Our own software · built and owned by Mind Links

One question, forty branches.

An operations assistant scoped to a branch, a role and a period, so a regional manager stops phoning around to build a picture that already exists.

Sector
Multi-branch retail, clinics, restaurants, rental
Company size
Ten branches upward
Type
Assistant over operational data, scoped by permission
First release
One month

01 — The problem

The data exists. Getting an answer takes two days.

A regional manager wants to know which branches missed target last week and why. The sales figures are in one system, staffing in another, stock in a third, and the why is in the head of a branch manager who will answer the phone this afternoon.

So they call four people, assemble it in a spreadsheet, and by the time the picture is complete the week is over.

Reports exist for the questions somebody anticipated. Every real question is the one nobody built a report for.

02 — The system

Ask in words. The permission model does the rest.

The assistant is scoped before it answers. A branch manager asking about staffing gets their branch. A regional manager gets their region. Nobody sees another branch's payroll because they phrased the question cleverly — scope is enforced at the data layer, not in the prompt.

An answer that shows what it counted
Saudization movement by branch, with the definition attached. Ask for another branch’s payroll and it refuses — scope is applied to the query before it runs, not asked of the model.

Screenshots of Mirsad, a working prototype we wrote for this page. It runs, every figure in it is invented, and we will click through it in a meeting.

It shows the query, not just the number

Every answer comes with what it counted and over which period. A number without its definition starts an argument in the next meeting, and the argument is about the tool instead of the branch.

It says when it does not know

If the answer needs data that is not connected, it says which system it is missing rather than approximating from what it has.

Follow-ups keep the context

“And the month before?” “Just Jeddah?” “Same but excluding the new branch?” That conversation is the actual work; a dashboard cannot have it.

Answers become saved questions

Anything asked three times becomes a pinned question on the branch page. The assistant gradually builds the dashboard nobody could specify in advance.

03 — The Saudi part

Two things every multi-branch operator here is watching

Then the ordinary ones: Sunday to Thursday, Ramadan trading patterns that look like a data error if you have not seen them before, and Arabic branch names that are spelled three ways across three systems.

Built with, and where it runs

The highlighted regions are live in the Kingdom today. Hosting is settled before the software is, because PDPL and your own client contracts settle it — not us.

04 — What we would build first

One month, one system, five questions

In the first month

  • One data source only — usually sales
  • The five questions your regional managers actually ask, from asking them
  • Scope enforced by role and branch
  • Every answer showing its query

Deliberately not yet

  • Staffing, stock and finance systems
  • Free-form questions across everything
  • Alerts and anomaly detection
  • Anything that writes back

One source answered exactly is trusted. Three sources answered approximately is abandoned in a fortnight, and you do not get a second attempt at credibility.

Whose software this is

Ours. Designed, written and owned by Mind Links. No client behind it and no real branch data.

Our engineers have built assistants scoped to organisational hierarchies for large operators. Those systems belong to those operators and nothing from them is here. What carries over is that the hard part is the permission model, not the language.

Talk to us about your version of this Back to work