The whole client portfolio in one system

We state it precisely, because the term gets worn out. Agency mode is three things at once: multiple legal entities in one system, data isolation at database level, and per-client configuration without a separate installation. The agency functions then run on top of that — the consultant's portfolio view, group actions, comparing obligation status per client, consolidation and intercompany. The order is not incidental: functions can be built onto a tenant model that stands, but a tenant model cannot be slipped underneath functions. A system that did not anticipate the mode solves it with a separate database per client, and at that point no screen can repair the fact that the upgrade runs forty times.

No, and that is the point. Clients live in one system with isolation enforced through row-level security in the multi-tenant core — at database level, not through an application filter that depends on every query being written carefully. Who sees what is decided by six-layer authorisation: roles, data scope, approval limits, segregation of duties, maker-checker and delegation. The same holds for AI — the copilot and OCR operate inside the user's permissions and the client's isolation, so there is no clever question that routes around the mode. The practical consequence for a firm is one upgrade for the whole portfolio instead of forty, and one application of a regulatory change instead of forty.

No, and that is deliberate. From a scanned or received invoice the system extracts fields, assembles a posting proposal, detects duplicates and proposes the link to the goods receipt — but the result is a draft a person confirms. What separates it from ordinary OCR is that confidence is measured per FIELD TYPE rather than as one number for the document: the amount, the date and the tax number do not share a threshold, and an uncertain field stays empty instead of being filled with a probable value. Approval and rejection both stay in the AI trace, so six months later you can see not only what was posted, but what the system proposed and who accepted it.

Through the channels they already use. Incoming invoices arrive straight into the system over the SEF channel — no scanning, no retyping; a document that is already structured goes to review rather than through OCR. Whatever comes on paper or by email goes through the scanner and an OCR proposal. For data in bulk — opening balances, bank statements, code lists, exports out of someone else's system — there is the import/export framework with a template builder, column-to-attribute mapping, a dry-run diff and per-row errors. The dry run matters more than it looks: the import is visible in full before it executes, so bulk entry is not a leap in the dark.

Yes — that is the purpose of agency mode. The compliance calendar holds filing deadlines, regulatory changes and obligation status per legal entity in one place, and the status is exactly as precise as the channel's answer: prepared means prepared, filed means filed, and accepted means accepted only once SEF or the portal confirms it. The system spells that distinction out instead of rounding it up into a green cell. A status that claims more than the system knows is worse than an empty field — a firm answers for someone else's books, so false certainty costs it more than the unknown.

Client by client, but configured once. The chart of accounts, posting groups and rules transfer as versioned configuration applied across the whole portfolio, with the differences held as per-client settings rather than as separate installations. Opening balances, partners, code lists and open items come in through the import framework: template, column mapping, dry-run diff and per-row errors, so every migration is visible before it executes. Firms typically start with a handful of clients through one full monthly cycle — from receiving the paperwork to the VAT return — and move the rest of the portfolio only after that, because by then it is proven on their own VAT rather than on demo data.