It’s the second week of the quarter. A client’s file is still missing three purchase invoices and a bank statement, and the reminder sent last week has gone unanswered. Two other clients have emailed questions that were answered in the spring, a colleague is out sick with several files halfway through, and the next VAT deadline is already on the calendar. None of this is accounting in the strict sense. It’s the chasing, sorting, explaining and handing over that surrounds it, and in many firms it takes up the hours that should go to reviewing figures and advising clients. That is the part of the work where accountants and bookkeepers are now turning to AI.
Research from Belgium and abroad shows how far that has already gone. In Belgium, a survey of 345 Belgian SMEs, commissioned by accounting software provider Horus Software, found that a third already use AI in their bookkeeping, for tasks ranging from invoice recognition and fraud detection to rewording emails and sending payment reminders. Accountants themselves see the opportunity. In the ITAA Barometer 2025, completed by 1,500 Belgian accountants and tax advisers, 71 percent said they see AI as a lever for more tailored advice, and 80 percent consider it their role to guide clients through digitalisation. A field study by researchers at Stanford and MIT combined a survey of 277 accountants with data from 79 small and mid-sized firms using AI-supported accounting software. Accountants using AI moved about 8.5 percent of their time from data entry toward communication and quality checks, and firms closed their monthly books 7.5 days faster. AI users also supported, on average, 55% more clients per week.
If you asked an accountant a year or two ago how they used AI, they probably would have said a chat window. They would ask a question about a VAT rule, or ask it to write a draft email, or a summary of a long document. While that is still common, a growing number of firms, from small bookkeeping practices to larger accounting offices and in-house finance teams, have moved a step further. They are setting up assistants for one specific job, giving them their own instructions and sources, connecting them to the mailbox or the files, and letting them prepare work that a human then checks.
This post is about what that looks like in practice. At the end, we explain how an AI agent works in IttesAI.
What counts as an AI agent
An agent is like a chat window, but you set it up once, for one set of tasks. It has its own instructions, such as how a document request should be worded or what a client summary must contain. It has its own reference material, such as procedures and a good example of a finished file. And it has a connection to the systems the team already works in, such as email, CRM, or cloud storage. It can then pull information from these systems, edit things, save things, send things, etc. (Always with your approval first.)
Where firms are using AI agents today
Collecting documents and knowing what’s missing
So much time in bookkeeping goes into gathering what’s needed before any actual work can even start: the receipt that never arrived, the payment without an invoice, the client who has to be asked for the same bank statement twice. This is where a lot of firms have already started. Some now have incoming emails linked to the right client file automatically, which saves a step that used to be done manually. The next step is an agent that compares what has arrived with what is normally expected for that client, such as last quarter’s documents, regular suppliers and bank payments without a matching invoice. It then lists the gaps and drafts the request to the client. The bookkeeper checks the list and decides what to ask for and when.
Reviewing figures and catching errors
An accountant preparing balance sheet figures for an upcoming meeting notices negative balances appearing where they shouldn’t be, so he builds a simple agent and limits it to a handful of trusted sources: official tax and reporting guidance and the documentation of his accounting package. He then shares his balance sheet, trial balance and account codes with it. The agent then finds the cause: transactions that should have been recorded as income received in advance had been booked as accrued income. Then it explains how to correct the entries.
Answering clients and following up
Client questions arrive all year, by email, by phone and often at the busiest moment. Many of them are variations on questions the firm has answered before: which costs are deductible, when the next VAT return is due, why a payment reminder came in, which documents are still needed. Each one takes time to look up and write out, and the answer can differ depending on which colleague happens to reply.
This is where many accounting firms are starting with AI. An agent connected to the firm’s mailbox reads the incoming question, looks at the client’s file and the firm’s own procedures, and drafts a reply in the firm’s tone and standard structure. The accountant reads the draft, adjusts it where needed and sends it. Questions that need a professional assessment are flagged rather than answered, so the agent prepares and the accountant decides.
The same approach works for follow-up. When a client has paid but the payment has not been matched, reminders keep going out and the client gets frustrated. An agent can compare the client’s message with the payment records, identify what went wrong and draft a reply and a correction for the team to approve. A dispute that would normally pass between several colleagues for days can be settled in a single exchange.
Monthly reporting and analysis
An AI agent can collect the month’s expense reports from storage systems like SharePoint, consolidate them, and check each cost line against the budget. It can then point out the figures that stand out, produce charts that summarise the month, and send the results to the finance manager.
That work can be split across several easy-to-build agents, each with its own task:
- one agent collects the reports;
- one cleans and combines the data;
- one compares the figures with the budget and looks for deviations;
- one turns the results into charts or a dashboard;
- one sends the report to the right people through the team’s usual channel
Firms building their own tools
Perhaps the most striking change is who is doing the building. Small firms, each with fewer than ten people, are building tools themselves with general AI assistants:
- A web form that asks prospective clients the right questions before the first call, after which the firm’s conversion rate rose by about a quarter.
- An approval flow in which an invoice sent to a dedicated mailbox triggers an electronic approval request. Once approved, it is booked with the signed approval attached as an audit trail.
- A working prototype of a new onboarding process, built in a few hours by a head of tax with no technical background.
Most firms don’t have a lot of time to spare in the middle of a quarter, which is why ready-made agents that can be set up without technical knowledge are becoming the more common route.
Building an agent with IttesAI
An IttesAI agent works across the tools your team already uses: your mailbox, your CRM and your files. You set it up once, for one task, starting from a template and describing what you need in plain language. To do its job, it needs three things:
- Instructions: how your firm words a document request, what a file overview must include, and where preparation stops and the accountant’s own assessment begins.
- Context: your procedures, templates and a good example of a finished file.
- Connections to tools you already use, like Outlook, Gmail, Salesforce, and more.
Through those connections it can look up information, draft emails and update records. Every one of those actions is shown to you as a draft first, and nothing is sent or changed until you confirm.
Setting up a first agent takes around ten minutes. Three agents accounting teams could build right now:
- A file intake agent. Give it a client’s emails and attachments for the quarter and it returns one overview: what has arrived, what is missing, and what needs attention first.
- A document request agent. Connected to your mailbox, it drafts the reminder for missing documents, explains why each one is needed, and sends it once you have read it.
- A client question agent. It answers recurring client or colleague questions in your firm’s standard structure, from your own procedures, and flags the questions that need an accountant.
Want to see what an AI agent would look like in your firm or finance team? Tell us how your files actually run, and we will tell you where an agent would make a difference.
