AI Automation for Law Firms: Intake Without Losing Control

AI automation for law firms is an intake and scheduling relay: it answers or texts back on a new inquiry, captures the facts needed for a conflicts check, flags language that sounds like a deadline without calculating one, and offers a consultation time, then stops. Deciding whether the firm can take the matter, what the case is worth, and whether representation has begun stays with an attorney every time, without exception.

TaskChad sells AI automation and implementation services to law firms and other service businesses, and this page is a description of what we build and sell, not an independent evaluator's assessment of legal intake technology. Nothing below reports a real firm's intake volume, acceptance rate, or fees, and this page is not legal advice for any specific firm or jurisdiction.

What this system does, and what it refuses to do

An intake automation for a firm has one job: get the caller's information into a structured form fast enough that an attorney can act on it, without the system pretending to be the attorney in the meantime. It does not tell a caller whether they have a case. It does not estimate what a claim is worth. It does not calculate a filing deadline from a date the caller gives it. It captures what the caller says, in the caller's own words where it matters, flags anything that sounds urgent, and routes the whole package to a person for a real decision. Firms that skip this boundary end up with a system that sounds confident about things it has no authority to be confident about.

Intake fields, kept inside conflicts-safe boundaries

Field	Captured how	Boundary
Caller and opposing party names	Asked directly for a conflicts check only	The system flags a potential match; it does not clear or resolve a conflict itself
Matter type, general category	Caller's own description mapped to a broad category like "employment," "personal injury," "family"	A category, not a legal characterization of the claim
Jurisdiction and county	Asked directly	Needed to route to the right intake team, not to assess venue strategy
Any date the caller mentions as urgent	Recorded verbatim, timestamped	Never converted into a calculated deadline by the system
Referral source and contact details	Asked directly	Standard intake and marketing attribution, nothing more

The state model: from first contact to attorney acceptance

• INTAKE_STARTED: a new inquiry arrives by phone, text, or web form and is recorded with a timestamp and channel.
• CONFLICTS_SCREEN_PENDING: the caller and opposing party names are checked against the firm's existing conflicts list.
• CONFLICT_FLAGGED: a possible match surfaces; the file is held and routed to whoever runs conflicts clearance before anything else proceeds.
• CONSULT_OFFERED: with no conflict flagged, the system offers real, open consultation windows.
• CONSULT_SCHEDULED: the caller picks a time, written to the calendar with a confirmation sent.
• ATTORNEY_REVIEW_PENDING: the completed intake file, including any urgency flag, sits with an attorney awaiting a decision.
• MATTER_ACCEPTED or MATTER_DECLINED: an attorney, not the system, makes this call.
• CLOSED: the file reaches a final state through acceptance, decline, or the caller withdrawing.

Conflicts screening is capture, not clearance

The most important design boundary in this whole workflow is the difference between flagging a possible conflict and clearing one. The automation compares names it captures against a list the firm maintains and surfaces a match or a near-match. It does not decide that a name is close enough to ignore, and it does not decide that a matched name is unrelated to the current matter. Both of those are judgment calls that belong to whoever the firm has authorized to clear conflicts, and a system that quietly waves through a near-match because the spelling was slightly different is a liability the firm will not discover until it is expensive to fix.

Deduplication: the same inquiry should not open two files

A caller who fills out a web form and then calls the same day describing the same problem should not generate two separate intake files fighting for two different attorneys' attention. The dedup check matches on caller name, phone number, and a short hash of the matter description within a defined window, and when a match is found, the new contact updates the existing file instead of opening a second one. This matters for more than tidiness: two open files for one matter can mean two different intake coordinators quoting different next steps to the same caller, which looks disorganized at exactly the moment a prospective client is deciding whether to trust the firm.

Two clocks that matter: consult holds and attorney response time

A held consultation slot needs an expiration, long enough that a caller finishing a normal conversation does not lose the time they were actively picking, short enough that an abandoned booking attempt does not block that slot from someone else. Separately, an ATTORNEY_REVIEW_PENDING file needs its own service-level clock: if a file with an urgency flag sits unreviewed past a defined number of hours, it should page the intake coordinator directly rather than waiting silently for an attorney to open their queue. These are two different problems. A scheduling system that only manages the first one will still let a genuinely urgent file sit unnoticed over a weekend.

Audit trail an intake coordinator can actually use

Every file should carry a timestamped record of the conflicts screen result and who cleared any flag, the urgency language captured verbatim, the consult offer and booking event, and the attorney's acceptance or decline with a timestamp. This is the record that lets a managing partner answer a specific question later, such as how long a flagged file actually sat before someone looked at it, rather than relying on someone's memory of a busy week.

