Why reading invoices is the easy part
Most tools, even basic ones, correctly extract an invoice number, VAT number, total amount, and supplier name from a PDF. That has been a solved problem for years. The question is not whether you can pull out those fields. The question is what you do with them once you have them. An invoice from supplier X for 4,840 euros only means something when you know which order or delivery it belongs to, whether the quantities match, and whether there is a pricing agreement behind it. That part cannot simply be read from the invoice itself. It lives in your ERP, your purchasing module, or sometimes in an email thread from three weeks ago.
What the learning curve looks like in practice
In the first weeks of an automation project, there is always a moment when the team takes a hard look at the matching logic. Suppliers write order references on invoices differently from how you refer to them internally. One supplier uses your purchase order number, another uses their own article code, and a third only mentions the delivery note. Sometimes there is no reference at all. That means you have to decide upfront: which fields do you use as the key for the match, what is the fallback when that key is missing, and when do you route it back to a team member rather than guessing? Those choices largely determine what percentage of invoices flow through automatically and how many end up in the review queue. Well-considered matching policy is the fastest way to raise your automation rate, not better extraction technology.
When automating invoice processing is not worth it
There are situations where it is better to wait before automating. If you have no structured purchasing process, no order numbers, and no fixed supplier agreements, automatic matching has very little to work with. The system reads the invoice correctly but has nothing to attach it to. The result is that almost every invoice still requires manual review, which saves no time at all. The same applies if you only receive invoices from a handful of fixed suppliers with stable formats and amounts. In that case, a simple template check or even manual review is more efficient than building a fully automated flow. Automation pays off most when there is volume, variation in formats, and recognisable patterns in order references.
How human-in-the-loop works for invoice matching
The model that works best in practice is not full automation, but controlled automation. dottle extracts the invoice fields, attempts to match them to the corresponding purchase order or delivery, and flags cases where that match is uncertain. A team member approves those cases; the rest flows through automatically. That is fundamentally different from handling everything manually: the team member spends their time on exceptions, not on retyping invoices that are perfectly clear. At Cabooter Group, dottle processes more than 70,000 documents per year, with 85 percent flowing through fully automatically. That percentage was not reached on day one; it is the result of precise matching rules refined throughout the implementation.
What to arrange before you start
Most of the time is not spent on the technical integration with your accounting package or ERP, but on cleaning up your supplier master and defining the matching rules. Which suppliers include a purchase order number as a reference? Which do not? Are your order numbers consistent, or does every buyer use their own numbering? Are there pricing agreements or contracts you want to validate on receipt? Answer those questions before you go live and the system will operate at a higher automation rate from day one. Discover them after go-live and you will be playing catch-up. Both routes work, but the first is considerably more comfortable.