Industry · Accountancy

AI for accounting firms. Custom-built, tested in practice.

We build AI applications that fit how an accountancy firm really works: invoice processing, audit assistance, client communication and compliance. Not a standalone AI package bolted on top, but an integrated flow in which Twinfield, Exact, AFAS or CaseWare keep doing what they do well, and AI speeds up the work around them. With an audit trail, AI Act governance and data sovereignty built in from the design stage.

DomainMid-market
DomainTax advice
DomainAudit & assurance
DomainTrust & notary services

Dutch accountancy in numbers.

~700
Accountancy organisations with an AFM licence
~22.000
RAs and AAs in the NBA register
~14M
Incoming invoices per year in SME accounting packages
71%
Expect concrete AI use within three years

Source: AFM register, NBA annual report 2024, accountancy sector survey 2025.

Standard AI tools cover about half.

The big vendors (Wolters Kluwer CCH iFirm, Visma, Exact, AFAS, CaseWare) have built AI modules into their products. For classic invoice processing, bank transaction categorisation and basic OCR, they work well enough. Until a firm has its first non-standard requirement: a proprietary audit risk framework, a sector-specific check for a healthcare or housing association client, an onboarding flow that needs to cover Wwft (Dutch anti-money laundering act), UBO and sanctions list screening in a single step. That is where the standard package stops.

Mid-market accountancy firms often get stuck on the combination of five or more systems: accounting software, CRM, document management system, e-invoicing platform and AML tool all need to work together as one coherent flow. Big-4 spinouts and challengers want their own LLM flow in which client data is guaranteed not to leak to OpenAI or Anthropic, and they often build their market position on that data sovereignty. Tax advisory firms have their own clause libraries and advice templates that generic AI can do nothing with. Bookkeeper cooperatives want a white-label tool that all affiliated members can use, without the cooperative having to set up its own IT team. Trust offices with an accountancy arm deal with AML and sanctions list screenings that differ per jurisdiction, and with UBO structures that are more complex than any standard onboarding can handle. Notaries with financial and governance advisory services face the same compliance requirements and can often reuse the same flows.

We build the custom layer for that. Not a replacement for Twinfield or CaseWare, but an AI layer on top that supports exactly the flow that sets your firm apart. With the right governance, a watertight audit trail and compliance from the first sprint. The starting point is always: what do you do better than other firms, and how does AI make that differentiator more scalable without the quality of your judgement suffering?

Which firms do we build for?

Mid-market accountancy firms

Ten to a hundred employees, their own character and specialisation, often with clients in one or two sectors for which the standard packages don't quite fit. This is where most of the demand for integrated AI flows across five or more systems comes from.

Big-4 spinouts and challengers

Firms founded by former partners of the Big Four who set themselves apart with a specific audit approach, their own risk framework or deeper sector expertise. They often build their market position on data sovereignty, and an AI layer of their own fits that perfectly.

Tax advisory firms

Their own clause library, their own advice templates, their own lines of interpretation. Generic AI has little to offer here; a private LLM flow trained on your own knowledge base does.

Bookkeeper cooperatives

Dozens or hundreds of independent bookkeepers who collectively want an AI platform, presented white-label by the cooperative. Build once, roll out many times, with central governance and local autonomy.

Trust offices with an accountancy arm

Complex UBO structures, AML screenings per jurisdiction, sanctions list checks that shift every quarter. Off-the-shelf packages are rarely sufficient here; for this type of firm, custom AI is more the rule than the exception.

Notaries with financial and governance advisory services

Notaries who, besides executing deeds, also provide governance advice and tax planning face almost the same compliance and AI requirements. The flows we build for accountancy firms lend themselves well to reuse here.

AI applications we typically build.

Four areas where we develop custom AI for accountancy firms in practice. For each area: the flow, the integrations and the governance required.

Invoice and bookkeeping automation

OCR is no longer a differentiator; the gains are in classification. An LLM that learns, per client, which cost centre belongs to which supplier, which VAT rate applies to which type of expense, and which ledger entry category the firm uses. The accountant only sees the exceptions, not the standard rules.

