I dont know who the target is for this. But if it's big enterprises, then this might not be as appealing as it sounds. I used to work for Datacert, a legal apps vendor for large companies(walmart, jpmc, novartis etc). They use e-billing with a well defined format(ledes) all items on an invoice is code with a specific code(UTBMS). We used a rules engine(Drools) to ensure proper billing. A proper use for machine learning in legal space would be simulations for different pricing models(Alternate fee arrangements(AFA) in legal parlance). If machine learning could evaluate different pricing models for a given case and provide inputs, that would be a big selling point.

You'll get 80% of the benefit just by looking at word frequency, highlighting outliers and then a weight based on factors such as length and secret-sauce weighting.

Bonus points if you're using multiple categorizations (using different weights for different industries).

NLP / statistical stuff is fun ;)

Are you scanning / OCRing the documents? I never managed to get the OCR to be good enough for invoicing, there always had to be a manual process to fix the (machine-learning-flagged) errors.

Or don't you need accurate-to-the-cent invoices?

Word frequency is in use at many larger insurance companies today. You can certainly find problematic bills with word frequency and the hours billed, but you end up with a lot of false positives so you still have to manually review everything. We get in deeper than word frequency.

And, yes! NLP + statistics is fun!