AI for Process Serving Companies: Service Attempts and Affidavits

An AI agent can help a process serving company keep service attempts, timestamps and proof-of-service paperwork complete and easy to find, and it can draft a checklist for each affidavit that shows what is still missing. The judgment about whether an attempt or service meets the requirements of a specific court or jurisdiction stays with licensed staff, who review the facts and decide what the record supports.

Key takeaways

• Log each service attempt as a structured record at the moment it happens, not from memory at the end of the day.
• Let the agent assemble an affidavit checklist from the logged facts and flag gaps for a human reviewer.
• Keep legal sufficiency decisions with a licensed person on every job, without exception.
• Give the agent read access and draft-only write access, with no ability to sign, file or send documents for the firm.
• Start with one workflow, confirm the records are reliable, and expand only after staff trust the output.

Where an agent fits in a process serving job

A process serving job moves through a predictable sequence: receipt of the papers, research on the party and address, one or more service attempts, a decision to serve or to report non-service, preparation of the proof of service, and return to the client or the court. Most delays and most errors sit at the handoffs. An attempt is noted on a scrap of paper, a time is written without a time zone, or a photo sits on a phone that nobody else can open. An agent connected to the firm's job system, the server's field notes and the document store can keep those handoffs visible. It does not serve papers and it does not replace the server. Its role is to keep an accurate picture of what was done and what remains.

Logging service attempts as structured records

A service attempt is only useful later if the record captures facts in a consistent order. Free-text notes are hard to compare across attempts and easy to lose. A structured record lets a supervisor see at a glance whether the diligence on a file is complete. A practical record for each attempt includes:

• The job reference and the matter or court it belongs to
• The date and time of the attempt, with the time zone stated
• The address or location, and how it was confirmed
• The server's identifier and whether the attempt was made in person
• What the server observed, described in plain words, such as occupancy signs or who answered the door
• Who was spoken to, the stated relationship to the person being served, and how that was established
• The outcome, recorded as served, not served, or unable to serve, with the reason
• References to photos, location data or notes stored in the firm's own systems
• The next planned attempt and any instructions received from the client

The agent can prompt the server for missing fields while the attempt is fresh and can keep entries in the same order for every job. The rule is that the record describes observations and actions. Conclusions about sufficiency belong in a separate step.

Drafting an affidavit checklist without drafting the conclusion

An affidavit of service is a sworn statement, and the person who signs it must be able to stand behind every line. The agent's useful contribution is a checklist built from the logged facts. The checklist lists the fields the firm's approved template requires for the relevant jurisdiction, marks each one as complete, incomplete or inconsistent with another entry, and links each line to the record that supports it. The firm maintains the template and its requirements. The agent does not interpret the rules on its own.

Illustrative scenario: a server logs three attempts on one matter, at different times of day. The first two entries include the address confirmation and the observations. The second entry has no description of the person spoken to. The agent drafts the checklist, marks the description as missing, and places the second attempt in a review queue. A supervisor asks the server to complete the entry from notes. A licensed reviewer then decides whether the full record supports the return. The agent never writes the sufficiency conclusion, and no line of the affidavit is filed until a human has checked it against the source notes.

The NIST AI Risk Management Framework describes this kind of arrangement, in which people keep oversight of consequential outputs and the system is designed so that its limits are visible. A checklist that exposes gaps is a more honest tool than a draft that looks finished.

A decision guide: what the agent does and what stays with staff

The boundary between drafting and deciding should be written down before any system goes live. A workable split looks like this:

• Organize attempt records from field input: the agent may do this, and staff check any entry that changes an outcome.
• Draft a checklist from the firm's approved template: the agent may do this, and a licensed person approves the template before use.
• Flag missing or inconsistent entries: the agent may do this, and staff resolve each flag.
• Decide that service was valid or that diligence was sufficient: licensed staff only.
• Sign, swear to, notarize or file an affidavit: a human only, outside the agent's permissions.
• Contact a party, a court or a client: only through approved channels, with human approval for each message.

If a task is not on the first two lines, it should not be delegated without a new review of the boundary.

Permissions, records and the audit trail

The agent should operate under least privilege. It reads the job records it needs, writes only to draft worksheets, and cannot change an outcome that a server has already recorded. OWASP's authorization guidance is a useful reference for setting those checks, since access should be decided on each request rather than assumed from the fact that the agent is inside the firm's network.

Field notes and uploaded documents are also a risk. A note or a scanned page can contain text that looks like an instruction to the agent. The OWASP guidance on prompt injection prevention describes why such content should be treated as data to be read, not as commands to be followed. Every action the agent takes should be logged with the request, the records it read and the draft it produced, so that a reviewer can trace how any line of a checklist was reached. The same discipline appears in our guide on internal audit workpapers, where follow-up depends on being able to reconstruct who saw what and when.

How Be Human runs this as a managed service

Be Human starts with one defined role for the firm. The engagement sets the memory the agent needs, which is the firm's approved checklist templates, job conventions and jurisdiction notes maintained by staff. It sets the tools the agent may use, such as read access to the job system and draft-only access to the affidavit worksheet. It records the approval boundaries described above and a success criterion that the firm can observe, such as the number of returned records that were sent back for missing fields. Managed operations then cover monitoring of the workflow, recurring evaluations against sample jobs, permission reviews, maintenance of the integrations, and incident response when a record is wrong. Templates and boundaries change only through a controlled review that a licensed person approves.

Choosing the first workflow to hand to an agent

The best first workflow is narrow, frequent and easy to check. Attempt logging for a single office is usually a better start than a full end-to-end process, because the output can be compared against the original notes within a day. Choose a workflow where the cost of an error is visible and recoverable, and where staff already agree on what a complete record looks like. The Be Human approach to intake and deadline work for law firms follows the same logic: define the boundary first, then let the agent prepare what staff would otherwise assemble by hand. Expand only after the records have been reliable for a period the firm itself considers sufficient. Take the free Be Human AI diagnostic to identify the first task in your firm that is worth handing to an agent.

Frequently asked questions

Can an AI agent decide whether a service of process was valid? No. The agent organizes observations and flags gaps, but the decision about whether an attempt or service meets the legal requirements of a jurisdiction belongs to licensed staff. Those staff review the source notes, apply the firm's approved rules and sign off on the outcome. The agent's output is a draft for that review, never a determination.

Does the agent write the affidavit itself? It drafts a checklist and a worksheet from the logged facts, and it can prefill fields that come directly from the record. The server or the person authorized to swear the statement reviews every line and signs it personally. Nothing is signed, notarized or filed by the agent, and the firm keeps that step outside the agent's permissions.

What personal information should we protect in these records? Attempt logs often include addresses, descriptions of household members and notes about third parties, so access should be limited to the people who need it and retention should follow the firm's obligations. Privacy rules differ by country and state, and for operations in Brazil the ANPD publishes guidance on personal data. The firm's counsel should confirm how the records are stored, who can read them and how long they are kept.

Do we need to replace our current systems before starting? Usually not. A first engagement works best when the agent connects to the job records, field notes and document store the firm already uses, with read access and draft-only write access to a worksheet. Structured fields can be introduced gradually as staff adopt them. A change of core systems is a separate decision that should be made on its own merits, not as a prerequisite for a first useful role.

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