# AI for accountants and fiduciaries: confidential use cases

> AI for accountants and fiduciary firms: the use cases that pay off first, what confidentiality rules require, whether ChatGPT can do accounting, and how to start with one workflow.

Language: en · Category: industry · Author: Nicolas Corod (Founder and chief executive officer) · Published: 2026-11-17

## How can accountants use AI without breaking client confidentiality?

Accountants can use AI safely when three things are decided before the first prompt: which client data may go into which tool, which tasks the AI prepares and which a qualified person signs off, and where the firm's own working methods live. AI for accountants is less a question of the best tool than of those three decisions.

Most guides on AI for accountants list tools: receipt capture, bank reconciliation, cash flow forecasts. Those tools are real and useful. What the lists leave out is the part a fiduciary is actually paid for, the judgement and the confidentiality. A Swiss survey makes the gap visible: according to the swissVR Monitor, [cited by PME in October 2026](https://www.pme.ch/invest/ia-en-entreprise-deloitte-decrypte-le-role-du-cfo/4ynpp4t), many companies have set up a governance framework for generative AI, but fewer than one in five have every piece of AI-generated content checked by a person.

## Which AI use cases pay off first in an accounting firm?

The first use cases to pay off are repetitive, document-heavy tasks where the result is easy to check: sorting incoming client emails and documents, preparing first drafts of client letters, reconciling bank statements and checking a file against the firm's own year-end checklist. They save effort without moving the professional decision to a machine.

| Use case | What the AI prepares | What stays with a person | Data sensitivity |
| --- | --- | --- | --- |
| Inbox and document triage | Sorting by client and topic, missing-document requests | Exceptions, client relationship | High |
| Bank reconciliation | Matching, list of open items | Booking decisions | High |
| Year-end file review | Check against the firm's checklist | Sign-off | High |
| Client letters and reminders | First draft in the firm's tone | Final wording, sending | Medium |
| Tax research | Summary of sources and open questions | Interpretation, advice | Low to medium |

A worked example, on sample data: a fictitious fiduciary firm in Lausanne receives about forty client emails a day. A skill connected to the shared mailbox files each email under the right client, lists the documents still missing for the VAT return and drafts the reminder. The assistant prepares; the accountant in charge of the client reads, adjusts and sends.

The value of such a workflow is not only time. Fewer forgotten documents and fewer late returns count as much, and they are easier for a partner to check.

## Can ChatGPT or Claude do accounting?

General assistants such as ChatGPT, Claude and Copilot can draft, summarise, explain a standard and structure a spreadsheet, but they do not replace accounting software or a qualified signature. They become reliable for accounting work when they follow the firm's own method and reach client data through controlled connections, not through files pasted into a personal account.

That last point is the confidentiality risk. In a Qualinsight study of 1,030 professionals in French-speaking Switzerland, [cited by PME in July 2026](https://www.pme.ch/strategie/ia-au-travail-72-des-romands-lutilisent-souvent-en-secret/rl801nw), 27% of people who use AI at work do so without approval from their company or its IT department. In a fiduciary, that is client data leaving through a personal subscription. An [AI policy for employees](/en/blog/ai-policy-for-employees/) that names the allowed tools and the forbidden data closes the most obvious gap.

## What do confidentiality rules require from a fiduciary using AI?

They require that client data goes only to providers the firm has a contract with, under conditions it can explain to a client. In practice: a business plan with no training on your data, a data processing agreement, a clear hosting location, and access limited to the people who work on that client.

The [General Data Protection Regulation (GDPR)](https://eur-lex.europa.eu/eli/reg/2016/679/oj) applies to European clients' personal data, and the revised Swiss Federal Act on Data Protection (FADP) applies in Switzerland. Many firms add a rule of their own: no client identifier in a tool hosted outside Switzerland or the European Union. Where hosting matters, [sovereign AI options in Switzerland](/en/blog/sovereign-ai-switzerland/) explain what can stay local.

Access rights matter as much as hosting. An AI agent that can read the whole client drive can leak more than a junior who only sees their own files. The principle of [least-privilege access for AI agents](/en/blog/least-privilege-ai-agents/) applies: the agent acts with the rights of the person using it, never more.

## Will AI replace accountants?

The evidence points to a change in the work rather than a replacement. AI takes over preparation, matching and first drafts; the accountant keeps judgement, advice and responsibility. Alexandre Buga, head of French-speaking Switzerland at Deloitte, wrote in his finance column in [PME in October 2026](https://www.pme.ch/invest/ia-en-entreprise-deloitte-decrypte-le-role-du-cfo/4ynpp4t): "The real ambition is not to automate as much as possible, but to automate what can be automated without handing responsibility over to a machine."

There is a second reason not to automate everything. Junior accountants learn the job on today's routine files. A firm that hands every reconciliation to an agent saves hours this year and may lack experienced reviewers in five. Choose what to delegate with the team, and keep some tasks human on purpose.

## How do you start with AI in a fiduciary firm?

Start with one workflow, one owner and one measure. Pick a weekly task that the team would gladly hand over, write down how the best person in the firm does it, and agree on a baseline (time, errors or missed deadlines) before you change anything.

1. **Choose the workflow.** Inbox triage or the year-end checklist are good first candidates.
2. **Write the method down.** The partner's checklist becomes a skill, the reusable instructions an assistant follows. The guide to [turning an SOP into an AI skill](/en/blog/sop-to-ai-skill/) shows how.
3. **Connect the data with limits.** One mailbox or one folder, read-only at first.
4. **Measure against the baseline** after four weeks, and decide with the team what to do with the freed capacity.

skilder lets a firm package that method as a role, with its skills, its scoped connectors from an internal, curated registry and its access rights, used from Claude, ChatGPT or Copilot. Each run leaves an execution trace attributed to the user, the audit log can be exported, and hosting is in Switzerland (Infomaniak) or the European Union. To choose your first workflow, a [free 30-minute exchange with an expert](https://koalendar.com/e/meet-with-nicolas-9?utm_source=skilder.ai&utm_medium=blog_ai_for_accountants) helps structure AI adoption around your client files.

To pick the first workflow for your firm, [book a free 30-minute exchange with an expert](https://koalendar.com/e/meet-with-nicolas-9?utm_source=skilder.ai&utm_medium=blog_ai_for_accountants).
