Skip to content
be-human.ai
← Back to the journal

AI by industry

Superintelligence Prep for Small Law Firms: Conflicts and Billing First

Before an agent handles intake at a small law firm, conflict checks and billing status must exist. Here is the build order and where legal advice stops.

Be Human typographic cover with the title: Superintelligence Prep for Small Law Firms: Conflicts and Billing First
Conceptual illustration created with AI · Be Human

Superintelligence prep for a small law firm starts at matter intake, the first point where a prospective client's names, adverse parties and fee terms enter your records. An agent can handle the routine collection and routing in that flow, but only after the firm has a conflict record it trusts and a billing-status record that shows what has been agreed, invoiced and paid. This article sets out that order and the point where the agent must hand the matter to a lawyer.

Key takeaways

  • Conflict checks must be complete and searchable before an agent touches a new inquiry.
  • Billing-status records should show engagement, retainer, invoice and payment state for each matter.
  • An agent can collect facts and route inquiries, but it should not clear conflicts, judge merit or give legal advice.
  • Every permission should be written down: what the agent may read, what it may write, and which decisions need a named attorney.
  • Start with one intake path and one measurable success criterion, then widen scope only after reviewing the logs.

Why intake is the right place to start

Intake is where a small firm's risk and revenue first become visible. Someone makes first contact, a decision is made about whether the firm can take the matter, a conflict search runs, engagement terms are set, and the matter is opened for billing. Each step depends on details that are easy to lose: the spelling of an adverse party, a former client who controls a related company, or a retainer that was discussed but never recorded as received.

Intake also repeats. A busy practice handles a steady stream of similar inquiries, and much of the work is collecting the same facts in the same order. That is exactly the kind of work an agent can carry well, which is why intake often comes early in a superintelligence prep plan. The risk is that an agent placed at this point will move quickly over data nobody has checked. Speed only helps once the records underneath are reliable. For a broader view of how family law practices structure this flow, see family law intake and court dates.

What must exist before an agent takes over intake

Before an agent is given any role at intake, the firm should be able to answer yes to each of these points:

  • Conflict records are searchable by party name, known aliases, related entities and former clients, and the search runs through a controlled step before any matter is opened.
  • Every inquiry has one opening record with a status field, such as inquiry, declined, conflict hold, engaged or closed, and status changes only through defined steps.
  • Fee arrangements are recorded as a type (hourly, flat, contingent or retainer) with the agreed terms linked to the matter, not kept only in an email thread.
  • Billing-status records show, per matter, whether the engagement letter is signed, whether a retainer is required and received, and the state of the last invoice and payment.
  • Trust or escrow handling is documented by someone qualified to describe it under your jurisdiction's rules, and the agent receives no access to those funds or their records.
  • Storage locations are named for the intake form, the engagement letter and the conflict result, along with the people allowed to read each one.
  • A named attorney owns each decision category, including who clears a conflict and who approves a change to fee terms.

If several of these are missing, the first project is record cleanup, not agent configuration. That work is less visible than a chat interface, but it determines whether the agent's output can be trusted.

A step-by-step build order for intake

The sequence below keeps the agent's scope small until the firm has evidence that routing is accurate.

  • Step one: map the current intake path on paper, from first contact to opened matter, and mark each point where information is written down and by whom.
  • Step two: clean the conflict list. Merge duplicate party records, add known aliases, and mark former clients. Expect this to take longer than the agent work itself.
  • Step three: define the billing-status fields and the allowed value for each. A status with no definition or no owner will confuse staff and the agent alike.
  • Step four: write the routing rules the agent may apply. For example, an inquiry that matches a party name goes to a conflict hold. The agent does not decide whether the match is a real conflict.
  • Step five: give the agent read access to the conflict and billing records, and write access only to intake notes and routing tickets. It should not be able to change matter status.
  • Step six: replay a set of past inquiries through the process in a test setting and compare the agent's routing with the decisions the firm actually made. Record every mismatch and fix the cause before going live.
  • Step seven: go live on one practice area or one inquiry channel, with a person reviewing each routing decision for an agreed period, then review the log before widening scope.

The replay in step six is the most useful test. It shows where the rules are unclear, which records are stale, and which kinds of inquiry the agent should never see.

Where the agent must stop: limits on legal advice

An intake agent can acknowledge an inquiry, summarize what the prospective client wrote, check whether a conflict hold is open, explain the firm's general process and draft a neutral acknowledgment for an attorney to approve. It must not say whether the prospective client has a claim, what deadline applies, how a contract or court rule reads, or whether a conflict can be waived. Those are professional judgments, and the firm's attorneys own them.

