Every major practice management platform has added an AI layer: Clio has Duo, MyCase has IQ, Filevine, Smokeball, and Lawmatics all ship AI features. They are convenient because your matter data is already sitting there. They are also generally shallower than a good general model connected to the same system, and they cost extra per user. Do not switch platforms to get AI.
I run 10 law firms and I have sat through the demo for most of these. Same pattern every time: a competent, narrow feature set, priced as an add-on, wrapped in a slide that says the future of legal work.
The features are real. The pricing question is the one nobody asks out loud.
What the built-in AI layers actually do today
Strip away the marketing and the feature lists across these platforms converge on five things.
Summarize a matter. You open a case and get a paragraph explaining what happened, who did what, and where it stands. Useful when a case gets handed to a new attorney.
Draft emails from matter context. The AI already knows the client name, the case type, and the last few events, so it writes a status update without you pasting anything.
Surface deadlines and next actions. It reads the calendar and the task list and tells you what is late.
Search your own documents in plain English. “Find the repair invoice from March” instead of clicking through folders.
Generate time entries from activity. This one is the most immediately profitable feature in the category, and I cover it separately in AI billing for lawyers. Smokeball has been doing automatic time capture longer than most.
That is the honest scope. Nothing on that list requires a special legal model. It requires a general model that can see your data.
The limitation nobody puts in the demo
These features are wrappers around general models with your matter data attached. That is not an insult, it is just the architecture. Clio, MyCase, and the rest are not training their own frontier models. They are calling one and giving it your case file.
Three consequences follow.
You pay extra for the wrapper. Most of these are add-on pricing, per user, on top of your existing per-user subscription. A 10-person firm feels that immediately.
You get only what the vendor built. If you want the AI to compare this month’s intake against last month’s and tell you which referral source is going cold, you cannot. That feature does not exist and you cannot add it.
You cannot customize it. With Claude you can build a Project that holds your firm’s demand letter voice, your fee agreement rules, and your case evaluation standards, then use it against your real data. The built-in gives you the vendor’s prompt, not yours.
The comparison
Pricing on these changes constantly and every vendor negotiates. Confirm current numbers with the vendor before you sign anything. Here is the shape of the market.
| Platform | What the AI does | Pricing model | Best for |
|---|---|---|---|
| Clio Duo | Matter summaries, plain-English search across your Clio data, drafting help, task and deadline surfacing | Add-on tied to higher Clio tiers, per user | Firms already fully committed to Clio |
| MyCase IQ | Document summaries, drafting assistance, transcription and insights on case documents | Add-on per user, bundled into upper tiers | Small firms wanting AI without leaving MyCase |
| Filevine AI | Document analysis, medical record and deposition summarization, demand support | Add-on, quoted per firm | Personal injury and mass tort volume shops |
| Smokeball | Automatic time capture, document automation, drafting inside Word | Included in higher tiers | Firms whose main leak is unbilled time |
| Lawmatics | Intake and marketing automation with AI drafting for client communication | Add-on per user | Firms optimizing intake and follow-up |
| Claude plus an MCP connector | Anything you can describe: cross-matter analysis, custom drafting, your own repeatable prompts, combined with email and documents | About $20 to $30 per user for Claude, connector is free | Firms that want capability instead of convenience |
That last row is the argument.
What I would do, and what I actually do
I would rather connect a great general model to whatever system already holds the data than pay a platform for a mediocre built-in. That is the entire reason the Clio MCP route exists and why I wrote it up. Fifteen minutes of setup, no per-seat fee, and the model can then answer questions the vendor never anticipated.
At my lemon law firm the useful questions are never “summarize this matter.” They are things like: which cases have been sitting in document collection past 30 days with no client contact, and draft the follow-up for each one. No built-in AI layer does that. A connected model does it in one pass because it can read the case list, apply my rule, and write the emails.
There is a real exception. When the built-in does something workflow-specific that a connector cannot reach, the built-in wins. Auto-populating court forms is the clearest example: the vendor has the form library, the field mapping, and the e-filing integration, and no amount of clever prompting replicates that. Same goes for automatic time capture that runs quietly in the background all day. Pay for those. Do not pay for a summarize button.
How to choose
-
Do not switch platforms for AI. Migration costs weeks of staff time and the feature gap between these platforms is not worth it. Whatever you are on, stay.
-
Ask your current vendor what their AI does before you buy anyone’s add-on. Get specifics, not a feature page. Ask for a live demo using a real matter from your own account.
-
Test the same task both ways. Take one real case. Ask the built-in to summarize it and draft a client update. Then do the same thing in Claude with the file attached or the connector live. Compare the two outputs side by side. That five-minute test settles the argument faster than any review will.
-
Connect rather than migrate. If your platform has an API or an MCP connector, use it. You keep your system and gain the better model. Start with intake, which is where most firms have the biggest gap between what they collect and what they act on.
The four questions to ask any vendor
Before you turn on any AI feature that touches client data, get written answers to these:
Is my matter data used to train any model, including the vendor’s own? Which model provider processes it, and under what agreement? Where is the processing done, and is any of it offshore? What do I keep, in what format, if I cancel tomorrow?
The last one catches people. Firms discover at cancellation that the AI-generated summaries and time entries do not export cleanly. Ask now, not in year three.
Your action for today: open your practice management platform, find the AI feature you are already paying for or being pitched, and run one real matter through it. Then run the same matter through Claude. Whichever wins, you will have stopped guessing.