JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

1 of 45

AI Integration in Legal Practice Necessitates Subscription Billing

Mathew Kerbis, The Subscription Attorney

2 of 45

In this session, you’ll learn:

(1) About the latent legal market

(2) How to correctly apply artificial intelligence for legal services

(3) Implications of AI efficiencies on billing time

(4) Tools and strategies for implementation

3 of 45

Disclaimer:

Everything today is for educational purposes only

Things are changing all the time in this fast moving legal AI landscape.

This is a snapshot of the present moment.

Companies pivot their products all the time anyway.

Not all of these companies will survive.

Discussing an AI tool is not a warranty or representation of the quality of the tool.

Do your own further research into each tool and use demos/trials.

4 of 45

Mathew Kerbis Bona Fides:

Launched Subscription Attorney LLC in March of 2022

Using AI since launch (still a daily user)

Using Generative AI since November 2022 when ChatGPT launched

Teaches lawyers how to ethically use AI in law practice

Teaches lawyers how to adopt and use hardware and software

Target market is latent legal market

5 of 45

My Software/Platforms:

Canva

LinkedIn Live

River

Meetup

6 of 45

My Software/Platforms:

Canva

LinkedIn Live

River

Meetup

7 of 45

Latent Legal Market:

According to Grand View Research, approximately $396.80 billion was spent on US legal services in 2024

https://www.grandviewresearch.com/industry-analysis/us-legal-services-market-report

World Justice Project estimates that 77% of US legal issues are unresolved by a lawyer (https://worldjusticeproject.org/)

Only 23% of the market is being served by lawyers

Potential legal market of approximately $1.328 trillion

Probably greater since this is based on most fees being billable hours

8 of 45

9 of 45

AI for Law Firm Commercials:

On May 20, 2025, Google launched Gemini Veo 3

This is a game changer

We don’t need more than 8 seconds for online ads

Here’s my 7th prompt in Veo for a currently fictional legaltech company:

An asian woman walking away from the camera in a city with skyscrapers. She turns her head around, looks right at the camera, and says, "Is your attorney using Practee? They should be."

Next, the unedited video, then my edit that took 2 minutes in Descript

10 of 45

11 of 45

12 of 45

Imagine the possibilities…

13 of 45

Old World:

Rent equipment

Hire a production team

Audition an actor (and pay them)

Write a script (kind of did this as a prompt)

Find a day to shoot

Hope that day would be sunny

Get lucky that passerbys didn’t ruin the shot

Edit it over days or weeks

14 of 45

New World:

Prompt AI tool new feature that I’m already paying for

Make minor tweaks in an AI-powered editing tool I already have

Publish to the world

15 of 45

Old World:

New World:

Spend a lot of time and money to get a complicated thing done

Don’t spend anything more and get it done faster and better

16 of 45

Same principle applies to legal services.

17 of 45

Hypothetical:

You charge $500/hour

10 hours of billable time makes you $5,000

Generative artificial intelligence can make 10 hours of billable work only take 10 minutes

Under the billable hour in a GenAI world, $5,000 worth of work is only billable at $100

What do you do?

Raise your rates?

18 of 45

Hypothetical:

To make .2 of an hour worth $5,000 means your billable hour rate would be $25,000/hour

That’s a 5,000% increase to your hourly rate (or 50x)

Will your clients be happy to be to pay that rate?

Are you ethically allowed to charge that high of a rate?

What does that mean for serving the latent legal market and access to justice?

What are your options?

19 of 45

Generative AI:

What makes AI generative?

It generates something!

Analysis is not generative but it can be powered by AI

GenAI can revolutionize the practice of law when used correctly

GenAI’s training data is not a source of facts or data

YOU or a DATABASE are the source of facts or data

It’s not perfect but it’s less prone to error than humans

20 of 45

Another Analogy:

Your hire a contractor to do you work at your home

You’re going to pay them by the hour

They show up with a manual screwdriver and saw

You know power tools exist — why didn’t they bring them?

You’re not paying them by the hour unless they use power tools

So the contractor comes back with power tools

Sets down a 2x4 to saw in half

Then uses his power drill to drill sequential holes to cut it in half

*I tried to use AI to generate this image but it was so nonsensical that it didn’t work

21 of 45

Another Analogy:

When you use ChatGPT to do legal research and then check all of your citations, this is EXACTLY what you are doing

At best, you get a case available online

At worst, it makes something up

Wrong tool for the job

It will take longer and have rough edges

AI is not all equal and cannot be used for all things

Use purpose-built AI for specific tasks

22 of 45

Retrieval-Augmented Generation:

What is retrieval-augmented generation and why does it matter?

