Clients send invoices as phone photos, supplier PDFs and forwarded e-mails, and the accountant keys them in by hand. AI reads the details, assigns the expense account and VAT rate, checks the invoice against SAF-T data and prepares an unapproved entry in SAP Business One, Sage or QuickBooks Online. The accountant approves.
Built for ISO 27001NIS2 DORA
In an accounting firm with dozens of clients, purchase invoices arrive in every possible way: the client's office manager forwards a supplier e-mail with a PDF, the director sends a photo of a receipt from a phone, and some invoices land in a shared firm mailbox in Outlook with no indication of which client they belong to. The accountant opens SAP Business One if the client is kept there, or Sage or QuickBooks Online if not, and keys in the supplier, number, date, amount ex VAT and with VAT, then picks the rate and the expense account from that client's chart of accounts. Then they check the tax authority's e-invoicing and SAF-T data to see whether the supplier reported the invoice, and if not, they write to the client. For one invoice it takes minutes, for a hundred clients in a month it is weeks of work, and under time pressure errors creep in: the wrong VAT rate, fuel booked as entertainment, the same invoice entered twice.
The solution takes over the keying and leaves the decision to the accountant. Invoices are collected from clients' Outlook e-mails, the firm mailbox and the client's SharePoint folder; photos and scanned PDFs are read into text. The model extracts the supplier, company registration and VAT numbers, invoice number, dates, amounts and VAT rate, assigns an expense account based on the supplier, the item description and that client's earlier entries, flags whether VAT is deductible and checks whether the invoice has already been entered. Each invoice is checked against SAF-T and e-invoicing data: if the supplier did not report it or the amounts do not match, that is flagged. The result appears in SAP Business One, Sage or QuickBooks Online as an unapproved entry with the original attached, and the accountant approves it, corrects the account or rate, or rejects it. We deploy private AI in MDP CLOUD's GPU cloud in Lithuania (EU): client invoices do not leave our data centre, the models do not learn from your data, and every recognition and every accountant decision is logged in line with GDPR. We connect to SAP Business One through an API or import file, to Sage through import and to QuickBooks Online through an API, depending on the version.
Four steps from input to a record in your system, with human approval wherever there is risk.
The solution watches the firm's and the accountants' Outlook mailboxes and clients' SharePoint folders. From the sender, the domain and the client list it works out which client an invoice belongs to, and sets unclear ones aside as a question for the accountant. Photos and scanned PDFs are read into text, and multi-page invoices are merged.
The model extracts the supplier, company and VAT numbers, invoice number, dates and amounts, assigns an expense account from that client's chart of accounts in SAP Business One or QuickBooks Online based on the supplier and item description, and flags VAT deductibility. It learns recurring suppliers from the accountant's earlier approvals, and flags duplicates and unusual amounts.
The invoice is checked against the tax authority's e-invoicing and SAF-T data: whether the supplier reported it and whether the amount and date match. Then an unapproved entry is created in SAP Business One, Sage or QuickBooks Online with the original attached and the result of the check. The accountant approves, corrects or rejects it, and only then does the invoice become an accounting entry.
If it is unclear which client an invoice belongs to, a VAT number is missing or the SAF-T data does not match, no entry is created and the accountant gets a question in Teams with an image of the invoice. Every month we review how many entries accountants corrected and, together with the chief accountant, tune the classification rules for each client.
Unapproved purchase entries are created in SAP Business One through an API or import file, in Sage through import and in QuickBooks Online through an API, in all the systems where you keep your clients at the same time.
Every recognised invoice is automatically compared with the tax authority's e-invoicing and SAF-T data, and mismatches and unreported invoices are flagged on the entry.
The recognition model runs in the MDP CLOUD GPU cloud in Lithuania (EU), client invoices are not passed to an outside provider, and the models do not learn from your data.
Every recognition, check and accountant decision is logged with a time and a user. Training for accountants and tuning the rules during the first month are part of the implementation.
Private AI in the MDP CLOUD GPU cloud + SAP Business One, Sage, QuickBooks Online. We choose the platform by the task and by where your data may be located. Read more about this service and its limits on the service page.
No. SAP Business One, Sage or QuickBooks Online receives only an unapproved entry with the original attached and the SAF-T check result. The accountant approves it, corrects the account or VAT rate, or rejects it. Only what a person has approved reaches the VAT return and tax reporting, and the accountant files the return through the tax authority's portal.
From the supplier, the description of the goods or service and that client's earlier entries in SAP Business One or QuickBooks Online. In the first month we review with the chief accountant what the AI proposed and what accountants corrected, and tune the rules for each client. For an unknown supplier the AI suggests an option and marks it as a guess.
No, it is a normal situation for an accounting firm. Each client is assigned its system and the entry is created there: with SAP Business One through an API or import file, with Sage through import, with QuickBooks Online through an API, with Xero and NetSuite through an import file. We have no official partnerships with these vendors; we work with their public interfaces and check what your versions allow before we implement.
A question about VAT deduction or a personal income tax relief costs half an hour of searching tax authority rulings, and the firm's procedures…
Bank statements are downloaded as XML or CSV and matched to invoices by hand, line by line, and unclear references are chased with the client.
After the 10th, accountants write to clients about missing invoices, bank statements and expense reports.
Clients ask in Outlook every day how much VAT and payroll tax to pay, by when and where their invoice is, and the accountant opens SAP Business…
This service is delivered by specialists with valid international certifications. Their expertise is confirmed by independent organisations, not by us.
AI systems are deployed with protection against data leakage and model abuse.
The AI security programme and its controls are managed following the ISACA methodology.
AI risks are identified and managed like any other IT risk.
We reply within 1 business day
