The first question partners ask
When a managing partner asks "where do we start with AI?" the honest answer is: stop trying to automate the whole practice. Start with the work that is repetitive, time-sensitive, and high-volume. The five workflows below are the ones we see paying off first in firms between five and forty lawyers. None of them replace judgment. They replace the first pass.
1. First-pass intake
The bottleneck is rarely the legal work. It is the opening of every new enquiry: reading the email, pulling names and dates from the attached documents, running conflict checks, and routing the matter to the right associate. A managed AI employee reads every new enquiry, extracts the key facts, runs the conflict check, and queues the matter for review. The associate opens a file that is already organized.
In a typical mid-size firm, this returns six to ten hours of associate and paralegal time per week, on work that was not generating billable output.
2. Document review and triage
When a new matter lands, the first document review is mostly classification, not judgment. Which contract is this. Which clauses are non-standard. Which exhibits are missing. A managed AI employee can pre-classify, extract, and flag exceptions before a lawyer ever opens the file. Lawyers still do the substantive review. They stop doing the clerical opening.
3. Conflict checks
Conflict checks have to be thorough and they have to be fast. They are also a workflow: search the matter database, search the email archive, surface related parties, and document the result. That is exactly the kind of structured, repeatable work an AI employee is good at. The lawyer still signs off. The lookup stops taking an hour.
4. Client status updates
Clients want a status update more often than firms have time to send one. A managed AI employee can draft the update from the matter file, send it on the cadence the firm chooses, and let the lawyer approve or amend before it goes out. This is one of the highest-return changes a firm can make for client experience, and it costs almost nothing to run once it is set up.
5. Billing follow-up and trust replenishment
The least favourite work in every firm is the monthly chore of asking clients for outstanding documents, signature pages, and trust replenishments. A managed AI employee handles the first pass: identifies the gaps, drafts the message, sends the reminder, and routes anything that does not resolve to a human. Collections velocity improves without anyone in the firm feeling like a collections agent.
What to ask before automating any of them
- Where does the input come from? (Email, case management, web form.) If the input is unstructured, the AI employee needs clear examples of "good" before it can run unsupervised.
- What does "done" look like? A workflow that ends in "a human reviews it" is fine. A workflow that ends in "an email goes out" needs a real review gate.
- Where does the data live after? If it has to land in a specific field in a specific system, the integration matters more than the model.
The takeaway
You do not need to automate the whole practice to get the value. Pick the workflow that costs the most hours per week, scope it tightly, and run a managed AI employee against it for a quarter. Most firms find that the first workflow pays for the engagement and surfaces the second one by itself.