The question I get asked most often
When I talk to business leaders and IT teams about Azure Document Intelligence, they almost always ask the same thing: do we need to build a custom AI model, or is there something ready out of the box?
It’s a great question, and the answer has a real impact on how quickly you can deploy and how much effort it takes to maintain. Let me break it down in plain terms.
The short answer: start with prebuilt models. Move to custom only when necessary.
Option 1: Prebuilt Models
Microsoft has invested enormous resources in training AI models on millions of real-world documents. The result: a library of prebuilt models that are ready to use today, with no training data required on your end.
Current prebuilt models include:
- Invoices vendor details, line items, totals, tax
- Receipts merchant name, transaction date, amounts
- Identity Documents: passports, driver’s licenses
- Business Cards name, email, phone, company
- W-2 Tax Forms employee and employer data
- Health Insurance Cards member info, plan details
If your documents fall into these categories, prebuilt is your fastest path to automation. You can be extracting accurate data in an afternoon, no machine learning expertise needed.
When prebuilt works best:
- Your documents match a standard format (invoices, receipts, common forms)
- You want to move fast and don’t have time to train a model
- Your team doesn’t include ML specialists
- You’re running a pilot or proof-of-concept before committing to a full build
Option 2: Custom Models
What happens when your documents are unique to your business or industry? That’s where custom models come in.
With custom models, you provide sample documents (Azure recommends at least 5, ideally more), label the fields you want to extract, and train the model. Azure handles the underlying ML — you’re just teaching it what to look for in your specific documents.
When custom models are the right call:
- Your documents have a proprietary or industry-specific layout
- You need to extract fields that don’t exist in any prebuilt model
- Accuracy requirements are high, and the document format is consistent
- You’re processing high volumes and need a purpose-built solution
Custom models can achieve impressive accuracy once trained on 20–50 representative samples. The initial investment pays off quickly at scale.
Option 3: AI Builder (for non-developers)
There’s actually a third path that’s worth mentioning, particularly for Power Platform users: AI Builder.
AI Builder sits on top of Document Intelligence and wraps it in a no-code interface inside Power Apps and Power Automate. Business users not IT teams can build document processing workflows without writing any code.
The trade-off? Less flexibility than working directly with the Azure API, but dramatically lower barriers to entry.
AI Builder is ideal when:
- Your team lives in the Power Platform ecosystem
- Business users (not developers) are building the automation
- Speed of deployment matters more than fine-grained control
A practical decision framework
Here’s how I walk clients through the decision:
Start by asking: Is my document type covered by a prebuilt model? If yes, use it. Don’t over-engineer.
If no: Is my document format consistent and repeated at volume? If yes, train a custom model. The upfront effort delivers long-term ROI.
And if your team is non-technical: Is Power Platform your primary toolset? If yes, AI Builder is your best friend.
You don’t always need the most sophisticated option. The right model is the one your team can actually deploy and maintain.
The bigger picture
One thing I emphasized at GPPB 2026: the goal isn’t to implement AI for its own sake. The goal is to eliminate manual, repetitive work so your team can focus on higher-value tasks.
Whether you use a prebuilt model for invoices or a custom model for your unique compliance forms, the outcome is the same: faster processing, fewer errors, and a team that’s working on things that actually matter.
The technology is ready. The question is just which path makes sense for where you are right now.
