I recently had the pleasure of attending the ABBYY AI Summit in London, where I was fortunate to listen to and chat with a wide range of AI experts. Naturally, the focus was on Document AI and Process Intelligence AI, with one of the key topics being the role of AI within Intelligent Document Processing (IDP) workflows.
Throughout my career, I’ve spent a good deal of time as a business process consultant listening to clients, mapping as-is/baseline processes, and helping to design best-practice workflows for the future. Often, the document sits at the heart of the process, but there are always steps before and after it that influence how (and why) that document is handled. It’s in these stages before and after the actual data extraction where we’re now seeing the greatest need for Generative AI (GenAI) and large language models.
One particularly effective use case that combines Document AI with GenAI is Customer Enquiry Management. The process of handling incoming sales orders often begins with mailbox management. Order processors are typically responsible for monitoring the sales inbox and triaging emails as they arrive often by manually moving them into subfolders for action by specific team members.
The types of requests that come into a sales inbox are many and varied:
- “Please place this new order.”
- “Where’s my order?”
- “Can I add these items to my order?”
And the list goes on. These types of requests often impact not just the sales order processing team, but also warehouse, finance, and beyond. Accurately categorising each enquiry and assigning it to the correct workstream is critical.
By leveraging GenAI and large language models, we can extract the email text and automatically categorise the request. Based on that classification, we can then:
- Determine whether the attachment should be sent to Document AI/IDP for data extraction.
- Identify the appropriate workstream or team to handle the request.
- Trigger a task or workflow in the Business Process Management (BPM) system.
We can take it even further by extracting key contextual details from the email, such as:
- Whether items are being ordered in free text (natural language).
- Whether there are special instructions, e.g.:
- “Please deliver this order to…”
- “As per quote number…”
- “Agreed with XYZ person…”
Typically, we implement this by extracting the body text from the email programmatically and sending it via API to our LLM for classification and enrichment.
These two three YouTube videos showcase these techniques in action and explore the process in more detail:
Using Abbyy Document AI API, python, Gmail and OpenAI:
Using Abbyy Vantage, O365 and OpenAI:
A presentation I gave on extracting and processing natural language orders from the body of an email:
