Can a shared protocol for data access replace the IT department a company does not have. Interim findings from projects we ran ourselves.
Because there is nobody to keep them running. A link between two systems is written once and lives exactly until the first update on either side: the export format changes, and a week later somebody notices the numbers no longer match.
In a company with an IT department that is half a day of work. In a company without one it is silence: nobody knows what broke until the gap shows up in money. So the owner picks manual work, which is at least predictable.
More than the owner names in the first conversation. Usually two or three come up: the till, accounting, maybe stock. Walking through the processes adds the bank, payroll, delivery, bookings, supplier ordering and the spreadsheets where everything else is reconciled.
What matters to us is not the count but the number of seams between them. Manual work lives on the seam, and the seam is what breaks on an update.
MCP (Model Context Protocol) describes one way to give a program access to a system data. Instead of a separate link for every pair of systems there is one layer that both systems and agents connect to.
For a company without developers the point is not that this is a new protocol. The point is that it removes maintenance. A connection is described once and reused: the next report, the next agent and the next system all work through the same layer.
That is what we mean by an integration standard: not a product and not a vendor, but an agreement on how systems hand over data.
It pays off where the same reconciliation is done every week and always by hand. That work is counted in hours, and after the connection the hours are visible.
It does not pay off where the process still changes faster than it settles. If the way of working is rewritten every month, an integration freezes last month and becomes one more thing to redo.
One-off jobs are a separate case. Moving data once is cheaper by hand, and that is what we tell the client instead of selling a connection.
Order in the books. If one item is spelled two ways and delivery notes are entered a week late, the protocol changes nothing: it simply brings the same data faster.
This is the most common source of disappointment. A company expects access to data to replace the quality of it, and gets a neatly assembled untruth.
The findings come from projects ITHS ran itself: implementations, integrations and agent launches in Latvia, 150+ projects per ITHS internal records as of 7 October 2026. Client data is not disclosed, only anonymised observations are used. Until the sample is closed we publish no aggregate figures on this topic - this note holds only what repeats from project to project.