It should turn every order into one job record, assign it to a server, log each attempt with time, place and notes, and produce the proof of service for the server to review and sign. AI can read incoming papers and draft summaries; it must not write an attempt that never happened or sign anything.
Of the software I read, only Process Server's Toolbox describes AI on its own pages, with a document-reading assistant it calls AiMY. ServeManager and Process Organizer showed none, and Proof, a service that serves papers through its own network rather than software you run, describes an autofill feature for uploaded documents. I could not test how well any of these features perform, so the claims are the vendors', and each tool should be checked on its own page before you rely on it, because features and prices move.
Nine jobs the software must do, in the order a serve moves
A tool earns its fee by handling the whole path from order to invoice with one record per job. Here is the path, with what to test for each step in a demo. Run every test with your own paperwork, not the vendor's sample.
| # | Job | What good looks like | Demo test |
|---|---|---|---|
| 1 | Intake | An order by email, PDF or web form becomes a job with parties, case number, court, documents, deadline and special notes | Send a real order and time how long a clean record takes |
| 2 | Duplicate check | The same defendant or address already in progress is flagged | Enter a duplicate and see if anything warns you |
| 3 | Assignment | You pick a server by area, load and rush level, and see acceptance | Assign to an outside contractor as well as a staff server |
| 4 | Routing and batching | Addresses group by area and time window | Load ten addresses and check the map and order |
| 5 | Attempt logging | Each attempt records time, GPS, photo, outcome and notes from the server's phone | Log one attempt with no signal, if you work in poor coverage |
| 6 | Audit trail | Edits show who changed what and when; attempt times cannot be altered silently | Edit an entry and look for the history |
| 7 | Proof of service | A template for your court merges job, case and service details, ready for the server to review and sign | Produce one with your court's wording |
| 8 | Client updates | Clients see status, upload papers and download finished proofs without calling | Log in as a test client and try to see another client's job |
| 9 | Billing and server pay | Invoices include rush fees and mileage; contractor pay is calculated; the ledger syncs | Invoice one job and calculate one server's pay |
Steps 5 and 6 matter most. The attempt log is the evidence behind everything the server later swears, so a tool that lets anyone tidy an entry afterwards without a trace is a risk, however good its other features look. For dispatch and routing in general, AI job scheduling and dispatch for field service teams covers the logic; the tools below handle the legal-specific parts.
What a clean job record holds
Most downstream problems start with a thin job record. Whether the order arrives by email, a client portal or an AI reading of the papers, check that the record holds these fields before dispatch.
| Field | Why it matters | A common slip |
|---|---|---|
| Party to be served, exactly as written, plus known aliases | The server must serve the right person, and the proof must match the papers | A shortened name that no longer matches the document |
| Address for service, with unit, access notes and other known addresses | Drives routing, and shows where attempts were made | A missing unit number that wastes a trip |
| Documents to serve, listed by title and page count | The proof must say what was handed over | A packet of 40 pages with no list |
| Case number and court, as printed | Needed for the proof template and the filing | A court name typed from memory |
| Hearing date or response period, as written | Sets the rush level, but never calculate a legal deadline in the field | A date entered in the wrong format |
| Service instructions and safety notes | Tells the server about hours, hazards and any limits set by the client | Instructions left in an email instead of the record |
| Client and billing reference | Lets the invoice and the status message find the right account | Two clients with near-identical names |
| Rush level and any price agreed | Controls dispatch order and the invoice | A rush fee agreed by phone and never recorded |
Make the fields required where you can, so a job cannot be dispatched with a blank address or a missing document list. Where the software reads papers for you, check the fields it fills first and treat every unfilled field as a task for a person.
When the job goes to an outside server, send a dispatch packet in one message: the job record, the papers, the instructions and the evidence you expect, such as a photo of the door and a note of any person who answers. Ask for the attempt log back through the software, not in a text message, so that the evidence sits in one place with a system timestamp. If a tool makes outside servers email you photos or call in results, you have added a re-keying step that the software was supposed to remove.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowFour tools read against that list
I read each vendor's own pages in November 2026 and recorded only what they state. "Not stated" means the page I read was silent, not that the feature is missing. Proof is a different kind of product: it serves papers through its own network of servers, so it is a service you could buy or subcontract to, not software to run your firm.
