Frequently asked questions
The key questions, answered
- Who is this for?
- Primarily accounting firms that manage documents for multiple clients and want to reduce manual bookkeeping preparation.
- How many clients can I manage?
- Unlimited. Most of our partners manage 5–30 clients, but the system scales to larger portfolios as well.
- What types of documents does it handle?
- Invoices, receipts, and bank statements as PDF or image files, in any language and any currency, with official ECB exchange rates applied automatically. Contracts and other business documents are filed and summarized too.
- How is this better than OCR?
- The system uses computer vision, which is more accurate than traditional OCR, and can identify relationships between data points, like matching transactions to invoices automatically.
- How accurate is the system?
- For machine-generated documents, internal testing across thousands of invoices shows 99%+ accuracy. For handwritten documents, 95%+. The system flags anything it's unsure about. You only review those.
- Do I need to review every document?
- No. The system flags uncertain items and explains the reason in plain language. You only review what's been marked.
- What happens when documents are missing?
- The system flags bank transactions without a matching invoice, and also identifies those that don't require one (internal transfers, loan payments, payroll, etc.).
- Does it work with custom accounting software?
- Yes, with any software that supports data import (XLS, XML, API). For specific systems, like DATEV or BMD, we build the integration directly into the product.
- How much setup is required?
- Minimal. Provide your chart of accounts and preferred category structure, or simply import partners and categories from a spreadsheet; the AI maps the columns for you. The system learns as you use it: no manual vendor-by-vendor setup.
- What happens above 400 invoices a month?
- Nothing stops. Invoices 401 to 1,500 cost €0.24 each, invoices 1,501 to 3,000 cost €0.23 each, and every invoice above 3,000 costs €0.21. Each rate applies only to the invoices inside its band. Bank transactions are not billed.
- How do my clients use it?
- Clients get their own login: they upload or email documents, answer your questions directly on the record, and see the reports you share. Roles keep the accountant and client sides cleanly separated: clients never see your other clients or your internal notes.
- How does it handle VAT?
- Every invoice line gets a VAT code resolved automatically from the partner group, product/service group, rate, direction, and calculation method, including reverse charge and non-deductible handling. Exports carry the codes, and invoices with incomplete VAT data are flagged before they can be exported.
- Can I connect AI assistants like Claude or ChatGPT?
- Yes. AI Finance Team has a built-in MCP server: connect the assistant you already use and ask about your clients' books in plain language, or let it handle routine work like categorizing and approving matches. It only sees what you can see, making changes requires explicit permission, and every action is logged. An organization admin can switch MCP access off entirely at any time.
- Can it import bank transactions automatically?
- Yes. Clients can connect their business bank accounts through their bank's own secure consent page (banks across the EU and UK are supported). Transactions then flow in daily with read-only access; AI Finance Team never sees banking credentials. Prefer not to connect? Statement upload works just as well.
- Where is our data stored, and is it used to train AI?
- In Frankfurt. Database, application, file storage and transactional email all run in AWS eu-central-1, and the Claude models are served from AWS Bedrock in the same region. Bedrock stores no inputs or outputs, passes nothing to the model provider and never trains on them; its abuse screening is fully automated, so no person reads a prompt, and Anthropic has no access to the account its model runs in. We do store your documents (that is the product), plus the prompt and response logs behind each AI decision, deleted automatically after 14 or 90 days depending on the call type, in Frankfurt. No provider we use trains any model on your data. Every firm gets a GDPR Article 28 data processing agreement and the full sub-processor register before the first upload; the register discloses one US provider used for text embeddings, under zero-retention and no-training terms.
