Legal intake automation
Most conversations about automating legal intake start with the wrong question — "can this be automated?" — when the useful question is "should this stage be automated, and what does it cost us when it gets the answer wrong?" This page breaks a personal injury intake into discrete stages and grades each one honestly. Our argument, stated up front: firms consistently over-automate case screening and under-automate follow-up and data capture, which is precisely backwards.
On this page
- Intake is not one process, it is nine stages
- The rule we grade by
- The grading, stage by stage
- Where firms over-automate: screening
- The specific risk of automated case screening
- Where firms under-automate: follow-up and data capture
- The order we would build it in
- The handoff, and what it depends on
- How to test an intake build before it speaks to anyone
- What this does not do
- Questions firms ask
- Go deeper
Intake is not one process, it is nine stages
"Intake" is a single word covering a chain of quite different tasks. Some of them are pure logistics. One of them is a judgement call that determines whether your firm takes a case. Automating them as a single block is how firms end up with a system that is simultaneously too aggressive in the place it should be cautious and absent in the place it should be relentless.
Broken apart, a personal injury intake is roughly this:
- Reaching the contact. Answering the call, acknowledging the form fill, picking up the chat — the act of the firm being present when someone reaches out.
- Identity and contact capture. Name, callback number, email, preferred contact method, and how they found you.
- Conflict check. Establishing whether the names involved already appear in your matter list.
- Incident fact capture. What happened, when, where, who else was involved, what treatment followed, whether an insurer has been in touch.
- Screening decision. Whether this is a matter your firm wants to pursue.
- Consultation scheduling. Getting a specific time on a specific attorney's calendar.
- Follow-up on non-responders. Everything that happens after the first attempt does not resolve.
- Engagement paperwork. Sending, chasing, and receiving the retainer.
- Handoff and reporting. Getting the record into your case management system and knowing, weekly, what came in and what happened to it.
Nine stages. Firms usually buy a product that addresses stage one, and then wonder why the signed-case number does not move. The full lead response system exists because stages one, six, seven and nine are where the compounding is, and they are only loosely related to each other.
The rule we grade by
Automate the stages where the correct action is fully determined by information the person in front of you supplies. Keep human the stages where the correct action requires judgement about facts nobody can verify at the point of contact. Everything in between is a partial: the machine gathers, structures, and flags; a person decides.
This rule is deliberately about verifiability rather than difficulty. Scheduling is not automated because it is easy — it is automated because the caller's availability and your calendar's availability are both knowable at the moment of the call, so there is nothing to get wrong that a person would get right. Screening is not kept human because it is hard — it is kept human because the inputs are a stranger's unverified account of an event, delivered in the hour after it happened, by someone who may be in pain, medicated, upset, or simply bad at describing things.
The second half of the rule is the one vendors skip.
The grading, stage by stage
Three grades: automate fully means no human is required in the normal path; automate partially means the machine does the gathering and flagging and a person makes the call; keep human means a person should own the decision even though a machine could technically produce an output.
| Stage | Grade | Why |
|---|---|---|
| Reaching the contact | Automate fully | There is no judgement in answering. The alternative to an automated answer at 11pm is not a better answer, it is voicemail. This is the lowest-risk automation in the whole chain and the one most firms should start with. |
| Identity and contact capture | Automate fully | Every field is supplied by the caller and read back for confirmation. A machine that never gets distracted, never mishears a number without checking, and never forgets to ask for the email is straightforwardly better than a tired human at this. |
| Conflict check | Automate partially | Name matching can be run instantly and surfaced. The decision about what a partial match means is your firm's, and it is not a decision to make on a fuzzy string comparison at 2am. |
| Incident fact capture | Automate fully | This is the most under-rated automation in intake. A structured set of questions asked identically every time, plus a verbatim transcript and recording, beats a paraphrase typed by someone trying to keep up. Capture is not evaluation. |
| Screening decision | Keep human | The inputs are unverified, the caller is unreliable through no fault of their own, and the cost of a false negative is a signed case handed to a competitor. Hard knockouts only — see below. |
| Consultation scheduling | Automate fully | Availability on both sides is knowable at the moment of the call. Every hour between "someone will call you" and an actual booked time is an hour in which the caller can book with someone else. |
| Follow-up on non-responders | Automate fully | The single largest under-automated stage. Consistency is the entire value, and consistency is the one thing a busy human intake team cannot supply. |
| Engagement paperwork | Automate partially | Sending, reminding, and tracking receipt are logistics. What is sent, to whom, and on whose authority is a decision your firm owns, and no vendor should be configuring it without an attorney signing off. |
| Handoff and reporting | Automate fully | Re-typing an intake record into case management is pure waste. Weekly reporting on volume, source, and outcome should be generated, not assembled by hand — assembled-by-hand reporting is reporting that stops happening in a busy month. |
Six full, two partial, one human. Notice that the one stage graded "keep human" is the stage most heavily marketed as an AI capability, and that three of the "automate fully" stages are the ones firms most often leave manual.
