A law firm evaluating Zanus AI should define a narrow administrative intake before testing real enquiries. Record how a caller reaches the appropriate person without allowing the assistant to interpret a legal issue, accept an engagement or promise an outcome. Suitability and professional obligations need the firm’s own responsible review.
Let the firm approve the scope
In the United States, the ABA’s explanation of Formal Opinion 512 highlights professional duties relevant to generative AI, including competence and confidentiality. It is not a vendor approval or a substitute for the rules applicable to a particular lawyer.
Have the firm’s responsible lawyers assess those obligations and approve the proposed tool, contract and script. Keep a real-data trial on hold until the required review is complete.
Write the minimum useful intake
| Administrative item | Boundary to approve |
|---|---|
| Caller contact details | How the firm may return the enquiry |
| Existing matter reference | How staff identify the right team without inappropriate disclosure |
| Broad enquiry purpose | Which limited questions the script may ask |
| Requested callback | Who owns it and what response window may be stated |
The firm should decide what information should not be collected through this channel. More detail is not automatically better intake.
Test statements outside the boundary
Use internal scenarios in which a caller asks whether they have a case, whether a deadline applies or whether the firm has agreed to act. Check that the response follows the firm’s approved escalation wording and creates an appropriate staff task.
Test a repeat caller and an uncertain identity match. Ask which matter information could be retrieved and whether that access matches the approved process. Do not treat a phone-number match as sufficient authority to discuss a matter.
Review the handoff with the lawyers who receive it
Give the responsible team the resulting test records and ask whether they contain enough permitted information to respond. Track missing context, excess information, incorrect promises and callbacks without an owner.
Keep the pilot’s success measure administrative: accurately routed enquiries and fulfilled response promises. Do not count an intake as a legal assessment or established client relationship. Maintain a way to suspend the workflow if it strays beyond the approved scope.
Questions and answers
Should an intake assistant decide whether the firm can take a case?
Keep that decision within the firm’s authorised professional process. The assistant should use only the approved administrative script and handoff.
Designing the Administrative Script Before Any Live Call
Once a firm has defined the outer boundary of what an AI intake assistant may and may not do, the practical challenge becomes translating that boundary into a working script that staff can actually review, rehearse and correct. A script written only at a high level leaves too much room for the assistant to fill gaps with responses the firm has never seen. The goal of this section is to help a firm think through that design process in concrete operational terms, covering the shape of individual exchanges, how responsibility passes from one step to the next, and what observable evidence would tell a reviewer that the workflow is staying within bounds.
The design work is worth doing on paper before any technology is configured. A written draft forces the firm's responsible lawyers and administrative staff to agree on wording rather than assuming agreement. It also produces a document that can be updated when circumstances change, which is a more durable basis for oversight than an informal understanding of what the assistant is supposed to do.
Mapping Each Exchange as a Discrete Step
A useful way to draft the script is to treat each exchange as a distinct step with a defined input, a defined output and a named owner for what happens next. Consider an illustrative scenario: a caller phones the firm and the AI assistant answers. Step one is a greeting that identifies the firm and the nature of the channel without characterising the assistant in a way that could mislead a caller about whether they are speaking to a human professional. Step two is collection of the caller's name and preferred contact number, with a statement that the firm will use that information only to return the enquiry. Step three is a single open question about the broad purpose of the call, worded in a way that does not invite the caller to provide detailed facts. Step four is a confirmation of the callback window the firm has approved, attributed clearly to administrative process rather than to any professional commitment.
Each of those steps should have an exit condition. If the caller provides information that falls outside the approved scope at any step, the script should route to a holding response and create a staff task, rather than attempting to respond substantively. Writing that exit condition explicitly, for every step, is the operational discipline that prevents the assistant from drifting into territory the firm has not sanctioned.
Wording That Does Not Imply a Legal Assessment
Certain phrases that sound routine in ordinary conversation carry specific weight in a legal context. Phrases such as "we can look into that for you," "that sounds like something we handle," or "you're in the right place" can reasonably be interpreted by a caller as an indication that the firm is considering or has accepted their matter. A firm evaluating any AI intake tool should review its proposed script specifically for phrases of that kind, because they are easy to introduce accidentally and difficult to retract once a caller has relied on them.
A useful exercise is to give the draft script to a member of staff who has had no involvement in drafting it, and ask them to read it as if they were a caller in a stressful situation. Ask them to note any phrase that left them with an impression about the firm's willingness or ability to help with their specific problem. That impression test is a practical check that sits alongside, rather than replacing, the responsible lawyers' professional review.
Where a caller asks a question that falls outside the administrative scope, the response should be a clear, neutral redirect rather than a silence or an abrupt termination. A response along the lines of "I'm not able to help with that through this channel, and one of our team will contact you to discuss it further" sets a clear boundary without leaving the caller without any next step. The firm should agree that wording in advance and treat any deviation from it as an event requiring review.
Tracing Responsibility Through the Handoff
The intake process does not end when the AI assistant completes a call. The moment a task is created and passed to a member of staff, a second set of responsibilities begins, and those responsibilities are worth mapping with the same care as the intake script itself.
What a Task Record Should Contain
A reviewable task record created by the intake process should contain the caller's name and contact detail, the time of the call, the broad category of enquiry as captured by the approved script, the callback window that was stated during the call, and the name of the person or team to whom the task has been assigned. It should not contain a summary of the caller's legal problem written by the assistant, an assessment of whether the enquiry falls within the firm's practice areas, or any statement that could be read as a preliminary view on the merits.
If the task record contains more than the firm has approved, that is a signal to review the script, not merely the record. Excess information in a task record indicates that the script allowed the caller to provide it, which means the boundary was not working as intended at the collection stage.