We build that flow on a custom LLM layer, connected to Twinfield, Exact, AFAS or another accounting package. This includes receipt extraction via a mobile app, batch categorisation of bank transactions, and anomaly detection in annual accounts across years and against sector benchmarks.

  • Smart invoice classificationOCR plus LLM classification of ledger entry category, VAT rate and cost centre, learning per client.
  • Receipts app for clientsMobile extraction with automatic matching to cost centre and project.
  • Mass categorisation of bank transactionsLLM batches with per-client training data, with no data leaking to public models.
  • Annual accounts anomaly detectionDeviations between financial years, sector benchmarks and industry-specific outliers flagged automatically.

Three other areas where AI makes the difference.

Audit assistance. In audit engagements, the work is shifting more and more from execution to judgement. An LLM that scans the bookkeeping for red flags on the auditor's behalf, audit sampling that combines statistical sampling with risk-based scoring, anomaly detection between financial years and sector benchmarks: that is where we see most implementations. MindBridge AI Auditor is often the commercial starting point; we only come into the picture when the firm wants its own risk framework that does not fit into an off-the-shelf package. Think of healthcare accountancy, where DBC (diagnosis-treatment combination) checks and NHC classifications need to be automated, or housing association audits where the split between DAEB (services of general economic interest) and non-DAEB activities must be verified separately. The audit team retains final responsibility; the AI does the groundwork and flags where human judgement is essential.

Client communication. A mid-size firm sends thousands of emails every week: questions about VAT returns, parts of the annual accounts, payroll administration details, tax changes. A large share of these are repetitive. A well-built client communication assistant reads incoming email, recognises the intent, and drafts a reply with references to that client's relevant files. The accountant or tax adviser reads and adjusts it; responsibility for the content remains entirely human. For mid-market clients, we often build a CFO-as-a-service layer on top: an automatically generated explanation of the monthly report, to which the firm adds its own style and advice before it goes to the client. It scales advisory work to many more clients without losing the feeling of personal contact.

Document classification and DAS. Incoming post, whether physical or digital, now needs to be routed automatically: a UBO notification, a Wwft (anti-money laundering) document, a VAT correction, an amended annual report. A classification model on top of the scanning software can do this with considerable accuracy. On top of that, we build Document Automation Solutions: standard reports, audit opinions and advisory templates that produce a first draft based on the client data, after which the accountant finishes and signs the document. It saves the most time in the assistant role, where capacity is currently scarcest. For VAT return and payroll administration checks, we build similar flows, where the LLM cross-checks collective agreement (CAO) application, holiday entitlement and shift allowances against the client's input and presents outliers to a payroll administrator.

Compliance from the first sprint.

AI in an accountancy firm touches more regulation than in almost any other sector. From the design stage onwards, we work within the frameworks of the Wta (Dutch Audit Firms Supervision Act), NV COS (Dutch Standards on Auditing), AFM supervision, GDPR, the AI Act, the Wwft and the forthcoming CSRD obligations.

Wta & NV COS

Professional rules for accountants

The Wta (Dutch Audit Firms Supervision Act) and the Dutch professional rules (NV COS) set requirements for independence, professional competence and file keeping. Our AI flows respect that separation: the AI supports, the RA or AA (registered or certified accountant) remains ultimately responsible and visibly performs the final step.

AFM supervision

Auditable AI decisions

The AFM (Dutch Authority for the Financial Markets) is increasingly looking at the use of AI by licence holders. We build an audit log per AI decision: which model, which prompt, which input, which output, which human review step. An AFM supervisor can reconstruct every decision.

GDPR Art. 35

DPIA and client data privacy

By definition, accounting involves PII and special categories of personal data belonging to your clients' clients. We deliver a DPIA covering data flows, retention and data subject rights. For LLM flows, depending on the firm, we opt for on-premise models, EU-hosted inference or a private endpoint with no data retention.

