FabCon Europe 2026: the announcements that reach your reports

This is the most concrete announcement of the week for our clients. Today, feeding a warehouse from Business Central requires an extraction layer: queries, scheduling, a night-time window, and someone checking in the morning that everything went through. Mirroring replaces that layer with a setting.

Two points keep all their importance. The first: what arrives in OneLake is the ERP's tables, with their internal data model. Fields calculated on the fly, FlowFields in particular, are rebuilt as measures rather than replicated as values. The second: replicating is not modelling. Raw tables placed in a lake remain raw tables, and the work of giving them meaning starts there.

The point worth reading twice: these answers rely on the same semantic model as your reports, with the same measures and the same definitions. The sales director who questions Copilot from a Teams conversation and the management controller who opens the report see the same revenue. That is exactly what we call a single version of the truth, and Microsoft has just made it the entry point of its assistant.

The consequence is plain: the quality of your semantic model becomes visible to the whole company, and no longer only to those who open the reports. Well-named measures, clean relationships and certified definitions produce good answers. An approximate model produces approximate answers, at scale.

Governance follows, which is welcome: sensitivity labels and DLP policies propagate into the Copilot experiences, and a dedicated setting lets the administrator decide which semantic models take part in AI experiences.

Writing back is a constant request. A margin report that displays the variances is useful; the same screen where the management controller enters their comment and their reforecast assumption becomes a working tool. Until now, this required a Power App connected alongside, or a business portal developed for the occasion.

We have been building business portals for several years, and this announcement leaves them their full place: an app generated from a semantic model covers input tied to analysis, whereas an approval workflow connected to an ERP remains a development in its own right. It does, however, cover very well the area where we used to see working Excel files circulating alongside the reports.

This lifts the objection we hear most often at the start of a project: reserved capacity is paid for from day one, whereas the value arrives in the third month. Being able to start a proof of concept, let it run for a few weeks and pay only for the compute actually consumed changes the way a project is launched, and above all the way it is decided.

Reserved capacity keeps its full meaning once usage is established and regular: it remains the most economical model at stable volume. The benefit is to now have both, and to move from one to the other at the right moment.