Assigning a Named Owner Before the Call Ends
A callback promise made to a caller without a named owner is an operational liability. The caller has a reasonable expectation that someone will call within the stated window. If no individual or team has accepted responsibility for that task before the call ends, the promise depends on a process that has not been confirmed. Firms should decide in advance which role or team receives tasks from the intake channel, confirm that the recipient has agreed to the response window, and build a review step that checks for tasks that have passed their window without a logged response.
In smaller firms, the assignment may be to a specific person by name. In larger firms, it may be to a team with an internal allocation process. Either arrangement is workable, but the arrangement should be documented, and the person who designed the intake workflow should confirm it with the people who will actually receive the tasks, not assume they are aware of it.
Handling a Task Where Identity Is Uncertain
A recurring operational challenge in legal intake is the caller whose identity cannot be confirmed with confidence. A phone number may be associated with more than one contact in the firm's records. A caller may give a name that is similar but not identical to an existing record. A caller may be calling on behalf of a person or organisation rather than as the primary contact.
The firm should decide, before live use, what the intake process does in each of those situations. The safest administrative approach is to treat an uncertain identity match as an unmatched caller and create a new task without reference to any existing matter. That approach errs toward caution rather than toward efficiency, but it avoids the more serious problem of associating an intake record with the wrong matter. The responsible staff member who follows up the callback can confirm identity through the firm's approved verification process before any matter-specific discussion takes place.
A phone-number match alone is not a basis for disclosing matter information. That principle should be stated explicitly in the intake policy, because the temptation to use a match as a shortcut will arise in practice, particularly under time pressure.
Running a Structured Acceptance Exercise
Before moving from internal testing to any live or pilot use, the firm should conduct a structured acceptance exercise with observable evidence rather than informal sign-off. The purpose is to produce a record that the responsible lawyers have seen the workflow in operation and that specific criteria were met.
Scenario Categories to Cover
A useful acceptance exercise covers at least four categories of scenario. The first is a straightforward administrative enquiry: a caller provides their name, contact detail and a general description of the purpose of their call, and requests a callback. The observable evidence is a correctly formed task record assigned to the right owner within the firm's approved process.
The second category is an out-of-scope enquiry: a caller asks a question that requires a legal assessment, such as whether a particular time limit applies to their situation, or whether the firm has agreed to act for them. The observable evidence is a redirect response using the approved wording and a task record that does not contain any response to the legal question.
The third category is an uncertain identity scenario: a caller's details partially match an existing record. The observable evidence is that the task is created without reference to the existing matter and that no matter-specific information is disclosed or recorded.
The fourth category is a distressed or persistent caller: a caller who provides more information than the script requests, or who pushes back against the redirect. The observable evidence is that the script does not expand its scope in response to the pressure and that the task record reflects only what the approved script was designed to capture.
Recording What the Lawyers Who Receive Tasks Actually Find
The lawyers or staff who will receive and act on tasks from the intake process should be included in the acceptance review, not just the people who designed the workflow. Their practical question is whether the task record gives them what they need to make a return call that is useful to the caller and consistent with the firm's professional obligations. If the record contains too little information for them to route the call appropriately, the script may need adjustment. If it contains more than the firm has approved, the script needs to be tightened before live use.
This review is worth conducting as a meeting rather than an email exchange, because the discussion often surfaces assumptions that neither side knew the other was making. The intake designer may assume that a particular field is self-explanatory; the receiving lawyer may read it differently. That kind of misalignment is much easier to resolve before live use than after.
Maintaining Oversight After Pilot Launch
Acceptance testing covers the workflow as it was designed. Ongoing oversight covers the workflow as it actually operates when real callers interact with it. The two are not the same, and a firm that treats a passed acceptance exercise as a permanent clearance is not exercising adequate oversight.
What to Review and How Often
The firm should agree on a review cadence before launch. A useful starting point is a weekly review of a sample of task records, focusing on whether the records fall within the approved scope and whether callbacks were completed within the stated window. The review should be carried out by someone with enough professional context to recognise a record that has drifted out of scope, not only by the person responsible for the intake workflow.
The review should also include a check for callers who have not received a callback within the stated window. That check is as much an indicator of operational health as any review of the script, because a missed callback is a concrete failure of the promise the intake process made to a real person.
Criteria for Suspending the Workflow
The firm should agree in writing on the criteria that would lead it to suspend the intake workflow, and the person who has authority to make that decision. Criteria might include a task record that contains a statement that could be read as a professional commitment, a pattern of callers who report confusion about the nature of the channel, or a missed callback rate that exceeds what the responsible team considers acceptable. None of those criteria require a technology failure; they reflect the possibility that the workflow is operating as designed but producing outcomes that fall outside what the firm intended.
Having the suspension criteria in writing before launch means the decision to suspend, if it arises, is a matter of applying agreed criteria rather than a judgment call under pressure. That reduces the likelihood that the workflow continues operating after a signal that it should stop.
A firm that approaches AI-assisted intake with this level of operational structure is not slowing down its evaluation; it is building the kind of documented process that will allow it to extend, adjust or suspend the workflow with confidence as its needs and the relevant professional landscape continue to develop.
Sources and next steps
This independent Sysevo guide proposes evaluation steps and reports no hands-on vendor test or compliance certification. Check the linked primary sources and Zanus AI’s current Front Office overview, then confirm the scope of your proposed deployment in writing.
Read Zanus AI call center human handoff for related context. To compare providers using the same requirements, explore Sysevo AI voice agents or book a Sysevo demo. Sysevo publishes this guide and offers its own business software.
Related workflow guides
- Zanus AI and GDPR: Map Call Data Before a Real-Data Trial
- Zanus AI Call Center Staffing: Divide AI and Operator Work
For current Sysevo subscriptions and usage options, compare plans and pricing.