AI Act Art. 4

AI literacy in the firm

Mandatory since February 2025: every employee who works with AI systems must be demonstrably AI-literate. Delivery includes a training package and documentation. Where relevant, we link to our AI literacy training.

Wwft

AML and client due diligence

The Wwft (Dutch Anti-Money Laundering Act) requires a risk-based approach to client onboarding. Our flows automate UBO checks, sanctions list screening and PEP checks, with human review for every high-risk flag. We build in sector-specific differences — trust, accountancy, legal — explicitly.

CSRD

Sustainability reporting

The Corporate Sustainability Reporting Directive is coming for accountants through the assurance role. AI helps with data aggregation from client systems, scope 1/2/3 classification and year-on-year consistency checks. We build this as a standalone module that plugs into your existing audit flow.

Not sure about a large project yet?

Test your idea first: a working prototype in 1 day

With OneDayBuild, we make your idea tangible in a single day for €1,150, so you know whether further development is worth the investment. Decide to go ahead with the full build? Then we deduct the cost in full.

View OneDayBuild →

Seamlessly connected to the accountancy ecosystem.

We don't replace your accounting package or audit tool. We integrate with it, add AI where it creates value, and let Twinfield, CaseWare or AFAS keep doing what they've done well for years.

Twinfield
Visma accounting package
Exact Online
Mid-market accounting
AFAS
ERP & payroll
CaseWare
Audit software
CCH iFirm
Wolters Kluwer suite
Klippa
Document & receipt OCR
Basecone
Invoice scanning and recognition
KvK + UBO
Trade register API

Custom where off-the-shelf stops.

For commodity flows, we almost always recommend the standard offering. Klippa or Basecone for pure receipt scanning is faster and cheaper than anything we can build — and we'll tell you so. We only come into the picture when more than three systems need to work together, or when the AI layer needs its own risk framework or industry-specific flow that standard packages don't cover.

Think of healthcare accountancy with automatic DBC checks, housing association audits with a DAEB/non-DAEB split, or a cooperative of independent bookkeepers that wants a shared white-label AI tool. For these kinds of challenges we build the custom solution and integrate it tightly with your existing stack, often drawing on our experience with custom LLM integrations and custom accountancy software.

The architecture always follows the same pattern. We place a thin abstraction layer between the AI model and the accounting package, so that both the model and the underlying stack can be replaced without rewriting the business logic. Prompts and model choice are version-controlled, tested with representative client data, and automatically evaluated for accuracy before a new version reaches production. That's deliberate: the LLM landscape changes every few months, and a firm running on GPT-4 today should be able to switch to Claude 5 or an open-source alternative within a year without hassle, if that suits it better on price, speed or compliance.

From first exploration to ongoing management.

An AI project for an accountancy firm has its own rhythm — different from a regular software implementation. Five phases that are identical for every firm.

01 · Audit

Workflow mapping

A day at the firm with a partner, controller and paralegal. Goal: which flows really cost time, where the AI potential lies, and where it doesn't.

02 · Design

AI scope and risk classification

For each use case, we determine the AI Act risk class, the governance level and the data flow. The result: a design that an AFM supervisor can follow.

03 · Build

Two-week sprints

One working AI flow delivered every sprint. Accountants test alongside us right away. We work with both public and private LLM endpoints, depending on data sensitivity.

04 · Rollout

Phased go-live

First one team or client group, then the next. With every rollout: training, documentation and the mandatory AI literacy module for users.

05 · Maintenance

Ongoing

We cover model monitoring, drift detection and regulatory changes in a maintenance subscription. First-line response within the same working day.

How the sector views AI.

NBA Vision 2025

"AI is changing the accountancy profession from executing work to forming judgements. Firms that invest in their own AI flows now are building a distinctive position with mid-market clients."

Accountant Adviseur, March 2025

"Custom AI is on the rise at firms that don't fit the standard packages. Sector-specific audit flows and combined Wwft screening are the two most requested applications."