Where firms over-automate: screening
The pitch is seductive: let the system qualify, so your intake specialists only speak to good cases. It sounds like leverage. In practice it usually means your firm has delegated its most consequential operational decision to a script that is executing on a stranger's five-minute description of an event.
The asymmetry is the whole argument. Consider the two ways an automated screen can be wrong:
- A false positive passes a case your firm would not take to an intake specialist. Cost: a few minutes of a person's time, and some mild irritation.
- A false negative turns away a case your firm would have taken. Cost: the entire value of that matter, plus the acquisition cost already spent on the lead, plus the fact that you will never know it happened. Declined leads do not appear in any report as a loss. They appear as efficiency.
Those costs are not remotely symmetric, so a screening threshold tuned to "reduce junk" is tuned against the wrong objective. We build to the opposite default: ambiguity escalates, it does not decline.
There is a narrow exception, and it is worth stating precisely because it is the only automated screening we are comfortable with. Hard knockouts — binary, factual, uncontested criteria that your firm has written down and an attorney has approved — can sensibly be automated. Practice-area mismatch is the clearest case: a caller describing a matter your firm does not handle at all can be given a courteous handling that your firm has specified in advance. Geography, where your firm's jurisdiction is a hard boundary, is another. What cannot be automated is anything that requires weighing: severity, likely value, how credible the account sounds, whether the facts "feel" like a case. Those are judgements, and the machine's confidence in them is not the same thing as their being right.
The specific risk of automated case screening
An over-tuned screen fails silently and asymmetrically. If your qualification logic is wrongly declining a share of viable enquiries, nothing in your reporting will tell you — those callers do not complain, do not appear in your CRM as lost opportunities, and are indistinguishable from genuine non-cases. Meanwhile your intake team's numbers improve, because they are handling fewer enquiries and converting a higher percentage of them. The metric goes up while the firm gets smaller. This is why we insist that every declined or deprioritised enquiry keeps a full transcript, and that someone at the firm reads a sample of them every week for the first month. If a vendor cannot show you the calls their system turned away, they are not showing you the thing you most need to see.
A second, quieter risk: screening scripts drift out of alignment with the firm. A criterion an attorney set in March is still executing in November after the firm's appetite has changed, because nobody owns reviewing it. A human intake specialist absorbs a shift in what the firm wants through hallway conversation. A configured system does not. Whoever installs this should be telling you that the screening logic needs a scheduled review, and if they are not, it is because reviewing it is unbilled work.
Where firms under-automate: follow-up and data capture
These are the two stages where automation is genuinely, unambiguously better than a person, and they are the two stages firms most often leave alone.
Follow-up. The common pattern is that a new lead gets one callback attempt. If the person does not answer an unknown number — which is ordinary, sensible behaviour — the lead ages out of anyone's attention without ever being formally closed. Nobody decided to give up on it. It simply stopped being at the top of a list. A defined sequence of attempts across call and SMS, with a stop rule, costs almost nothing to run and requires no artificial intelligence at all to be worth doing. The automation's contribution is not intelligence, it is that the fourth touch happens on a Tuesday when the intake team is buried. Sequence design and the compliance questions that are genuinely yours to answer are covered on the follow-up page.
Data capture. Compare two records of the same enquiry. One is a set of form fields typed by a person listening and typing at the same time, compressing a caller's account into a summary sentence. The other is the same structured fields, asked in the same order every time, plus the caller's own words preserved verbatim in a transcript with the audio attached. The second is better on every axis that matters — completeness, accuracy, and the ability for an attorney to go back to what was actually said rather than what somebody wrote down. It is also better for your marketing reporting, because source attribution gets asked every time instead of when someone remembers.
- Ask the same questions in the same order. Consistency is what makes intake data comparable across weeks and across sources. Human intake produces good records and bad records depending on the day.
- Keep the verbatim account. A transcript is a materially different asset from a summary field. Firms that have both stop trusting the summary.
- Capture source on every enquiry. Not because it helps intake, but because it is the only honest input into decisions about marketing spend.
