Choosing the process
Together we pick the process that repeats most often, has a clear definition of "done" and has its data in your systems. We define what the AI does on its own and what it does only after a person approves.
Client invoices arrive in Outlook as phone photos and PDFs, bank statements are keyed in by hand, and month-end close starts with reminder e-mails. AI reads the documents, checks them against SAF-T and e-invoicing data and prepares unapproved entries in SAP Business One, Sage or QuickBooks Online, and the accountant approves. Models run in a private cloud in Lithuania (EU).
Built for ISO 27001NIS2 DORA
For an accounting firm serving dozens or hundreds of clients, every month starts the same way. Client invoices land in Outlook: phone photos, supplier PDFs, forwarded e-mails with no comment. The accountant opens SAP Business One, Sage or QuickBooks Online, depending on which system that client is kept in, keys in the supplier, amount, VAT rate and expense account, and then checks the tax authority's e-invoicing and SAF-T data to see whether the supplier reported that invoice. Bank statements are downloaded as SEPA XML or CSV and matched to invoices line by line. After the 10th the reminders to clients begin: three invoices, an expense report and a statement are missing, while clients write in asking "how much VAT do I pay and when?". In the evening the accountant is still looking up what the tax authority ruled on VAT deduction for car leasing, and at the start of the year everyone prepares the annual financial statements for the business register together.
AI does not file returns or decide tax questions in place of the accountant. It takes over the keying, checking and reminding. Invoices from Outlook, SharePoint and photos are recognised, classified, checked against SAF-T and e-invoicing data and appear in SAP Business One, Sage or QuickBooks Online as unapproved entries. An agent follows the month-end calendar, reminds clients about missing documents and files what arrives into client folders in SharePoint. Bank statement lines are matched to invoices and prepared as payment entries. An assistant answers accountants from VAT and income tax legislation, tax authority rulings and the firm's procedures, with a source, and answers clients from their own accounting data and filing deadlines for VAT returns and payroll filings. The accountant approves every entry, amount and client e-mail. Because this is clients' financial data, the models run in MDP CLOUD's private GPU cloud in Lithuania (EU) or as Claude through Amazon Bedrock in an EU region, with no training on your data, with action logs and GDPR controls. We integrate with SAP Business One, Sage, QuickBooks Online, Xero and NetSuite through APIs, import files or the e-mail flow, depending on the version, and for firms that run an ERP we also offer ERP hosting in a cloud in Lithuania (EU): ERP hosting. The first step is always the same: which staff reach which clients, where the data is stored and what the AI may prepare without an accountant.
The 5 solutions we implement most often in this industry. Each one starts with a single process, has a human approval step and is measured.
Invoices from clients' Outlook e-mails, phone photos and PDFs are read and classified: supplier, amount, VAT rate and expense account from that client's chart of accounts. Each one is checked against the tax authority's e-invoicing and SAF-T data and appears in SAP Business One, Sage or QuickBooks Online as an unapproved entry. The accountant approves, corrects or rejects it, and nothing reaches the books without them.
We build it with: Private AI in the MDP CLOUD GPU cloud + SAP Business One, Sage, QuickBooks Online
Following the firm's month-end close calendar, an agent reminds clients from Outlook about missing invoices, bank statements and expense reports, sorts the files they send into client folders in SharePoint and checks the list of received invoices against SAF-T data. The accountant gets a list of gaps for each client in Teams instead of hunting for them.
We build it with: Copilot Studio agent + Outlook, SharePoint, SAF-T data
Lines from SEPA XML and CSV bank statements are matched to clients' sales and purchase invoices by payment reference, amount and past history, and the prepared payment entries go to SAP Business One or Sage with the status "awaiting approval". Lines that are not recognised stay in a separate list for the accountant. The accountant approves the reconciliation.
We build it with: Private AI in the MDP CLOUD cloud + SAP Business One and Sage import
The assistant answers accountants' questions from VAT and income tax legislation, tax authority rulings and the firm's internal procedures in SharePoint, citing the article, document and date with every answer. If a source is older than a later amendment, the assistant flags it. It does not make tax decisions for the accountant, it only shows where the rule is written.
We build it with: Claude via Bedrock in an EU region + SharePoint procedures, tax authority rulings
For recurring client questions in Outlook (how much VAT, payroll tax and social security to pay, by when, where is my invoice) the assistant prepares a draft reply from SAP Business One or Sage data and the filing deadlines for VAT returns and payroll filings. At month end it drafts a summary e-mail for each client. The accountant reviews and approves before sending.
We build it with: Claude via Bedrock in an EU region + SAP Business One / Sage data + Outlook
From the first conversation to a working solution, with measurement and ongoing support.
Together we pick the process that repeats most often, has a clear definition of "done" and has its data in your systems. We define what the AI does on its own and what it does only after a person approves.
We put access to files and systems in order, give the AI its own identity with the least privileges it needs, turn on an action log and choose where the model runs: your Microsoft 365 tenant, Amazon Bedrock in an EU region or our GPU cloud in Lithuania (EU).
We build the solution with Copilot, Claude or ChatGPT, whichever fits the task, connect it to your systems and test it on historical documents and enquiries until accuracy reaches the agreed threshold.
We launch with a person in the loop, review the logs and exceptions and adjust the rules. Once the first process runs reliably, we move on to the next one on the list.
These solutions are a starting point for a conversation, not a complete list. Tell us which processes repeat in your company, and in one conversation we will tell you what is worth handing to AI and what is not.
No. In implementations for this industry the models run in MDP CLOUD's private GPU cloud in Lithuania (EU) or as Claude through Amazon Bedrock in an EU region: invoices, statements and accounting data are not used to train models and are not shared with anyone else. Access is limited to the clients the accountant can see in SAP Business One or QuickBooks Online, and every action is logged in line with GDPR.
No, and that is a deliberate choice. The AI only prepares entries, checks data against SAF-T and e-invoicing records and reminds you of deadlines. The accountant files returns through the tax authority's portal, submits payroll filings through the social security portal and files annual financial statements with the business register. Every amount and every tax decision remains a human responsibility.
Yes, but the way we integrate depends on the version: with SAP Business One through an API or import files, with Sage through import, with QuickBooks Online through an API, and with Xero and NetSuite through import or export. For a firm that keeps different clients in different systems, the solution works with several at once. We have no official partnerships with these vendors; we work with their public interfaces and check what your version allows before we propose anything.
An assistant that knows your procedures, contracts and instructions, and answers in seconds.
A clear answer on where AI will deliver value in your business, and where it is not yet worth touching.
AI is useful when it runs inside your systems, not in a separate window next to them.
Calls over lunch, Messenger messages, a photo with the question "how much will this cost", mechanic notes on a scrap of paper and insurer forms.
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