How to automate client intake in a law firm

What does an automated intake actually do, step by step?

It takes an enquiry from arrival to a lawyer's approval queue without anyone retyping anything. The shape that works in most small practices:

1. A structured form is the front door. Not a mailto link. Capture the fields a conflict check and a fit assessment actually need: full legal name and any former names, entity names where relevant, the other side's names, a plain description of the matter, the area of law, and how they found you. Keep it short, because every extra field costs completions.
2. A record is created immediately. A prospect or provisional matter record in your practice management system, with the enquiry stored as structured data rather than as a forwarded email. This is what makes everything after it possible.
3. A conflict check is assembled, not decided. The workflow searches your client, matter and party records for the names supplied, including reasonable variants, and produces a candidate list showing the matter, the date, the acting lawyer and the relationship. It presents that list to a lawyer with a clear prompt to make and record a decision.
4. An acknowledgement is drafted and held. Written from the enquiry's own details, in the firm's voice, confirming receipt and setting out what happens next and by when. It sits in the responsible lawyer's queue, not in the client's inbox.
5. A lawyer reviews, edits and sends. One click if the draft is right, thirty seconds of editing if it is not. The decision to send is a human act, every time.
6. The outcome is written back. Accepted, declined or referred, with the reason, so the record is complete and your intake numbers are real.

Steps 1, 2 and 4 are machine work. The decision in step 3 and the act in step 5 are not, and no amount of confidence in the tooling should change that.

Where does the conflict check fit, and who decides?

The conflict check runs before any substantive response, and a lawyer decides the outcome. The automation's job is to make the search complete and fast, so the decision gets made properly rather than skipped because it was tedious.

Three ways it goes wrong, all avoidable.

Silent clearing. If the workflow reports "no conflicts found" and the file opens automatically, you have automated the obligation away rather than the work. Always require an affirmative human decision, including when the search returns nothing, and record who made it and when.

Shallow matching. An exact string match will miss trading names, misspellings, former names, related entities and directors. Search wider than you think you need to and accept false positives: a lawyer glancing at six irrelevant matches costs far less than one missed relationship.

Incomplete sources. If half your history sits in an old system, a share drive or someone's email, a search of the current system is not a search of the firm. Either bring those sources into scope or have the workflow state plainly which records it searched, so the reviewing lawyer knows its limits.

If you use a language model here, use it to expand names for searching or to summarise what a matched matter was about. Do not use it as the search itself, and do not let it form the view.

What are the confidentiality guardrails?

Treat every enquiry as confidential from the moment it arrives, including from people who never become clients, and design the workflow so fewer systems and fewer people touch that information than in your current manual process. A well-built automation usually reduces exposure: it replaces forwarded emails and copy-paste with one controlled path.

The guardrails worth writing into the build:

• Your accounts, your keys. Every system in the chain sits in the firm's name, with the firm holding the credentials. If a supplier's departure would strand your data, solve that before the build.
• Least privilege. The workflow gets the specific records and fields it needs, not the whole practice management system. Its access should be auditable and revocable independently of any person's login.
• Know where the data goes. For each tool, establish where data is stored and processed, who the subprocessors are, and the retention and deletion positions. Get it in writing from the vendor, not from a marketing page.
• Check the training position. Where a model is involved, confirm the vendor's contractual position on whether inputs are retained or used for training, and use business or enterprise terms under your own account. Consumer chat tools are not appropriate for identifiable client information unless you have assessed them and concluded otherwise.
• Send the minimum. If a model only needs to draft an acknowledgement, it does not need the other side's details or the substance of the dispute. Reducing what leaves your system beats any policy.
• Respect information barriers. If your firm uses ethical walls, the automation must honour them. An automated search that surfaces a walled matter to the wrong person is a breach that would not have happened manually.
• Keep an audit trail. Who approved, what was sent, when, and what the conflict decision was. This is the record that makes the process defensible later.
• Fail loudly. If a step cannot complete, the workflow stops and alerts a person. It should never guess, retry silently, or send a partial message.
• Understand your obligations before you build. Depending on your practice these may include duties under the Privacy Act and the Australian Privacy Principles, the notifiable data breaches scheme, your insurer's requirements, and the solicitors' conduct rules in your jurisdiction covering confidentiality, competence and supervision. This article is not legal advice. Confirm how they apply with your state or territory legal regulator, your professional body and your insurer before anything goes live.

What must a lawyer still do personally?

Everything involving judgement, advice or an obligation to the client. Write that line down before the build rather than negotiating it afterwards.

• Decide the conflict position and record the reasoning.
• Decide whether to act at all.
• Review and send anything that goes to a prospective client.
• Handle costs disclosure and the engagement documents. A workflow can prepare and populate them; the content and the sending are legal acts.
• Supervise the process, including reviewing a sample of what the automation produced, especially in the first month.

The acknowledgement draft deserves particular attention. It should confirm receipt and next steps and stop there: no characterisation of the matter, no indication of prospects, nothing that could be read as advice or as an assurance that the firm is acting. Draft that wording once, carefully, because it will be used hundreds of times.

How do you build it without turning it into a project?

Build the narrow version, run it alongside the current process for two weeks, then switch. The failure mode is designing a complete intake system for a firm that has not agreed its own steps.

1. Write down the current intake exactly as it happens, including the parts that only happen when someone remembers.
2. Fix the form first. A structured front door improves intake on its own, before any automation exists behind it.
3. Add the record creation. Now every enquiry is countable.
4. Add the conflict search as an assist, with a mandatory human decision.
5. Add the drafted acknowledgement last, and keep it in draft for longer than feels necessary.
6. Review the first thirty days of output, adjust the rules, and only then consider what else the record could trigger.

What to do next

Start with the form and the record. They are the smallest useful build, they carry almost no professional risk, and they make everything after them possible. Once every enquiry lands as structured data in a system you control, the conflict search and the drafted acknowledgement become incremental additions rather than a leap. If you would rather map and price the whole sequence first, that is what a Shift Flow Map is for.