An example of what we build with AI automation
An example of what we build: an AI back office for a UK accountancy practice
The client here is an accountancy practice; the approach is not limited to accountancy.
At a glance
Client: a UK accountancy practice
Work: client filing, document processing, accounts desk, client history, approval-first client chasing
Approach: done for you, on the practice's own office network, with the practice approving everything that goes to clients
Messages everywhere, information missing, work done twice
The practice's clients sent paperwork and questions by WhatsApp and email, on paper and by post. Years of correspondence and records sat in mailboxes, older backups and shared folders, not always filed against the right client. The team spent a large part of each week reading messages, working out who they belonged to, chasing clients for what was still missing, and redoing work when information arrived late or incomplete.
What Kutchi built, using Agentic OS
Kutchi's AI agents did the work under Agentic OS, our operating layer in which agents propose and a person approves before anything runs.
- A filing home for every client.
A consistent folder for each client on the practice's own office storage (Inbox, Working, Archive, Scans), created add-only so nothing existing was moved or overwritten.
- Intelligent document processing for the email backlog.
Historic emails and attachments were matched to the right client and filed. Anything the agents could not match with confidence was held in a review list for the practice rather than guessed.
- Older records brought into the same structure.
Files from earlier backups were matched to clients and copied into their folders without overwriting anything, with sensitive items set aside for the practice to decide on.
- An accounts desk on the practice's own network.
A login-protected desk on an office machine with a profile for each client, opening onto their records and accounts work (ledger, VAT, payroll and returns workings) for the practice's accountants to review.
- A history for every client.
A per-client history built from the practice's own records, refreshed nightly, so anyone can see what has come in and what is outstanding.
- Deadline checks from the public register.
Company filing dates checked against public Companies House pages, with upcoming and overdue items flagged to the practice.
- Approval-first client chasing.
A monthly process that works out what each client still owes and prepares one request per client, listing everything missing, as a draft in the practice's own email account. The practice reads it, edits it if needed, and presses Send. The system has no ability to send. WhatsApp reminders are planned only for clients who have given consent.
Human-in-the-loop throughout. Nothing was deleted, nothing was sent to a client and no money was spent without the practice's explicit approval. A qualified accountant at the practice keeps sign-off on all client work.
What's next
WhatsApp handling with approval, the monthly chasing process going live, and AI-powered bookkeeping workflows switched on one at a time.
The same approach works for your business
Accountancy is one example. Any business whose back office runs on messages, documents and follow-ups can work the same way: a Kutchi client drone on your network, Agentic OS agents doing the work, and your people approving what matters.