The queue also needs a named owner for each status, not merely a shared inbox. The intake coordinator owns incomplete contact records and consultation logistics. A designated conflicts reviewer owns every CONFLICT_FLAGGED file. The assigned attorney owns ATTORNEY_REVIEW_PENDING and is the only role allowed to select MATTER_ACCEPTED or MATTER_DECLINED. If a person is unavailable, an approved backup role should be explicit in the routing table. That ownership map prevents the automation from advancing a file simply because a timer expired, and it gives the firm a concrete place to send an aging alert without broadcasting sensitive intake details to staff who do not need them.

What always requires an attorney

Legal advice of any kind never comes from this system, including an answer to "do I have a case." Merit assessment never comes from this system. Deadline calculation never comes from this system, because a captured date and a calculated statute of limitations are not the same thing, and getting that wrong has real consequences for a real person's claim. Fee discussion beyond a firm's publicly stated, pre-approved general policy never comes from this system. And no confidentiality assurance is made to a caller before the firm's own engagement process has actually established a relationship, since promising confidentiality prematurely can create expectations the firm has not agreed to.

ABA Formal Opinion 512 is ethics guidance, and this is not legal advice

The American Bar Association's Formal Opinion 512 addresses a lawyer's ethical duties when using generative AI tools, including competence, confidentiality, communication with clients, and reasonable diligence (ABA, Formal Opinion 512). The opinion and NIST material cited below were reviewed on August 13, 2026. It is professional ethics guidance from a national bar association, not a statute binding every jurisdiction, and individual state bars can adopt, adapt, or diverge from it. Whether and how it applies to a specific firm's intake automation depends on that firm's own jurisdiction and licensing rules. This page is not legal advice, and a firm should have its own bar counsel review how any AI-assisted intake process fits its actual ethical obligations before relying on it.

NIST's framework, used as a checklist and nothing more

The NIST AI Risk Management Framework is voluntary guidance organized around four functions, Govern, Map, Measure, and Manage, meant to help an organization reason through an AI system's risks across its lifecycle (NIST, AI Risk Management Framework). It is not a law, not a certification, and not proof that any intake system is safe or accurate. Used here, it is a prompt to ask who owns the conflicts-screen rules, what happens when a near-match gets missed, how the firm would find out, and who fixes it.

A hypothetical intake, worked through

Consider a hypothetical employment-law inquiry: a caller describes being terminated and names a former employer. The name check against the firm's existing conflicts list comes back clear, and the system offers three real consultation windows this week; the caller books one and a confirmation goes out. A second hypothetical caller mentions a deadline "coming up in a couple weeks" tied to a filing they read about online. The system records that phrase exactly as spoken, timestamps it, and routes the file to ATTORNEY_REVIEW_PENDING with an urgency flag attached rather than telling the caller anything about what that deadline actually is or when it falls. Neither example is a real case outcome. Both show the rule structure only.

Failure-path tests before this goes live

Before real inquiries reach this system, test a caller name that closely resembles, but does not exactly match, an existing conflicts entry, confirming it flags for a person rather than clearing automatically. Test two contact channels used for the same matter within a short window, confirming the dedup rule merges them into one file. Test a caller who mentions an urgent-sounding date, confirming the system records it verbatim without attempting to calculate anything. Test a held consultation slot that expires without a completed booking, confirming it releases back to availability rather than blocking it indefinitely. Test an ATTORNEY_REVIEW_PENDING file that sits past the defined response window, confirming it actually pages the intake coordinator.

What to measure in the first 30 days

Track inquiries captured by channel, the conflicts-flag rate, consultations booked versus completed, the average time a file sits in ATTORNEY_REVIEW_PENDING, and the acceptance-versus-decline split. At 30 days, review flagged files specifically, since that is where a screening rule is most likely to be too loose or too strict compared to how real callers actually spell and describe names. This shows whether the rules match reality. It does not, by itself, prove the automation changed how many matters the firm accepted, which needs a fixed, agreed-in-advance before-and-after comparison rather than a one-time review.

Where this fits with the rest of your intake

AI lead qualification workflow covers the general qualifying pattern this intake model builds on, and bilingual lead intake automation and voicemail-to-CRM automation cover two of the channels a new inquiry commonly arrives through. The law firm AI receptionist page covers the voice product this workflow can run on top of, and our law firm Reddit roundup covers what attorneys are actually asking in public forums before they buy anything. Speed-to-Lead covers the broader response-time system this fits inside, the receptionist page covers live call handling directly, and Marketing Automation covers what happens to a contact once a matter closes.

The bottom line

A law firm does not need an intake system that sounds like a lawyer. It needs one that captures a clean file fast, screens for conflicts honestly, flags urgency without pretending to calculate it, and gets a real attorney looking at the right file quickly. Everything that requires legal judgment stays exactly where it already belongs.

If you want a clear picture of where your own intake process is actually losing inquiries or delaying attorney review, run the Revenue Leak Score. The score runs on the page without booking and returns a ranked starting point before you decide what to fix.