- Timestamp everything. You cannot fix a response time you have never measured. Measuring it from your own data is a separate exercise and a prerequisite for knowing whether any of this worked.
The order we would build it in
Grading the stages tells you what to automate. It does not tell you what to do first. Our sequencing is deliberately conservative, because the failure mode of a fast rollout is a bad call to a real prospective client that nobody catches for a week.
- Reaching the contact, after hours only. The narrowest scope with the largest gap behind it. The comparison is against voicemail, not against your best intake specialist, and it reverses in minutes. How that scope is set up is its own page.
- Capture and handoff. Structured fields, transcript, recording, and a record that lands where your team already works. This is where the intake quality improvement shows up.
- Scheduling. Only once the capture is trustworthy, because a booking on top of a bad record is worse than no booking.
- Follow-up sequences. Added when the firm has agreed the stop rules and the consent handling, not before.
- Hard knockouts, if at all. Last, narrow, attorney-approved in writing, and reviewed with transcripts in hand after the first month.
A firm that stops after step two has captured most of the available value. We would rather say that plainly than sell five steps to someone who needs two.
The handoff, and what it depends on
An intake record that has to be re-typed into your case management system has automated nothing. It has moved the typing.
Depending on your platform and what your plan's API actually permits, a completed intake can arrive as a calendar invite with the transcript attached, a structured email or webhook to your intake inbox, or a record created directly in the system your firm runs on. What is achievable differs meaningfully between platforms, and between plan tiers within the same platform.
We do not claim a native, pre-built integration with a platform until we have built and tested it for your firm specifically. If your case management system needs custom work to receive a structured intake record automatically, you hear that during the audit, before you commit — not during implementation. When you evaluate any vendor, ask them which exact fields they write, what happens when a record cannot be matched to an existing contact, and whether the transcript arrives as an attachment, a link, or not at all.
How to test an intake build before it speaks to anyone
Testing is the step firms compress when they are keen to go live, and compressing it is the most common cause of a bad launch. This part is on your side, and it cannot be delegated to us.
- Call it as your worst-case caller. Mumble. Give a partial address. Interrupt. Change your mind about the date halfway through. The polished test call proves nothing.
- Call it as someone who is not a case. Confirm that the handling of a non-matter is something you would be comfortable seeing quoted back to you.
- Call it as an existing client. Existing clients ring the main number. Confirm they are routed, not intaken.
- Ask it a legal question directly. "Do I have a case?" "What's this worth?" It must decline and route, every time, in every phrasing you can think of.
- Read the record, not the demo. Open what landed in your case management system afterwards and ask whether a paralegal could work from it cold.
- Test the reversal. Turn the forwarding off and confirm calls land where they used to. Do this before go-live, not during your first incident.
What this does not do
It does not give legal advice, evaluate the merits of a claim, estimate case value, comment on limitation periods, or say anything that could be read as forming an attorney-client relationship. It captures facts and executes operational rules your firm has written; it does not exercise legal judgement. It does not create attorney capacity — if your firm's constraint is that it cannot work the files it already has, automating intake will make that worse rather than better, and we will say so on the audit call rather than after you have signed.
Questions firms ask
Is it not the whole point that the system qualifies leads for us?
That is the usual pitch, and it is the part we would push back on. The system should capture everything a qualification decision needs and present it cleanly, so the decision takes a minute instead of a callback cycle. Making the decision itself is different, and the cost of getting it wrong falls entirely on you. We will build hard, attorney-approved knockouts if you want them. We will not build a system that weighs case quality.
What is the single highest-return thing to automate first?
Answering after hours, then structured capture with a transcript. Those two together change what lands on an intake specialist's desk at 8am from a phone number to a workable file. Follow-up sequences are close behind and are the most commonly skipped.
Does an automated intake handle sensitive information appropriately?
It is configured around your firm's own intake questions and does not solicit more than your script asks for. Your specific confidentiality and data-handling requirements get walked through during setup. We do not claim a blanket compliance certification — we would rather be precise about what has actually been built for your firm than vague about it.
Who is responsible if the screening logic declines a case we would have taken?
Your firm owns the criteria; we build the mechanism to them. That is exactly why our default escalates ambiguity rather than declining it, why every declined enquiry keeps a transcript, and why we ask someone at the firm to read a sample weekly during the first month. If a vendor is comfortable owning that decision on your behalf, ask them what happens when they get it wrong.
Can we automate some stages and leave others alone?
Yes, and most firms should. The stages are independent decisions. A firm that automates answering, capture, and handoff and leaves screening, paperwork, and everything else untouched has taken most of the available value with almost none of the risk.