AFM publication on AI in the financial sector

"The AFM expects licence holders to ensure that AI decisions are verifiable, traceable and ultimately the responsibility of a human. An audit trail and governance documentation are not optional but a starting point."

Answers for the firm just starting with AI.

The questions we hear most often from directors, controllers and IT leads in accountancy.

What does AI actually do in an accountancy firm?
In practice, we see four main uses: invoice processing and bookkeeping automation (OCR plus LLM classification), audit assistance (anomaly detection and sampling), client communication (generated draft replies that the accountant verifies) and compliance (Wwft screening, UBO checks, AML). AI is also increasingly used for CSRD reporting and in a CFO-as-a-service role for mid-market clients, such as automated commentary on monthly reports.
Will AI replace the accountant?
No, and the regulations would not allow it anyway. Under NV COS and the Wta, a RA or AA remains ultimately responsible for the opinion. What AI does is speed up the repetitive, executional tasks. The accountant does more assessment and advisory work, and less data entry. For mid-market clients, that often means more room for the strategic conversation.
Which AI tools already work well in accountancy?
For commodity flows, the off-the-shelf packages are usually the right choice: Klippa and Basecone for invoice OCR, Twinfield and Exact for accounting AI, MindBridge AI Auditor for audit-specific analysis, CaseWare for audit working papers. We actively recommend them when they fit. We only come into the picture when the flow requires deep integration between five or more systems, a custom risk framework, or sector-specific logic that no package offers, such as healthcare accountancy with DBC checks.
Privacy: does client data leak to OpenAI or Anthropic?
That depends on how you build it. In our standard design, client data never goes to the public APIs of OpenAI or Anthropic. We work with EU-hosted private endpoints (Azure OpenAI with a no-data-retention contract, Anthropic via a private endpoint, or on-premise models such as Llama and Mistral for the most sensitive flows). For every implementation, we explicitly record which data is processed where, in a data flow document that forms part of the DPIA.
How do we combine GDPR, the Wwft and the AI Act at the same time?
By treating them as one coherent framework rather than three separate compliance requirements. The GDPR determines the data flow and retention periods, the Wwft determines which checks are needed at client onboarding, and the AI Act determines the risk classification and governance requirements. For every implementation, we deliver one integrated document covering all three frameworks, with explicit choices per flow. For the organisation-wide approach, we often refer to enterprise AI implementation.
What determines the cost of an AI project for an accountancy firm?
Three factors: the number of systems we integrate with (one bookkeeping package is a different matter from five systems working together), the governance level that matches the AI Act risk class (high-risk requires considerably more documentation and testing), and the extent to which the firm wants its own LLM layer (a private endpoint or on-premise model costs more than a public API with a DPA). We work with a fixed sprint budget and give a concrete price for the full build after the design phase.
How does implementation work in practice?
We start with a workflow audit: a day at your office with the partner, controller and paralegal. We then design the AI scope and risk classification. We build in two-week sprints, with accountants testing directly, and the first working flow is ready after a few sprints. Rollout happens in phases per team or client group, with training and the mandatory AI literacy module. After that, we move to ongoing management with model monitoring and maintenance for regulatory changes.
Do you also work for smaller firms, or only the mid-market?
Custom AI is rarely the right choice for a firm with fewer than about ten employees. Twinfield, Exact or a Klippa subscription usually provide enough AI functionality for the price. We mainly get involved with firms of ten to a hundred employees, Big 4 spin-offs, bookkeeper cooperatives that want a shared white-label tool, and trust offices with an accountancy arm where regulatory complexity outgrows the standard package. A good AI strategy often starts with a broader conversation about AI strategy.

Ready to make AI concrete in your accountancy firm?

A half-hour introduction with a partner and a controller or IT lead. We listen, ask questions and give you an initial direction: which AI use cases would deliver the most value for you, and where you're better off sticking with the standard package. No obligation, no sales pitch.

Edit content