What does an OCR template actually do?
An OCR template defines where a particular field appears on a document: the VAT number is always on line 4, column 2; the total amount is always in the bottom right. The system reads those coordinates and retrieves the values. This works well when your supplier always sends the same form, in the same layout, without deviations. Think of a large retail chain that delivers the same purchase order in the same PDF format every week. One template, thousands of documents, minimal error rate. Low cost in use, fast to process, deterministic: you know exactly what the system does.
Where templates break down: variation is the enemy
The problem starts as soon as a supplier reformats their invoice, a carrier uses a different CMR layout, or a new customer delivers orders in a format you have never seen before. The template no longer recognises the field, or worse: it reads the wrong value without you noticing. In practice, this means you need to build and maintain a separate template for every sender. A transport company working with dozens of carriers sees this add up quickly: thirty carriers, thirty CMR templates, and every time a layout changes someone has to update the template. That is not automation, that is template management.
What document AI does differently
Document AI does not read a document based on coordinates, but based on context. The system understands that a number following the word 'total' or 'amount due' is likely the invoice amount, even if that number appears in the bottom left rather than the bottom right. It recognises a delivery address as a delivery address, even on a CMR from a carrier you are seeing for the first time. This makes document AI suitable for varying layouts, unknown suppliers, and multiple document types arriving together, such as orders, packing slips, and invoices all coming in through the same inbox. You do not build templates per sender; the system reads what is there and places it in the correct field.
Concrete examples: CMR freight documents and purchase orders
CMR freight documents are a good example of the kind of variation that breaks templates. Every carrier has its own layout: some are digitally generated, others are filled in by hand and scanned, and field positions differ by country of origin. A template-based system requires a separate configuration per carrier. Document AI reads the CMR field by field based on what is written, regardless of layout. The same applies to purchase orders from dozens of suppliers in wholesale: every company has its own format, some send Excel, others PDF, and the occasional one still sends a Word document. Templates do not scale to that; document AI does. That is not a theoretical advantage, it is the difference between a working workflow and a continuous maintenance task.
Honest about the trade-offs: when should you still choose templates?
Templates are not inherently worse. If your document mix is small and stable, you work with few suppliers, and layouts have been the same for years, a template-based approach is cheaper and easier to manage. Document AI requires a higher upfront investment per document type and makes less sense when there is little variation to bridge. The real question is: how much variation is there in your document flow, and how often does it change? Little variation, stable formats, known senders: templates get the job done. High variation, changing carriers or suppliers, multiple document types: with templates you pay for maintenance that you do not incur with document AI. The threshold is not a specific number of documents per year, but the ratio between volume and variation in your own mix.
The case: templates do not scale with variation
The core difference is scalability under variation. A template-based system grows linearly with the number of senders: more suppliers means more templates, more maintenance, more fragile configurations. Document AI scales with volume without variation creating extra work. A business with ten suppliers today that adds fifteen more next year will hit a maintenance wall with templates. With document AI, nothing changes operationally. That is why document AI is not a replacement for templates when documents are stable, but becomes the better choice as soon as variation is the bottleneck.