AI by sector · Fiduciaries
What can AI do in a fiduciary firm that keeps its accounting software?
At quarter-end in a fiduciary firm (an accounting and tax practice), the documents for each client mandate arrive by email, by scan and sometimes as a photo taken on a phone. The VAT returns all fall due together. Every year-end begins with chasing missing documents. Those hours of collecting and sorting can go to a supervised AI agent.
In brief
What does AI do in a fiduciary firm?
Djtal, the operational AI specialist for Swiss companies, deploys a supervised AI agent for a fiduciary firm's repetitive work (client documents, bank reconciliation, VAT returns, year-end closings). It works around your accounting software (Abacus, WinBiz, Crésus), the system of record. Human approval points are set in writing at scoping, confidentiality and the Swiss Federal Act on Data Protection (FADP) first.
Use cases
Five workflows, and a clear split between the agent and your staff.
| Workflow | What the agent prepares | What your staff keep |
|---|---|---|
| Accounting documents for client mandates (email, scan, photo) | Requests missing documents, files them by mandate, reads each one and prepares the allocation | Approves allocations outside the agreed framework, settles the ambiguous cases |
| Bank reconciliation | Imports the statement (CAMT.053 format), matches the entries one by one and lists the discrepancies | Settles the discrepancies, approves the corrections |
| VAT returns | Prepares the quarter's return and checks that the VAT codes are consistent across the period | Reviews, signs and files |
| Chasing late payers for client mandates | Spots the due dates and drafts the reminder in the tone agreed with each client | Approves each reminder before it goes, deals with sensitive cases |
| Year-end preparation | Lists the missing documents, runs the consistency checks and prepares the closing entries | Finalises the accounts and advises the client |
The timing helps here. The Swiss financial centre is migrating to new payment formats (pain.001 from 30 September 2026 and SIC 5.3 on 13 November 2026, according to SIX), and eBill has passed 4 million users (SIX, August 2026). A fiduciary firm whose incoming flows are already read by an agent takes these changes in its stride without hiring. The dates are set out in e-invoicing in Switzerland, what changes in 2026.
Your software remains
The agent adapts to Abacus, WinBiz and Crésus.
A fiduciary firm has years of charts of accounts, settings and habits built into its software. Its staff know it inside out, and its mandates live there. Replacing it just to ‘do AI’ would destroy more value than the AI could add. So the agent works around it. It reads the mailbox, the documents and the bank statements, uses imports and APIs where your vendor offers them, and prepares the work directly for your staff where no interface exists.
Document recognition is improving at most vendors, so we switch on what your licence includes before adding a dedicated agent. Zia at Zoho and the AI built into ERPs get the same treatment.
The method
Four steps, and a metric set before we begin.
- 1.
Map the real situation
Half a day with your accountants to see where the work piles up, which client data they touch (confidentiality and the FADP first) and which interfaces your software offers.
- 2.
Pilot one pain point
Document collection or bank reconciliation, on two or three mandates. The success metric is defined before the first line of configuration.
- 3.
Measure before and after
Processing time per mandate, share of cases passed to a person, missing documents detected earlier. Your own mandates supply the figures for the next decision.
- 4.
Extend, workflow by workflow
Each new workflow inherits the framework: a written scope, human approval at the steps that commit the firm, an action log. Our embedded engineers hand autonomy over to your teams at the end of the engagement.
Since May 2026, Émile, Djtal's finance agent, has been importing our own CAMT.053 bank statements and matching the entries one by one. When a supplier falls outside the auto-approved list, processing stops and waits for approval. The split of work described on this page is the one our finance agent applies every week.
Frequently asked questions
AI in a fiduciary firm, your questions answered.
Do we need to replace Abacus, WinBiz or Crésus to use AI?
No. Djtal's AI agent works around your accounting software. It reads the documents, the mailbox and the bank statements, then prepares the entries through the interfaces available (imports, exports, APIs where your vendor offers them, or work prepared for your staff). Your software remains the system of record, and your years of configuration keep their value.
How does the FADP apply to AI handling fiduciary client mandates?
Confidentiality applies twice over in a fiduciary firm. The Swiss Federal Act on Data Protection (FADP) covers your own data and that of your clients, for which you are contractually responsible. Djtal maps, mandate by mandate, what the agent reads and where that data is processed and hosted, and limits the scope to what the processing needs. These choices are documented in writing and can be shown to a client who asks.
What does the agent do alone, and what does a member of staff approve?
The agent collects, files, prepares allocations, matches entries and drafts letters. Anything that commits the firm (an entry outside the agreed framework, a filed return, a reminder sent to the client of a mandate) goes through human approval. The autonomy dial is then turned up, in writing, category by category, on measured cases.
Will AI replace fiduciaries in Switzerland?
AI shifts where the value lies in a fiduciary firm. Data entry and manual matching shrink, while review and advice take their place. A firm that processes its documents faster spends the hours it gains on what its clients already ask for (dashboards, forecasts, support through filing deadlines). Your own staff, equipped with these tools, lead that move up-market.
Where should we start, and how long does it take?
Djtal starts with a pilot on a single pain point, often document collection or bank reconciliation, with a metric defined before work begins (processing time, share of cases passed to a person). A pilot is measured in weeks, on two or three mandates. Extension to other workflows and other mandates is decided on that result.
To frame the project as a whole, the AI strategy audit draws up the map of workflows, the agent's permissions and the pilot plan.
Getting started
Start with one client mandate's quarter-end.
Bring the file of one client mandate. Together we pick the workflow where an agent saves your staff the most hours, and look at what the confidentiality of your mandates means for the data. You decide afterwards whether to go further.
Last updated: