When someone has just been arrested, exploring AI for criminal defence lawyers highlights why after-hours intake matters so much. Their partner is standing in a police station car park at 2 am with a phone and a Google search. They ring the first firm: voicemail. They ring the second: voicemail. They ring the third and a voice answers, takes the details, and says a lawyer will call back within the hour. That third firm has the client. That is the entire competitive dynamic, and it is why this is not a novelty conversation any more. It is a switchboard conversation.
This post covers what AI can genuinely do for a defence practice, what it must never touch, where the privilege and conduct risks sit, and how to work out whether an after-hours system pays for itself in your matter mix.
Key Takeaways
- AI for criminal defence lawyers earns its keep at intake, not advice: answering, qualifying and routing urgent after-hours calls while you sleep.
- Criminal work is bought at the worst moment of someone’s life, usually within an hour of the arrest, and almost always by whoever picks up first.
- An AI receptionist can take instructions, flag custody and bail urgency, and text you the summary, without ever giving legal advice.
- Privilege is a design question. Handle it by limiting what the AI collects and where the recording lives, not by avoiding the technology.
- Conflict checks stay human. AI can pull the names and run the first database pass, a lawyer signs off before anyone is engaged.
- Dormant client lists in a defence practice are worth working, but only with careful messaging, as criminal matters are not a repeat purchase to advertise.
- The first system to build is the one covering the hours you are unreachable, because that is where the lost instructions actually are.
Why the 2 am Call Decides Who Gets the Client
Criminal defence has an unusual buying pattern. Almost no other professional service is purchased in a police station car park by a frightened family member who has never needed a lawyer before. There is no consideration period. There is no comparing quotes. There is a phone, a list of names, and whoever answers.
Research on lead response consistently shows the first responder wins the overwhelming majority of the time, roughly 78% in the InsideSales and Harvard Business Review work on response time. That research was done on B2B web enquiries. In criminal defence, the effect is stronger, because the urgency is real and the caller has no reason to wait.
So look honestly at your own coverage. Business hours, someone answers. Evenings, it goes to a mobile that may or may not be picked up. Weekends, patchy. Public holidays, forget it. Yet arrests do not respect the roster. Friday night, Saturday night, and the early hours are peak. Your firm is least reachable precisely when the demand spikes.
The usual fix is a rotating after-hours mobile. It works, in the sense that calls get answered some of the time. But it means a partner sleeping with a phone on the pillow, a duty roster nobody wants, and a slow slide into everyone ignoring it. That is not a staffing problem you can hire your way out of. It is a systems problem.
What AI Should and Should Not Do in a Defence Practice
The fastest way to get this wrong is to imagine AI giving legal advice. It should not, and the moment you let it, you have a conduct problem rather than an efficiency gain. The line is clean: AI handles the conversation before the lawyer, never the conversation that is the lawyer.
Here is where it sits comfortably.
Answering and qualifying the call
An AI voice receptionist picks up on the first or second ring, any hour. It introduces itself as an assistant, confirms it is not a lawyer, and takes the details: who has been arrested, where they are being held, what they have been charged with if known, whether a bail hearing is scheduled, and a callback number. It then texts or emails that summary to whoever is actually on call, with an urgency flag.
That is a receptionist’s job, not a solicitor’s. The proof that this works is not theoretical. Dr Claire, a dental practice owner, was losing 47% of inbound calls despite having two receptionists. After a voice AI receptionist went in, missed calls went to zero and booked appointments rose 44%. Different industry, identical mechanic: the phone gets answered, the enquiry gets captured, and the business stops leaking at the front door.
Triage and routing
Not every after-hours call is a custody matter. Some are existing clients asking about a court date. Some are process questions. Some are people who want a different area of law entirely. An AI system can sort those in the first thirty seconds and only wake you for the ones that genuinely cannot wait until morning. That alone changes how sustainable an on-call roster is.
Intake paperwork and file opening
Once a matter is accepted, the same captured detail flows into your practice management system rather than being re-typed from a scrawled note. Names, contact details, the station, the charge, the hearing date. The AI drafts the engagement letter and client care information for a lawyer to check and send.
