How A.I. Invoice Review Actually Works (Legal Tech Lesson)
To help get the conversation back on track, we’d like to explain the application we know best: A.I. powered legal invoice review. That way you can more confidently assess what the technology might mean for your business.
But before we do…
Key Definitions To Know
Artificial Intelligence (A.I)
The development of computer systems that carry out tasks typically requiring human intelligence, such as visual perception, voice recognition, and language processing.
A specific application of A.I. that directs computer systems to autonomously interpret data and continuously learn from experience to improve the accuracy of their outputs over time.
The most common version of machine learning, which trains computer systems to recognize inputs and predict outputs by supplying it with examples of known input-output pairs.
Artificial Neural Networks
Inspired by biological neural networks, these systems learn to perform various tasks by considering sample sets of data, typically without being programmed with any initial guiding rules.
A.I. Applied: Legal Invoice Review
Most corporate legal professionals don’t need convincing on the benefits of reducing their role in the tedious process of invoice review. But what does it actually look like when A.I. picks up the slack instead?
A well-trained system will rapidly read invoices as they are submitted and automatically interpret which legal services the individual line item narratives are describing. The resulting outputs can then be compared against your law firm billing guidelines to make invoice approval decisions.
It takes significant supervised learning time to match the reliability of human reviewers, though. The system needs a steady supply of example inputs and outputs to analyze.
At Brightflag, we’ve spent the past five years feeding our platform a balanced diet of legal service classifications. And along the way, our team of legal analysts has been continuously refining the system with human feedback. Correct categorizations are rewarded, mistaken predictions are penalized, and the algorithm gradually grows sharper.
A.I. Invoice Review Example
The process begins by categorizing legal data into objects and activities within the narrative lines. The machine learning engine trains and re-trains on legal service descriptions that correspond to specific billing categories. For example, if we look at these line items:
The phrases “pro hace vice motion” and “bill of sale” are categorized as objects and the words “discuss” and “review” are categorized as activities.
We can then use this data to train the system to recognize future cases that may not have recognized previously. In other words, we take what a computer would normally see as jumbled data in which there is no visible pattern and use machine learning algorithms to identify patterns and sequences.
For example, the model used to categorize the invoice below would have to be trained to figure out the relationship of the person mentioned, John Smith, to the customer within the matter. This provides you with context that wouldn’t normally be detected without A.I.
The Real A.I. Advantage
Our immediate goal is for the A.I. to interpret the substance of an invoice and notify you when a potential violation to your billing guidelines has occurred. But the greater purpose of A.I. invoice review is to not only save time and identify potential savings, but also generate contextual data you can use for strategic decision-making.
Instead of simply confirming whether or not your team was able to review all invoices in a timely fashion, you can start asking more valuable questions like:
- What type of legal work accounts for the majority of our spend?
- Which matter phases represent the bulk of the work?
- Who are all the contributors associated with a specific matter?
- How much are we likely to spend on this matter per month?
- When are we going to hit the budget?
- Are the rates in these invoices abnormally high?
- How would the typical firm resource this matter internally?
- Which firms are most efficient?
Too important to ignore, too tedious to tolerate. That’s the tension most in-house legal teams feel when…