Workshops: AI Automation & Agents

Workshop: AI Document Processing for Invoices & Contracts

In this workshop finance, procurement and legal teams learn how AI reads, checks and structures documents such as invoices, delivery notes and contracts. You build a processing chain with OCR, a language model and human review that fits your existing systems.

Who is this workshop for?

  • Accounting and accounts payable teams
  • Procurement and logistics with high document volumes
  • Legal and contract management
  • Process owners in shared service centres
  • IT teams planning ERP and DMS integrations

What you will take away

  • You understand how classic OCR and multimodal language models work together.
  • You extract invoice data such as supplier, amounts, tax rates and line items into a fixed schema.
  • You analyse contracts for terms, notice periods and deviating clauses.
  • You add plausibility checks and review steps for uncertain results.
  • You choose between standard products, automation tools and custom development.

Agenda

  1. Module 1

    Teaching AI to read documents

    • OCR, layout detection and multimodal models
    • Structured vs. unstructured documents
    • Placing e-invoice formats such as XRechnung, ZUGFeRD and Peppol
  2. Module 2

    Invoices and receipts

    • Extraction with Claude, GPT or Gemini into a JSON schema
    • Matching against purchase order and delivery note
    • Exercise: an invoice batch with edge cases
  3. Module 3

    Contracts and long documents

    • Finding clauses and comparing them to your template
    • Turning deadlines and obligations into an overview
    • Limits: AI does not replace legal review
  4. Module 4

    Building the processing chain

    • Intake via email or scan, processing with n8n, Make or Power Automate
    • Confidence scores and review queue
    • Handover to ERP, DMS or accounting software
  5. Module 5

    Compliance and operations

    • GDPR, retention and bookkeeping requirements
    • Audit-proof logging
    • Metrics: cycle time, error rate, rework

Why language models change document processing

Classic OCR solutions needed a template for every supplier layout. Multimodal models such as Claude, GPT or Gemini understand meaning rather than position and cope far better with new layouts.

They are not infallible, though: digits can be swapped and line items missed. That is why you learn to validate results automatically and pass only uncertain cases to people.

In the workshop we use your own anonymised sample documents, deliberately including problem cases such as poor scans, multi-page invoices or credit notes – because these determine whether automation holds up in daily work.

Typical use cases

In the workshop we work on the documents that cause the most manual effort in your organisation.

  • Capturing and pre-coding incoming invoices
  • Matching delivery notes with purchase orders
  • Checking contracts for deadlines and risk clauses
  • Digitising forms and applications
  • Reading travel expense receipts

Data protection and confidentiality

Invoices and contracts contain personal and confidential data. We cover business plans without training use, EU hosting, data processing agreements and scenarios for local models. Tax and commercial law requirements should be confirmed with your tax advisers.

Frequently asked questions

How accurate is AI extraction?

That depends on document quality and types. We measure it with your sample documents in the workshop and show how validation rules reduce the error rate.

Does this replace our existing invoice software?

Not necessarily. AI often complements existing systems for edge cases or document types that have been handled manually so far.

Can AI assess contracts legally?

AI can find clauses and flag deviations; the legal assessment stays with your lawyers.

Workshop

Get in touch

How many hours a month does your team spend retyping documents?

Tell us a little about your team – we will get back to you within 24 hours with a proposal for content, format and dates.

Prefer to talk?

We only use your details to handle this request. See our privacy policy.