What stays human, always
Advice on charges, plea, or bail. Anything approaching a view on the merits. The conflict check sign-off. The decision to accept or decline the matter. Any contact with police or the court. Sentencing strategy, negotiation, and advocacy. These are not efficiency targets, and any vendor telling you otherwise should worry you.
Privilege, Confidentiality and the Questions Your Insurer Will Ask
This is the section most articles about legal AI skip, and it is the one that actually determines whether you can use any of this.
Legal professional privilege does not automatically evaporate because a system handled the first call, but confidentiality obligations are absolute, and a poorly configured tool creates real exposure. Treat it as a design problem with four questions.
Where does the data live? Call recordings, transcripts, and intake summaries need to sit somewhere you can point to on a map, and somewhere your professional obligations allow. Ask the vendor for the region, not the marketing copy. For NZ and Australian firms, this usually means insisting on local or clearly disclosed hosting.
What does the AI collect? Less is safer. The intake script should gather what is needed to triage and call back. It should not invite a distressed caller to explain what happened or say anything about the facts of the alleged offence. A good script actively steers away from that: “Don’t tell me the details of what happened; I just need to know where you are and how we reach you, and a lawyer will call you back.”
Who can hear the recording? Access controls, retention periods, and a deletion schedule. Decide how long transcripts are kept and enforce it. Most firms land on keeping the intake summary in the file and deleting the raw recording after a short window.
Is the caller told? Disclose that they are speaking to an AI assistant. Every time, in the first sentence. Aside from the conduct considerations, callers respond better to honesty than they do to discovering it later.
Handled this way, the risk profile is comparable to using an external after-hours answering service, which plenty of firms already do. The difference is that the AI answers every call rather than the ones the service gets to, and it never misreads a name.
Building It: Start With the Hours You Are Unreachable
The mistake I see most often is a firm trying to do everything at once. Someone reads an article about legal AI, decides to look at document review and research tools and intake and marketing simultaneously, and six weeks later nothing has been implemented because it turned into a project.
Do one thing. Cover the hours you are currently unreachable.
Work out the actual gap first. Pull your phone records for the last three months. Count inbound calls outside business hours, then count how many were answered and how many converted to a matter. Most firms are shocked by the unanswered number, because nobody ever counts it. The call that rings out does not appear in any system. It is invisible loss, which is why it persists for years.
Then set up the AI receptionist on the after-hours line only. Business hours stay exactly as they are. One channel, one time window, one clear measure of whether it worked: answered calls outside hours before and after.
From there, the obvious extensions present themselves. Overflow during the day when reception is on another line. A second language if your client base needs it. Automated first-contact texts to confirm the callback time. Reminder sequences for court dates, which reduce failures to appear and the admin storm that follows one.
If you want a structured way to see what else in the practice is eating time, the task automation approach is the same one I use with any founder-led business: list every recurring task, score each one for whether a system can take it, and start at the top. For a defence practice, the list usually includes court date reminders, file opening, disbursement tracking, and client status updates, well before anything touching legal reasoning.
The Dormant File Question
Every established defence practice sits on a list of former clients and former enquirers. Several hundred at least, often thousands. The instinct, borrowed from other industries, is to run a reactivation campaign.
Be careful here. Criminal defence is not a repeat purchase you can advertise into. Texting a former client to ask if they need a lawyer again is, at best, tone deaf. At worst it is a complaint.
Where there is genuine value is in the enquiries that never converted. People who called, got a quote or a callback that came too late, and went elsewhere or self-represented. And in the referral layer: the accountants, immigration advisers, and family lawyers who send work your way when they remember you exist. That list is worth a system, and the messaging is professional rather than promotional.
For context on how reactivation works when the list is appropriate, one finance client had 319 dormant contacts the team had written off entirely. A structured multi-touch sequence recovered $49,000. The mechanic transfers. The messaging absolutely does not. Adapt it, or leave it alone.
