Fiduciary firm: making 18,000 invoices a year run smoothly
A fiduciary firm of 40 people spends 2,700 hours a year entering and checking client invoices, a cost of CHF 216,000. By putting the flow in order and using AI only where it helps, it frees up about CHF 151,000 a year for an investment of about CHF 28,000.
The company
A fiduciary firm of about 40 people that processes 18,000 invoices a year for its SME clients, covering bookkeeping and VAT.
The problem and its cost
Invoices arrive by email, scan, supplier portal or a simple phone photo, in very different formats. Each document is opened, read, keyed in by hand, then filed, often by a qualified accountant. Missing documents are chased case by case, coding rules live in the heads of a few people, and quarter-end and year-end peaks overload the team.
The cost can be measured: 18,000 invoices × 9 min ÷ 60 × CHF 80 = CHF 216,000 a year, or 2,700 hours of low-value work. On top of this direct cost come VAT coding errors that need correcting, closing delays that grow longer, and the lack of time for advisory work, which is what clients value most.
The strategy
We decided not to start with a tool. Automating a disorganised flow only speeds up the disorder. So we first measured the actual time spent on a sample of client files, then standardised how documents arrive and how they are coded, and only then introduced automatic invoice reading. Steering came last, once volumes and exceptions were visible.
This order is what secures the gain. Once the flow is in place, manually handled time falls by about 70%, a saving of about CHF 151,000 a year. The investment, about CHF 28,000, covers the analysis, the documentation of rules, the configuration and the support of the team. It pays back in roughly two months. Everything was built on the accounting tools already in place, with no new software.
The execution
Team: We brought together a dedicated contact, our core team (operations, finance, data, AI) and a sector expert with the profile of a fiduciary head of accounting, familiar with Swiss VAT.
- →A single document intake channel per client, with a list of expected documents and automatic reminders when one is missing.
- →A documented coding reference per client (accounts, VAT rates, recurring suppliers), shared and maintained by the team.
- →A step that reads invoices and proposes the booking entry, connected to the existing accounting software, with a validation queue for uncertain cases.
- →A monthly dashboard of volumes, exception rate and average time per invoice.
AI reads the invoice fields and proposes an account and a VAT code, with a confidence level. Only low-confidence cases go to an accountant, whose corrections then feed back into the reference.
Most of the gain comes from a single intake channel, written coding rules and clear responsibilities, with no AI involved. These organisational measures are also what makes the automation reliable.
The result
Expected annual saving: 70% of CHF 216,000, about CHF 151,000. For an investment of about CHF 28,000, payback comes in 2 months.
The gain is 70% and not 100% because atypical documents, tax judgement calls and sample-based controls remain human work.
- Fewer VAT coding errors and fewer corrections at period end.
- Faster closings and more predictable answers to clients.
- Staff refocused on review and advice rather than data entry.
The assumptions
- Volume: 18,000 invoices processed per year.
- Current duration: 9 minutes per invoice, from receipt to filing.
- Loaded hourly cost: CHF 80.
- 70% of the time saved, with the remainder staying as human work.
Scenario type, built from real missions and anonymized.
In the same cocoon