RAG is when Generative AI (GenAI) draws from a specific source of truth when responding to a prompt.

The purpose is to improve the accuracy and relevance of responses.

External knowledge sources include documents, databases, and APIs.

Reduces if not completely eliminates “hallucinations.”

LLMs are not necessarily trained on the sources of truth.

23 of 45

24 of 45

GPT Wrapper:

Provides a user-friendly interface to interact with and utilize LLMs

Sits on top of the core GPT language model, providing a more accessible and purpose-built user experience

Makes it easier for users to leverage the capabilities of GPT models without having to directly interface with the underlying API or complex prompting

Pre-built templates, industry-specific/specialized models, and mechanisms for user feedback to improve the outputs

Simplify user experience and making GPT more accessible

25 of 45

GenAI is like…

The smartest person on the planet

Who is your assistant

Has read everything on the Internet

Works for free

Or for a much lower cost than a human assistant

But they’re an entry level worker

This is their first job

And it’s their first day on the job

And it’s like that every time you start a new conversion with them

26 of 45

GenAI is like…

What would you do?

You’d provide a lot of context

And follow up and iterate on whatever the work product is

You would not give up if they didn’t get it right the first time

Would you trust this assistant as a source of truth?

Of course not!

This assistant is also not socially well adjusted and doesn't know how to say no or ask follow up questions (unless it does as a GPT wrapper)

27 of 45

GenAI is not…

A calculator

2+2=4 is an immutable truth, and once programmed it always gets it right

Language is all made up and the meaning changes with context

GenAI is a probabilistic model guessing the next most likely word

This is why it can never be relied on, on its own, as a source of truth

28 of 45

General Use vs. Legal-Specific AI:

What makes an AI legal-specific?

The vendor will tell you it’s designed for legal.

Is that enough?

No - but’s a good start.

Inputs/uploads/prompts will not train underlying LLMs.

Each user and/or client will have a separate layer/instance from the LLM and can be replicated through the firm or per client.

Higher security for compliance such as with HIPAA/SOC 2.

Company reps cannot access your data or prompt history.

29 of 45

General Use AI:

Underlying large language models powering the GenAI is training on all documents and prompts put into the tool.

Default typically doesn’t use RAG.

Company can access your prompt history and past uploads.

Less reliable results.

Broader use cases.

Perplexity Pro, NotebookLM Pro (and other Gemini tools), and Descript

30 of 45

31 of 45

32 of 45

33 of 45

Legal AI Categories:

Assistants

Research

Analysis

Drafting

Litigation

Database/Knowledge Base

Intellectual Property

Augmentation

34 of 45

Legal AI Categories:

Assistants

Research

Analysis

Drafting

Litigation

Database/Knowledge Base

Intellectual Property

Augmentation

35 of 45

Legal AI Assistants:

Paxton (proprietary Legal LLM)

LawDroid (customizable)

GC AI (in-house)

Legalyze (practice area agnostic)

Callidus (practice area agnostic)

Centari (transactional)

Eve (Plaintiff litigation focused)

Alexi (litigation focused)

Legora (formerly Leya)

Harvey

Filevine AI

Clio Duo

MyCase IQ

etc.

36 of 45

Augmentation:

Screens AI (acquired by Agiloft)

Gavel (formerly Documate)

PatternBuilder (NetDocs)

Josef Q

SixFifty

37 of 45

Prompt Engineering Tips:

Leverage GenAI like it’s a superpowered personal assistant

Direct it like you’d direct an associate or staff-person

“Review this document and summarize it for me”

“Distill the three most persuasive arguments and explain why”

“Put all of the deadlines and dates in a table with explanations of the obligations next to the date”

“Analyze these three cases, synthesize them in three paragraphs, and list in bullet points the differences between the facts of each case”

You will have results in seconds or minutes compared to hours or days

38 of 45

AI Tips:

Written/spoken natural language is the interface

Follow up, follow up, follow up

Context windows matter

The power of personas

The future is agentic

Non-legal tasks

Semantic search

Subscriber benefit is access to models trained on firm data

Internal version for firm team members to use to increase productivity

39 of 45

The era of the billable hour is over…

40 of 45

ABA Model Rule 1.5:

Terms of Payment

Comment: [5] a lawyer should not enter into an agreement whereby services are to be provided only up to a stated amount when it is foreseeable that more extensive services probably will be required, unless the situation is adequately explained to the client…A lawyer should not exploit a fee arrangement based primarily on hourly charges by using wasteful procedures.