What Changes When the Phone Is Always Answered
The direct gain is obvious: instructions you were previously losing. For a practice where a single matter can be worth several thousand dollars, recovering two or three a month from the after-hours gap changes the numbers meaningfully, and the running cost is nowhere near the cost of a part-time hire.
The less obvious gain is what it does to the partners. The on-call phone is a low-grade tax on every evening and weekend. You are never fully off. Once a system answers first and only escalates the genuine custody matters, the phone stops ringing for the things that could have waited, and the calls you do take are the ones worth taking.
That is the shift worth aiming for. Not replacing lawyers with software. Building a business with a brain of its own, so the practice keeps functioning when nobody is at a desk. The phone is where it starts because the phone is where the loss is most measurable, but it is the same principle applied to every recurring piece of the operation: capture what the business knows, let a system handle the repeatable parts, keep the judgment with the people qualified to exercise it.
Wrapping Up
AI for criminal defence lawyers is not about the technology being clever. It is about a frightened person in a car park at 2am getting a human-sounding answer instead of a voicemail beep, and your firm being the one that answers. Everything else- the intake automation, the triage, the court date reminders- follows from fixing that one gap.
Start with the hours you are unreachable. Keep the AI firmly on the intake side of the line and the lawyers firmly on the advice side. Get the privilege and data questions answered in writing before you sign anything. Then measure the one number that matters: calls answered outside business hours, before and after.
If you want to work out what that gap is actually costing your practice and whether an after-hours system stacks up in your matter mix, book a 30-minute discovery call. No pitch deck, just a straight conversation about where the calls are going. If you would rather look at the wider picture first, the post on after-hours call answering covers the general case, and AI agents for small business explains how these systems fit together.
Frequently Asked Questions
Can AI give legal advice to criminal defence clients?
No, and it should not be configured to try. An AI intake system takes contact details, the location of the person in custody, the charge if known, and any hearing date, then routes that to a lawyer. It should state clearly that it is an assistant and not a lawyer. Advice on charges, plea, bail, or merits stays with the admitted practitioner, without exception.
Does using AI to answer calls affect legal professional privilege?
Privilege attaches to communications for the dominant purpose of obtaining legal advice, and an intake call generally sits inside that relationship. The practical risk is confidentiality rather than privilege. Limit what the intake script collects, keep recordings in a known jurisdiction with access controls and a deletion schedule, and disclose the AI to the caller. Get the vendor’s data handling in writing before you go live.
How much of a criminal defence practice can realistically be automated?
Intake, triage, file opening, court date reminders, disbursement tracking, and client status updates are all fair game. That is usually a substantial share of the administrative load. Legal reasoning, advocacy, negotiation, conflict sign-off, and the decision to accept a matter stay human. The realistic target is removing the admin around the legal work, not the legal work itself.
Will clients be put off by talking to an AI at 2 am?
Some will, which is why disclosure and a fast human callback matter. In practice, a caller in distress cares most about being heard and being told when a lawyer will ring back. An assistant that answers on the second ring and promises a call within the hour beats a voicemail every time. Keep the script short, calm, and free of any attempt to sound human when asked directly.
What does an AI receptionist cost a small law firm?
Less than a part-time receptionist and considerably less than an after-hours roster paid in goodwill. The honest way to assess it is against the value of recovered matters. Count your unanswered after-hours calls over three months, apply your average matter value and a conservative conversion rate, and compare. For most defence practices, one or two recovered matters cover a year of running it.
How do conflict checks work when AI handles the first call?
The AI collects the names involved: the person in custody, the caller, and any complainant or co-accused mentioned. It can run an initial search against your existing client database and flag apparent matches. A lawyer reviews that result and makes the call on whether a conflict exists before anyone is engaged. The system speeds up the search; it never makes the determination.
How long does it take to set up AI call handling for a law firm?
A single after-hours line with a tight intake script, escalation rules, and a text or email handoff is typically a matter of days rather than months, once the script and the data handling are agreed. The slow part is not the build. It is deciding what the AI is allowed to ask and getting the confidentiality settings signed off internally.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.