently asked questions
Yes, the pattern applies widely: fetch data, merge it, calculate, explain and send. Think of revenue per branch, open orders or a weekly marketing overview.
A combination of a workflow tool and code, connected to the existing systems. For each step we choose what is most reliable.
The workflow is monitored. If data is missing, we get an alert and no incomplete update is sent.