Document processing
Documents from the inbox into the ERP,processed by an AI agent.
The AI document processing agent takes over an entire workflow, end to end: it receives invoices, delivery notes or confirmations from email or a shared folder, extracts the data, compares it with your orders and rules, then sends it to the right system and the right person. Documents that do not match go to a colleague, with the reason explained.
Documents move, but someone has to push them from one step to the next
The problem is not just reading a document, but its entire journey: someone finds it in the inbox, opens it, copies the data, checks whether the order exists, saves it in the right folder and lets the colleague who needs to approve it know. As volume grows, documents get lost between steps, and transcription errors are only discovered at reconciliation.
- Documents arrive through different channels and in different formats, with no single point of entry
- Data is copied by hand into the ERP or spreadsheets, with a risk of transcription errors
- Checks against orders and contracts happen late or not at all
- It is hard to see what stage a document is at and who is working on it
Who it is for
- Administrative and procurement departments that receive invoices, delivery notes and order confirmations every day
- Logistics companies that handle transport documents from many partners, in different formats
- Companies that manually enter data from PDF or scanned documents into the ERP
- Back-office teams that want a clear, traceable path for every document received
What the document processing agent does
- 01
Takes in incoming documents
It monitors the designated email addresses and folders, identifies the document type and logs it.
- 02
Extracts the relevant data
It reads the supplier, document numbers, dates, line items and totals, including from scanned PDFs of acceptable quality.
- 03
Checks against orders and rules
It compares the extracted data with the order, goods receipt or contract in the system and flags any differences.
- 04
Routes the document
It sends it to the ERP or the appropriate approval workflow and notifies the person responsible.
- 05
Handles the exceptions
It sets aside incomplete, illegible or mismatched documents, with an explanation of the problem, for a colleague.
- 06
Keeps a trail for every document
It records what was extracted, what was checked and where it was sent, for internal audit.
How it works
A document's journey through the agent
The steps marked in yellow are approved by a person.
- Input
- AI
- Action
- Human in the loop
- Result
- 01
Input
Document received
An invoice or delivery note arrives by email or in the shared folder.
- 02
AI
Classifies and extracts
The agent recognises the document type and extracts the required fields.
- 03
Action
Checks in the ERP
It compares the data with the existing order and goods receipt in the system.
- 04
Human in the loop
Colleague resolves exceptions
Differences and unclear documents are checked by a person.
- 05
Action
Routing and logging
The validated document enters the ERP and the approval workflow.
- 06
Result
Complete trail
The document's status and history are visible to the team.
What information and systems the agent uses
- The mailboxes and folders dedicated to incoming documents
- Real examples of each document type and from the main partners
- Access to orders, goods receipts and suppliers in the ERP, for checking
- The validation rules: tolerances, mandatory fields, who approves what
- An extraction module for PDFs and images, configured on your documents
Possible integrations
- Microsoft 365 / Outlook
- Google Workspace / Gmail
- SharePoint or Google Drive
- SAP Business One
- WinMentor
- Saga
- SmartBill
- RO e-Factura (Romania's national e-invoicing system, through your invoicing system)
- Your own ERP's API
- Microsoft 365 / Outlook
- Google Workspace / Gmail
- SharePoint or Google Drive
- SAP Business One
- WinMentor
- Saga
- SmartBill
- RO e-Factura (Romania's national e-invoicing system, through your invoicing system)
- Your own ERP's API
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What stays in human hands
- Documents that differ from the order, goods receipt or contract
- Illegible or incomplete documents, or those from new partners
- Approving payments and accounting entries, which stays with authorised people
- Changing the validation rules and accepted tolerances
Limits and how we manage them
- Poor-quality scans, handwriting or stamps over text can lead to incorrect extraction
- A new format from a partner is processed reliably after adjustment and testing on real examples
- The agent cannot check what does not exist in the system: if an order is missing from the ERP, the document goes to exceptions
- It does not replace the accountant: accounting entries and tax compliance remain the specialists' responsibility
How we measure processing quality
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Accuracy of extracted fields
- We compare the extractions with the real values on a set of documents checked manually by the team.
- Accuracy of routing
- We track how many documents are redirected by hand after the agent has sent them.
- Share of documents sent to exceptions
- We analyse the reasons for exceptions to see what can be clarified in the rules or sources.
- Time from receipt to logging
- We measure the interval before and during the pilot, on the same document types.
- Differences discovered late
- We count the mismatches found only at reconciliation, compared with the previous period.
How we implement the document workflow
- 01
Document inventory
We establish which document types come in, from where, in what volume and what is done with each one today.
- 02
Rules and connections
We define the fields, checks and path for each type and connect the ERP with minimal permissions.
- 03
Pilot on a single document type
We start with the most frequent type, in parallel with the current process, and compare the results.
- 04
Gradual expansion
We add other document types and partners only once the first workflow runs reliably.
Use cases
The scenarios below are conceptual examples, not delivered implementations.
- FMCG distribution
Supplier invoices
The agent picks up invoices from email, compares them with the orders and goods receipts in the ERP and sends only the matching ones to accounting, with the rest going to a colleague.
Conceptual example - Road haulage
Transport documents
The agent extracts data from the transport documents received from drivers and partners and attaches them to the correct trip in the internal system.
Conceptual example - Construction
Delivery notes and progress reports
The agent logs material delivery notes against the correct site and flags quantities that exceed the order.
Conceptual example
Frequently asked questions
How is it different from a simple data extraction tool?
Extraction is just one step. The agent takes over the whole workflow: receipt, checking against your systems, sending documents on and handling exceptions. If you only need the extraction engine, we can build it separately.
What does the agent do when it is unsure about an extracted value?
It does not pass it on. It flags the uncertain field, attaches the original document and moves it to exceptions, where a colleague confirms it.
How are the documents and their data protected?
Documents stay in your systems and folders, the agent only gets access to what it needs, and every operation is logged. Together we decide on retention and where processing takes place.
Does it work with scanned documents too?
Yes, as long as the scan is legible. Poor-quality or handwritten documents are automatically sent for human review.
Does the agent make accounting entries?
It can prepare the data for entry, but accounting and tax validation stays with your specialists.
Which documents should we start with?
With the most frequent and most standardised document type, usually invoices from your main suppliers. We run it in parallel with the current process and add other types only once the first workflow runs reliably.
Can’t find the answer? Write to us and we’ll reply directly.
Want documents to reach the right place on their own?
Send us the list of document types you process every day and we will look at which workflow an agent could take on, with exceptions checked by the team.
Related solutions
- Intelligent document processingThe engine that extracts, classifies and summarises documents in batches and delivers structured data, ready to review and import.
- AI agent for workflowsRuns multi-step processes that need decisions along the way, where fixed automation stops at the first exception, with human approval at the sensitive points.
- AI agent for financial reportingPrepares periodic reports from your financial and operational data and flags variances, which the finance team then analyses.