Use this decision guide when you write the routing rules:

  • If the inquiry asks about merit, deadlines, strategy or what to do next, route it to an attorney and send only the standard acknowledgment that no advice has been given.
  • If a conflict search returns any match, including a partial one, place the matter on conflict hold and route it to the conflict owner. The agent does not clear it.
  • If the prospective client may be adverse to a current client, or appears in a former-client record, route it to the responsible attorney without a substantive reply.
  • If the inquiry concerns money held for a client, a refund, or a fee dispute, route it to billing staff.
  • If the matter falls outside the firm's practice areas, send the approved decline message and log the reason.
  • If the inquiry mentions a court date or an imminent deadline, escalate through the agreed urgent channel at once.
  • If the sender's identity, capacity or authority to instruct the firm is unclear, stop and route to a person.

The limit should be built into the instructions and the tools, not left to the model's judgment. The agent should have no tool that sends a merit assessment, and its drafts should pass through an approval step before anything leaves the firm.

Permissions, approvals and the audit trail

Treat every intake message as untrusted input. A web form or email can contain text that tries to override the agent's instructions, so the firm's rules should live in the system configuration, and the content of an inquiry should be handled as data to be classified, not as instructions to follow. The OWASP guidance on LLM applications describes this class of risk in detail, and it is worth reviewing with whoever maintains your integrations.

Permissions should follow the principle of least privilege. The agent's connection to the conflict database should carry a read scope, its connection to billing should carry a narrower read scope, and its write actions should be limited to intake notes and routing tickets. Each connection should be reviewed when staff roles change, and revoked when a tool is retired. Authorization decisions should be checked on the server side, not only in the agent's prompt.

Every action needs a record: the time, the inquiry identifier, the records consulted, the routing decision, the rule applied, and any human approval that followed. This log serves two purposes. It lets the firm answer later questions about how a matter entered the system, and it gives you the evidence for recurring evaluations. A review each month or quarter should look at mismatches, escalations that were not needed, and any inquiry where the agent came close to giving advice. The NIST AI Risk Management Framework is a useful structure for deciding which of these checks to run and who owns them.

Illustrative scenario: a two-attorney firm receives a web inquiry in which the sender names a company as the opposing party. The agent searches the conflict list, finds that the company's former name appears on a closed matter, and places the inquiry on conflict hold without drafting any reply about the dispute. The responsible attorney reviews the closed file the next morning and decides how to proceed. The log shows the match, the hold, the rule applied and the time of the attorney's decision.

How Be Human approaches intake, and where to start

In a managed Be Human engagement, the first step is defining one useful intake role, the memory and tools it needs, the approval boundaries and one observable success criterion. The work then covers the record sources the agent depends on, read and write scopes, recurring evaluations against past inquiries, permission reviews as staff change, and incident response when routing goes wrong. Be Human operates this remotely, in English or Portuguese, and keeps attorney approval points where the firm has defined them. The firm keeps the legal judgment; the operating system keeps the records in order.

If you want to find which intake task in your firm is ready for an agent and which records still need work, take the free Be Human AI diagnostic to find the first task worth handing to an agent.

Frequently asked questions

Can an AI agent run conflict checks for a law firm? It can search the conflict records you define and report any possible match quickly, which helps the firm hold matters early. It should not clear, waive or assess a conflict, because that decision belongs to a responsible attorney under your jurisdiction's professional rules. The agent's role is to surface the match and keep the matter on hold until a person decides.

What billing information does an intake agent need? It needs recorded facts: the fee arrangement type, whether the engagement letter is signed, whether a retainer is required and received, and the state of the last invoice and payment. It should read these fields and report them accurately. It should not forecast collections or estimate fees, since those depend on judgment and on information outside the intake record.

Does an intake agent give legal advice? It should not, and its configuration should make that impossible in practice. It can acknowledge an inquiry, collect facts, explain the firm's general process and draft a neutral acknowledgment for approval. Questions about merit, deadlines, strategy or what a contract means go to an attorney, and the agent should have no tool that produces those answers.

How long does preparation take? The timing depends on how clean your existing records are, and no fixed duration applies to every firm. Practices with duplicate party entries, inconsistent status labels and unclear retainer records often spend most of their preparation time on those records before any agent configuration begins. A short diagnostic can show which of those gaps matter most for your intake path.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →