Every invoice that lands on a reviewer’s desk is a quiet drain on legal resources – and there are thousands of them.
At Onit’s AI Center of Excellence we ask an important question: Can Large Language Models (LLMs) outperform human professionals in legal invoice processing? The answer turned out to be more decisive than even we anticipated.
Here we breakdown the major findings from the full research paper conducted by the AI Center of Excellence, available here.
1.AI Is Beating Humans
In our study on contract review, Better Call GPT, seasoned professionals still had the edge over AI. This time, the results are different.
Every LLM tested scored higher than every human reviewer category.
Experienced lawyers topped out at accuracy F score of 72%, while top performing LLMs achieved a score of 92%.
The disparity only grew at the line-item level. Gemini 2.0 led with a score of 81%, almost double the experienced lawyers’ 43%.
Takeaway: AI driven invoice review has real potential to improve the accuracy and quality of our invoice review. They are not here to replace human output, instead, they are here to enhance it.
2. AI is 50-80x Faster
When it came to speed, LLMs were in a different league. The fastest LLM model completed reviews in 3.6 seconds – over 50 times faster than experienced lawyers’ review time of 194 seconds.
Takeaway: At scale, the throughput advantage of LLMs has huge implications for transforming legal operations. What might take a team of reviewers weeks to process, AI can complete in hours.
3. AI is 99.97% Cheaper
Cost is where the conversation gets really interesting.
Human invoice review does not come cheap. In our study, human review rates ranged from $4.17 to $4.47 per invoice. In contrast, LLMs operated at between $0.01 - $0.13 per invoice. While exact figures, the magnitude of that gap is unlikely to close anytime soon.
Takeaway: The financial case of AI assisted invoice review is overwhelming – the scale of the savings makes it difficult to ignore.
What This Means for Legal Teams
Our most nuanced finding was that human reviewers who exercised discretionary judgement improved their accuracy by 6%. Professional judgement still has an important role.
The best path forward is not wholesale replacement, but a smarter division of labour. AI handles the volume consistently and cheaply, while human reviewers focus where it matters.
The Future of Legal Spend Management is AI Native
Our study on contract review, Better Call GPT, made it clear that AI has a serious potential to streamline legal operations tasks. This research goes further: AI isn’t just showing potential, it’s ready to deploy.
Legal teams that integrate AI now will process more invoices, enforce billing guidelines more consistently, and stay ahead of a field that is moving fast. Those that don’t risk falling behind.
The question is no longer whether AI can handle legal invoice review. It’s whether your organisation is ready to let it.
Want to explore the full findings? Download our Better Bill GPT whitepaper.
