The three processes that eat the most time in a typical firm
In most accounting firms, the biggest time drain isn’t the actual calculation work — it’s three repetitive tasks: keying data from documents into the accounting system, checking whether a client has actually sent everything needed for the month, and answering the same client questions over and over. None of this requires expert judgment. It requires attention and consistency, which is exactly the kind of work where AI earns its keep from day one, without forcing a firm to redesign its whole accounting process.
It’s worth separating two things early on. One is entering and organizing data — here AI acts as a reader and classifier. The other is the actual calculation: taxes, social security contributions, settlements — which needs to stay deterministic, because a mistake there costs real money and carries real legal consequences. We wrote about this split of roles in AI proposes, code decides — the same principle applies to accounting: AI reads and proposes, the engine calculates.
This distinction isn’t academic. Firms that skip it eventually run into a case where the model produced a figure, nobody checked it, and the responsibility still lands on the firm. So before talking about time savings at all, it’s worth deciding which tasks stay entirely with AI, and which have to pass through a deterministic engine or a person.
Data entry from documents: savings you notice immediately
This is the process where AI delivers the fastest, most measurable effect. Invoices, receipts, and payment confirmations arrive at a firm in wildly different shapes today: PDFs exported from sales systems, scans, phone photos, emails with attachments. A language model can read the content of a document like this — the amount, the counterparty, the date, the VAT rate — and turn it into structured data, instead of forcing someone at the firm to retype every line by hand.
In practice, this means the accountant doesn’t start working on a document from zero — they start from a ready proposal they only need to confirm or correct. It still requires their attention: the model can get it wrong on an unreadable scan or an unusual invoice layout. But it shortens the most mechanical first step of working with a document. We covered the mechanics of this kind of automation in Document workflow automation with AI.
Completeness checks: AI keeps watch, a person decides
The second process is the everyday headache of every accounting firm: has this client sent everything for the month? Is one cost invoice missing, or a payment confirmation for social security contributions? Keeping track of this by hand across dozens of clients is tedious and easy to get wrong, especially as the month-end deadline gets closer.
AI is good at comparing what’s come in against what should have come in, based on the pattern from previous months, and flagging gaps before they become a problem when closing the period. What matters is that this stays a signal to check, not an automatic decision to close the month without a person involved. Escalating to a person whenever there’s uncertainty is a principle we apply across all our implementations — we wrote about it in Decision automation with human escalation.
Client communication: it helps, but it doesn’t replace
The third area is client questions — mostly the same ones, recurring every month: “can I expense this,” “when is the social security payment due,” “are my documents for March complete yet.” An AI assistant with access to a client’s current status can answer most of these without pulling an accountant away from their work.
Here, though, the line is clearer than in the previous two areas. Questions about a specific tax interpretation, an unusual client situation, or anything outside routine should go to a person — not because AI can’t phrase an answer, but because the firm, not the language model, carries the responsibility for that interpretation.
In practice, this means AI-assisted client communication is worth starting with a narrow set of questions the firm already knows well from experience. Only once that narrow scope works reliably does it make sense to widen it. Trying to cover every possible client question at once usually ends with the firm losing track of what AI is actually saying on its behalf.
Where AI still doesn’t help — and why that matters
We’re deliberately not claiming that AI “calculates taxes” or “closes the month on its own” — because it doesn’t, and promising that would be dishonest toward firms that remain accountable to their clients and to the tax office for getting it right. The actual calculations — rates, deadlines, amounts due — should come out of a deterministic engine, not a language model. This distinction mattered enough to us that we built it as the foundation of our own accounting product, Qkwit.
What this looks like in practice: Qkwit
In Qkwit, AI reads documents and extracts the data, and a deterministic engine calculates taxes and social security contributions from it — not the other way around. The product is live today for four segments: sole proprietors, limited companies, payroll, and accounting firms, which get one panel showing every client company at once. AI books for you in the automatic mode you switch on — or prepares everything for one click.
For an accounting firm, that means client documents book themselves and the team only reviews exceptions, so the same firm can genuinely serve more clients without growing headcount.
How to tell if it’s actually paying off
Rather than counting how many processes got automated, it’s worth tracking what actually disappears from an accountant’s calendar because of it, and whether the time saved gets reinvested where their judgment is actually needed: interpretations, dealing with the tax office, unusual cases. That’s a more honest measure of progress than rolling out another tool.
What this means for your firm
A good starting point is one narrow process with a visible payoff, measured on a small group of clients before it spreads across the whole firm. We describe how we design this kind of rollout step by step on our AI implementation page.
Frequently asked questions
Will AI replace the accountant at an accounting firm?
Not when it comes to decisions and responsibility. AI speeds up the mechanical parts: reading documents, tracking completeness, answering routine questions. Accounting and tax decisions stay with the person and the deterministic engine doing the calculations.
Where should an accounting firm start with AI?
With one narrow process that has a clear payoff — usually document data entry. Once that’s settled, it’s worth extending automation to completeness checks and client communication.
Can AI misread an invoice?
Yes, especially with unreadable scans or unusual formats. That’s why AI-proposed data should go to a person for confirmation, not straight into the books.
Does Qkwit support accounting firms yet?
Yes. Qkwit is live for four segments — sole proprietors, limited companies, payroll, and accounting firms — with one panel where a firm sees every client company and only reviews exceptions.
Does adopting AI require replacing the firm’s accounting system?
Not necessarily. Automating data entry or completeness checks can run as an added layer on top of the existing process, without replacing the system the firm already keeps its books in.