The challenge: a spike in documents every quarter, handled by hand

At heart, the business is about verification: making sure its clients never pay more than what was voted at the owners' general meeting or set out in the lease. But before any checking could happen, everything had to be collected and re-keyed.

At the end of every quarter, service charge demands arrive in bulk. Up to four or five documents per site: regular charges, works, one-off calls for funds. They come by email, sometimes only on the managing agent's portal, and two weeks later by post.

For every document, an accountant had to:

spot the right email among meeting notices and day-to-day building messages;
find the site it relates to, with every managing agent using its own layout;
pick out the right amount and the balance still due;
rename and file the document;
fill in two Excel files side by side: the per-site tracker and the payment order sent to the client;
log in to the portals whenever a document was missing.

The result: a team under pressure at every peak, tracking that relied on each person's memory, and volume growing faster than the team. Meanwhile, the review work clients actually rely on took a back seat to data entry.

The end of every quarter was a scramble. The team spent its days hunting for documents and keying them in.

Managing director, property portfolio management company

The solution: AI reads and understands, the accountant checks

No set of rules like "with this managing agent, the amount is at the bottom right" could cover every case: every managing agent has its own software, layout and wording, and they change. Each document has to be read the way an accountant reads it.

What the AI does	Why it takes AI
Retrieves the documents published on managing agents' portals	More than 150 portals, each with its own interface. An AI agent browses like an accountant: it reads each page to work out where to click, and a redesigned interface doesn't throw it off.
Sorts emails: service charge demand, works demand, reconciliation, or irrelevant message	The subject line says nothing. The body and the attachment have to be read.
Reads each document with intelligent OCR and extracts the amount of each demand and the balance due	Standard OCR copies the text without knowing which amount matters. Here there is no fixed template: every managing agent presents amounts its own way, negative balances included.
Distinguishes routine charges from major works	Both are often headed "call for funds". The difference lies in the line items.
Matches each document to the right site	Addresses look alike and names vary. The AI cross-checks unit references, the address and the history.
Flags duplicates and unusual amounts	Received by email and then scanned, the same document becomes two different files: the content has to be compared.

Once the AI has understood a document, it renames it and files it in the document management system, and the tracker and payment order are filled in automatically. On the portals, the agent logs in with the client's credentials, kept in a secure vault, and checks dozens of portals in parallel at every due date.

Trained on the team's history. The AI learned from documents already processed and approved by the accountants: their sites, their managing agents, the way they name and record things. That history was its training ground, just like a new hire trained in-house.

Familiar tools. Everything lands in the document management system and the Excel files the accountants already used. Sanso runs the system day to day.

The accountant approves. When an amount looks off, the accountant checks the meeting minutes and calls the managing agent. The AI prepares, the accountant decides.

The AI does the sorting and the data entry. We keep the decisions: that is exactly the split we wanted.

Managing director, property portfolio management company

What the AI surfaces for the accountants

Major works to verify

Nearly one service charge demand in three concerns major works: facade renovation, roofing, lifts. The AI isolates them so the accountant can check them against the meeting minutes.
Money to recover

Nearly one year-end reconciliation in two is a refund in the client's favour, to be offset against upcoming demands. By hand, these amounts are easy to miss.

The results: more clients, with the same team

The main result: the company can grow. Today, two accountants track around 500 sites and more than 1,200 documents a quarter. Sorting, reading and data entry are done by the AI: a new client adds documents to the system, not hours of data entry for the accountants. Their time goes into checking, and that time now sets how many sites they can track.

Costs that stay flat as volume grows. For each new client, Sanso adds its sites and portal credentials, and the AI adapts to its documents, learning from the first batches the team approves. The accountants receive the processed documents directly.
A company ready to grow. Management can respond to new tenders and accept new portfolios.
The equivalent of one full-time role freed up. The time once spent sorting and re-keying now goes into checking and client follow-up.
Always-current tracking. Each document is processed as soon as it arrives: the accountants know at any time what has been received, recorded and is ready for payment. Quarter after quarter, demands are recorded before their due date.
Every document retrieved. More than 150 portals are checked systematically.
Continuity of service. Tracking lives in the system, not in someone's head: it carries on through holidays and sick leave.

Before, every new client meant more data entry. Today, we can say yes to new portfolios with the same team.

Managing director, property portfolio management company

A full-time role freed up to take on new clients

The challenge: a spike in documents every quarter, handled by hand

The solution: AI reads and understands, the accountant checks

What the AI surfaces for the accountants

Major works to verify

Money to recover

The results: more clients, with the same team

Let's talk about your documents