A prospective client calls a law firm on a Monday morning. They have a real matter and they're ready to hire someone.
Here is what usually happens next. The receptionist takes a name and a few notes. Someone emails the notes to an attorney, who is in a deposition. A paralegal enters the parties into the conflicts system — the prospective client, the opposing party, the related entities if anyone thought to ask about them. The search comes back with three possible hits that need a human to look at. The attorney who could clear them is out until Wednesday. Meanwhile, nobody has sent an engagement letter, because you can't send an engagement letter until conflicts clear.
The prospective client, who wanted a lawyer on Monday, has spent two days wondering whether this firm is interested. Some of them wait. Some of them call the next name on the list.
None of this is anyone's fault. Every step is necessary. The problem is that intake is a relay race where the baton sits on the ground between every runner.
Intake is the firm's front door
Firms tend to treat intake as administration — a cost center that happens before the real work starts. But intake is the first experience a client has of the firm, and it sets their expectations for everything after. A firm that takes two days to say "we can help" is telling the client something about how the matter will go, whether it means to or not.
We've written about this dynamic outside legal, in The Intake Form Is Where Deals Die and the case for speed to lead. The legal version has one twist that makes it harder: some of the steps genuinely require a lawyer's judgment, and those can't be rushed or skipped. The goal is not to rush them. The goal is to make sure the lawyer's judgment is the only thing anyone is waiting on.
What can be automated safely
Break intake into its parts and most of it turns out to be gathering, formatting, and routing — work that needs accuracy, not judgment.
Capturing the intake completely, the first time. A structured intake — by form, by phone with a transcribed summary, or both — that asks for every party and related entity up front, so the conflicts search isn't run twice because someone forgot to ask about the parent company.
Running the search immediately. The moment the parties are captured, the search runs. Not when a paralegal gets to it.
Assembling the hits into a decision packet. Each possible conflict, with the prior matter, the relationship, the dates, and the relevant notes pulled into one page — so the attorney who clears it reads for two minutes instead of digging for twenty.
Routing to whoever can decide now. If the responsible attorney is out, the packet goes to the backup the firm has already designated, with a clear flag of what's waiting.
Drafting the engagement letter in parallel. The letter can be prepared from the firm's template while conflicts are pending, so it goes out the minute they clear instead of starting then.
Keeping the prospective client informed. A short, human-sounding acknowledgment that the firm has the matter, what happens next, and when they'll hear back. Silence is what sends people to the next firm.
What must stay with a lawyer
Some decisions in this process belong to an attorney, full stop.
Whether a hit is a real conflict. Whether it can be waived and what the waiver requires. Whether the firm wants the matter at all. What the engagement terms should be. Those are professional judgments with ethical weight, and no automation we'd build would make them — or would be allowed to send anything to a client that implies they've been made.
This is the design principle we build into every AI employee: the system gathers and prepares, a person decides. In legal work the line is simply brighter, and the audit trail matters more. Every search, every packet, and every decision should be logged with who made it and when.
A worked example
Here are round numbers for a hypothetical mid-sized firm, just to show how the math works. Say it opens around forty new matters a month. Say intake, conflicts, and engagement currently take about ninety minutes of combined staff time per matter, spread across four people, and a few calendar days end to end.
Automating the gathering, packeting, and drafting might cut the staff time to around thirty minutes per matter. That's forty hours a month returned to people whose time is worth a great deal more than data entry — call it $60 an hour loaded, and it's roughly $28,800 a year before you count a single matter that no longer walks out the door while it waits.
That second number — matters retained because the firm answered in hours instead of days — is usually larger, and it's the one nobody measures. Your numbers will differ. That's why you run them before building anything.
Built with privilege in mind
Law-firm automation has requirements that most business automation doesn't: privilege, retention schedules, ethical walls, and client confidentiality that has to survive every integration. Our legal practice is led by a president who is an attorney and former General Counsel, with engineers who have delivered for DLA Piper, and those rules shape the design from the first conversation. Where the data can't leave the firm, the models can run on the firm's own hardware — the approach we laid out in When the Data Can't Leave the Building.
The build itself usually lives in our Document & Back-Office Automation work: intake capture, conflicts packeting, and document assembly are all variations on reading, structuring, and routing.
Where to start
Time one intake. Not the process document — an actual matter, from first call to signed engagement, with timestamps. Most firms find that the working time is small and the waiting time is enormous, and that the waiting clusters at two or three handoffs.
Those handoffs are your candidates. If you want a structured way to see whether they're worth automating, the AI Automation Scorecard walks through it in a few minutes.
The conflict check is never going to be instant, and it shouldn't be. But the only thing a new client should ever be waiting on is a lawyer's judgment — not a form sitting in someone's inbox.