automated legal intakeWhere Automation Belongs

Automated legal intake should handle repeated preparation work while an authorized person controls conflicts, engagement, qualification standards, legal conclusions, filing, and advice. The safest design collects information, preserves each source, proposes structured facts, and routes uncertainty to a named reviewer. An unknown answer must remain different from a question that was never answered.

What automated legal intake can do well

Automation is most useful when the firm can describe the task, the expected input, and the correct exception. A system can present case-type questions, request documents, send reminders, check for likely duplicates, classify uploads, propose values from a document, and assign work to a queue. These are preparation steps. Their output should be reviewable rather than treated as a legal decision.

Start with a recurring task whose failure creates review work instead of irreversible action. A reminder that cannot be delivered can enter an exception queue. A document that cannot be classified can wait for staff. A proposed incident date that conflicts with the claimant's answer can be marked for comparison. Each outcome leaves the firm in control.

The first useful workflow often covers four jobs:

1. Ask and collect. Present the configured questions and request the records expected for that case type.
2. Prepare the upload. Classify the file, propose relevant facts, and keep the page or passage behind each proposal.
3. Route the exception. Send missing answers, contradictory dates, duplicate records, and low-confidence values to the right person.
4. Continue follow-up. Remind a prospect about promised information while retaining the request, response, and delivery history with the intake record.

That scope can remove calendar chasing and repeated data entry without allowing software to decide whether the firm should represent someone.

Build around states, owners, and stop points

A working intake process needs explicit states. "New," "awaiting documents," "ready for staff review," and "ready for attorney review" give staff clearer direction than one open-ended status called "in progress." Each state should identify an owner, the evidence required to leave it, and the conditions that stop automatic movement.

Write the stop points before configuring automation. Common stop conditions include a possible conflict, a changed identity, a missing consent record, contradictory event dates, an unreadable upload, an unsupported case type, or a request for legal advice. The system should retain the intake and send the issue to an authorized person. It should not invent a value to satisfy a required field.

This is also where the firm distinguishes three different conditions:

• Answered: the prospect or a source supplied a value.
• Unknown: the person responded and does not know the value.
• Unanswered: the question has not received a response.

Those conditions lead to different work. An unanswered question may justify a reminder. An unknown date may require a record request or attorney review. Treating both as blank produces inaccurate completion reports and can send the wrong follow-up.

Keep every proposed fact attached to its source

An extracted value is a proposal until the firm confirms it. Intake may contain a questionnaire answer, a call transcript, a medical record, a calculation, and a later correction. Store those as separate sources. Do not overwrite all of them with the latest value.

For an important proposed fact, a reviewer should be able to see:

• the proposed value
• the source type and location
• the time it entered the record
• any confidence or parsing warning
• competing values from other sources
• the person who confirmed or corrected it
• the reason for the correction when the change matters

This record gives staff enough context to resolve two different incident dates without guessing which value displaced the other. It also supports later quality review. If one document type produces frequent corrections, the firm can inspect that specific step.

The NIST AI Risk Management Framework Core organizes risk work around governance, context mapping, measurement, and management. Applied to intake, that means the firm defines the permitted use, tests the workflow with representative records, measures corrections and exceptions, and changes the configuration when results show a weakness.

Legal judgment stays with authorized people

Software can prepare a file for a decision. It cannot receive authority simply because every required field contains a value. Conflict resolution, engagement, case qualification, legal conclusions, filing choices, and advice remain with the lawyer or authorized firm staff under the firm's rules.

The ABA's comment to Model Rule 5.3 explains that lawyers with managerial or supervisory authority must take reasonable measures to make nonlawyer assistance compatible with professional obligations. Intake software should support that supervision with assigned owners, visible review history, and enforced approval gates.

AI-assisted extraction or summarization requires another layer of review. ABA Formal Opinion 512 discusses duties that can arise when lawyers use generative AI, including competence, confidentiality, communication, supervision, and review of output. A firm applying those principles to intake should define who may use the tool, what information may enter it, and who checks a proposed fact before it affects a legal decision.

Protect prospective-client information

Intake begins before the firm has necessarily accepted a representation, but it can still collect sensitive personal, medical, financial, or dispute information. Limit collection to what the current stage requires. Restrict access by role. Record exports and material changes. Set retention rules for declined and incomplete inquiries.

For client information, ABA Model Rule 1.6 includes a duty to make reasonable efforts to prevent unauthorized access or disclosure. Firms should evaluate how an intake provider stores data, controls access, handles subprocessors, responds to incidents, and returns or deletes information. Applicable state rules, privacy laws, court orders, and client commitments may require additional controls.

Test the exception queue before expanding scope

Choose one repeatable case type and build a test set from the inputs that consume staff time. Include an incomplete questionnaire, a duplicate inquiry, a blurry scan, a file in the wrong category, a changed answer, and two sources with different dates. Use synthetic or properly authorized records for testing.

Measure what reaches the reviewer:

Measure	What it can reveal
Correction rate	Fields or source types that produce unreliable proposals
Missing-evidence rate	Requests or stage rules that do not match the case type
Duplicate-review rate	Identity rules that are too broad or too narrow
Exception age	Queues without clear ownership or service expectations
Unauthorized advancement	A control failure that should stop expansion
Review time	Whether preparation gives the reviewer usable context

Review the correction queue before adding another task. If staff regularly reconstruct the source trail, the workflow has moved work rather than prepared it. If exceptions remain open because no one owns them, adding more automatic routing will increase the backlog.

A practical rollout sequence

Document the current handoff first. Name the person who receives a new inquiry, the information needed at each stage, the legal review gates, and the systems that hold documents or notes. Then configure the smallest complete path from first contact to review-ready intake.

Run that path with representative exceptions. Confirm that staff can open every source, correct a proposal without deleting the original, and see why a record stopped. Train reviewers on both the expected workflow and the escalation path. Only then should the firm expand to another case type or automate another task.

The result should be a more inspectable intake record. Software performs repeated preparation. People receive the source material, the uncertainty, and the decision at the right point.

Book a demo →