AI for Sacramento Law Firms: Intake, Drafting, and Getting Hours Back

Field notes. John Arndt, Soxoa. Published April 14, 2026.

Talk to a partner at a small Sacramento firm about AI and you usually hear one of two reactions. Either “we can’t use that, confidentiality,” or “I heard about the lawyer who filed a brief with made-up citations.” Both concerns are legitimate. Both are also about the wrong use cases.

The firms getting something real out of this are not asking the software to write briefs or make legal arguments. They are pointing it at the operational layer around the legal work: intake, scheduling, document assembly, status updates, billing narratives. That layer takes a large share of a small firm’s week, very little of it bills at full rate, and most of it follows patterns a machine can learn.

Intake is the highest-leverage starting point

Most small firms lose potential clients in the gap between first contact and first conversation. A prospective client fills out a web form or leaves a voicemail on Thursday evening. Someone reviews it Monday morning. By then, that person has already called somebody else.

An intake system that responds within minutes, at any hour, changes the shape of that problem. It collects the facts of the matter in a structured way, checks for obvious conflicts against your client list, screens for practice-area fit, and puts an organized summary in front of the attorney. The attorney still decides whether to take the matter. The automation only makes sure the decision happens while the client is still listening.

Where the week actually goes

None of the following is legal judgment, and all of it is automatable to a meaningful degree. If you want to know which one to look at first, time them for two weeks and see which is eating the most non-billable hours.

Intake and conflict checks

First contact, structured fact-gathering, a conflicts pass against your own client list, and a fit screen before anyone spends attorney time on it.

Routine correspondence

Engagement letters, demand letters, client updates, and the standing set of letters where only the facts change.

Document assembly from precedent

Pulling the right precedent, filling in matter-specific details, and producing a shell an attorney edits rather than writes.

Chasing signatures and documents

The follow-up sequence a paralegal currently runs from memory and a sticky note.

Time-entry narratives

Reconstructing the week from calendar, mail, and document history, usually on a Friday afternoon.

Case status answers

The “where is my matter” call, which is almost always a question about a record you already have.

Drafting: first drafts, not final drafts

The citation-hallucination stories all share one feature. Someone asked a general-purpose chatbot to do legal research and filed the output without checking it. That is not a deployment. That is malpractice with extra steps.

A drafting system built properly works differently. It starts from your templates and your precedents rather than the open internet. It fills in matter-specific details from your case management system. It produces a first draft of a demand letter, an engagement letter, a discovery response shell, or a client update, and it routes that draft to an attorney for review before anything leaves the building.

The value is not that the machine writes better than the attorney. It does not. The value is that the attorney starts from a structured draft instead of a blank page, on documents whose structure is identical every time and whose facts are the only variable.

Nothing generated by a machine should reach a client, a court, or opposing counsel without attorney review. That rule keeps a firm out of trouble, and it is also what makes the rest of this worth doing, because it settles who is answerable before anyone builds anything.

The billing narrative problem

Every attorney knows the ritual of reconstructing the week’s time entries from memory. It is universally disliked, it is frequently inaccurate, and it costs the firm money in two directions: vague narratives get cut by clients, and rushed reconstruction undercounts hours.

This is a good candidate because the raw material already exists. Calendar entries, mail sent, documents edited, calls logged. A system that reads those signals can produce draft time entries with plausible narratives for the attorney to confirm or correct, which is a different task from writing them from nothing.

Whether it is worth building is arithmetic you can do yourself, and I would rather you did it before talking to anybody. Take last month’s entries, ask each attorney how many they reconstructed after the fact rather than logged as they worked, and put your own realization rate against it. If the answer comes out small, the honest recommendation is to leave the workflow alone.

What about confidentiality?

This is the question that stops most firms, and it deserves a straight answer. Consumer tools on default settings are not the place for client information. That is a deployment choice rather than a property of the technology.

Business-grade APIs offer zero-retention terms, meaning your data is not stored or used for training. A system can be built so client data stays in your own database and only the minimum necessary context is sent for processing. The State Bar of California’s guidance on generative AI points in the same direction: understand the tool, protect confidentiality, supervise the output. Read the guidance yourself, and have whoever owns professional responsibility at your firm approve the architecture before it touches a matter. Nothing in this article is legal advice, and no automation relieves an attorney of the duty to supervise.

One more thing worth saying plainly. A ban with no approved path tends to produce unapproved use rather than no use. If your firm has decided AI is off limits, the thing to check is whether that decision has actually held.

Where AI does not belong

Legal strategy. Settlement judgment. Anything that turns on weighing credibility or reading a room. Novel research where the cost of a wrong answer is a sanction. The technology is not there, and a vendor telling you otherwise is selling something.

That is a smaller loss than it sounds, because those were never the constraint. In a firm of a few attorneys the constraint is the operational drag around the legal work, which is exactly where this technology is strong.

Where to start

Pick the workflow that hurts most and has the least judgment in it. For most firms that is intake response time or billing narratives. Then run it in this order.

If you would rather not do the first pass alone, a Live Build Workshop is one day spent building a single working automation with your own team, on your own approved matters, with the people who do the work in the room.

1. Choose one workflow

   One, named, with a person who performs it today. Not a category, and not “document automation.”

2. Pull real examples

   Have that person select thirty to fifty approved real matters, with an explicit instruction to include the ones that ruined an afternoon. Tidy samples pass every test and tell you nothing.

3. Name the reviewing attorney

   Before the tooling conversation, not after. Give them a deadline, a way to reject, and enough context on screen to decide quickly.

4. Write down the bar and the stop condition

   Decide what working means, and what result would make you abandon it, while nobody is invested in the answer yet.

5. Run it for a month, then expand

   One live workflow people actually use beats four half-built ones. Measure it against how the manual version performed before you add the next.

If you want a read on your own firm

I would rather tell you which of these to skip than sell you all of them. Scope and price are agreed before work begins, and a recommendation to leave a workflow manual is a real result, not a failed sale.

Start with the readiness assessment: five dimensions, self-scored, it runs in your browser and your answers are not stored. Schedule a strategy call and bring the workflow you think is worst. Thirty minutes, and an honest answer about whether there is anything worth building.