| Tool | Intake and dispatch | Attempts and proof | Client access | AI on its pages | Prices shown |
|---|---|---|---|---|---|
| ServeManager | Law firms order through an InfoTrack integration; agencies enter jobs or take them through an order form (one-time setup fee); assign jobs to other servers; map and route jobs | GPS-logged attempts, photos and electronic signatures in the app; affidavits and proofs from templates that merge job, case and service details; custom templates built on request | Invite clients to enter serves, upload documents and view status | None mentioned | Agencies $59 a month (75 jobs), $119 (150), $219 (300), 500 or more with details on sign-up; law firms per order with no subscription; free ServerHub account for individual servers |
| Process Server's Toolbox | Quick job entry from saved contacts, client integration or custom online forms; automated routing in the app; job trading with other servers | Affidavits generated and signed in the app; the mobile app works offline | White-label client dashboard | "Ask AiMY" reads documents to pull out case and job details; the vendor claims it can cut job entry time by 50% or more | $40 a month with 50 jobs, then per-job tiers; unlimited users and active servers; 30-day free trial |
| Process Organizer | Clients submit work requests through the portal; automated job assignments to servers | Automated affidavits and field reports for each attempt; servers record serves and attempts in real time on a mobile website, with native apps | Portal to check status, submit work and download completed affidavits | None mentioned | None shown |
| Proof (a service) | Upload documents; "Autofill" pulls names, addresses and court details into the serve request for you to review; a vetted server from its network takes the job | Geolocation and photos for each attempt, added as exhibits; Proof's team drafts the affidavit, and the server certifies it or signs a notarised version, depending on local rules | Real-time updates and chat with the server | Autofill for reading uploaded papers; affidavits pre-populated from serve details and reviewed by its in-house team | A fee per serve that varies by speed and address; the price card I was shown ran from $50 for posting to $95 standard (4 attempts within 10 days) and $225 same day |
Read the table for gaps, not winners. Process Organizer lists automated job assignment but says nothing about routing, and the Process Server's Toolbox page I read did not describe a separate attempt-logging feature, only attempts mentioned in a customer comment. Neither gap proves the feature is absent, but each is a demo question. ServeManager's pages listed no AI at all, which is a fact about the pages I read and not a statement about its roadmap.
A 140-job month priced on two vendors' published plans
Two vendors publish prices you can compute. ServeManager sells agency plans by job count. Process Server's Toolbox charges $40 a month including 50 jobs, then $0.90 a job for jobs 51 to 100, $0.40 for jobs 101 to 500, $0.20 for jobs 501 to 1,000 and $0.15 for jobs 1,001 to 2,000. I have assumed the tiers stack the way the page's layout suggests, so confirm that with the vendor.
| Jobs a month | ServeManager agency plan | Process Server's Toolbox working | Process Server's Toolbox total |
|---|---|---|---|
| 75 | $59 | $40 + 25 × $0.90 | $62.50 |
| 140 | $119 (the 150-job plan) | $40 + 50 × $0.90 + 40 × $0.40 = $40 + $45 + $16 | $101 |
| 300 | $219 | $40 + $45 + 200 × $0.40 = $40 + $45 + $80 | $165 |
The flat plan wins at 75 jobs by $3.50, and the per-job tiers win at 140 and 300. Neither number is the whole bill. ServeManager's website order form has a one-time setup fee, and custom affidavit templates are built on request, so ask whether they cost extra; for Process Server's Toolbox, the pricing page does not mention AiMY, so ask whether it is included in the $40. Add payment-processing fees, server pay and any integrations you need.
What an attempt log needs so the proof holds up
Whatever the software, the quality of the proof depends on what the server records at the door. Each attempt needs a system timestamp, GPS position, the address as visited, a short factual description of what the server did and saw, a photo where useful, and an outcome code. Record observations, not conclusions: "no answer after three knocks, lights on upstairs" is evidence, while "obviously avoiding service" is an opinion. What a court requires in the record varies, so ask your lawyer or the court's guidance for the wording it expects.
A filled-in log for an illustrative job shows the difference between thin and useful entries.
| Attempt | Day and time | GPS and photo | Entry |
|---|---|---|---|
| 1 | Tuesday 6:40 p.m. | Matches address; photo of front door | Knocked three times and rang the bell. No answer. Car on the driveway, lights on downstairs. |
| 2 | Thursday 7:55 a.m. | Matches address; photo of door and car | Rang twice. No answer. Same car, bonnet cold. |
| 3 | Saturday 11:20 a.m. | Matches address; photo of door | Adult at the door said the recipient was out and declined to give a name. Left my card. Asked when to return; no answer given. |
| 4 | Monday 5:50 p.m. | Matches address; photo of door | Recipient answered to the name on the papers, accepted the documents in hand. Description recorded: [fields]. |
The weak version of attempt 2 would read "Tried again, no one home". It has a timestamp and nothing else, so the server must rebuild the story from memory when the affidavit is written. That is exactly when a drafting assistant is most tempting, and most dangerous. A good log means any drafting tool, human or AI, has facts to work from.
Where AI fits: six jobs, ranked by risk
AI is useful at the edges of this work, where it handles reading and typing, and risky near the middle, where the server's sworn account lives. Rank each job by what a mistake would cost, and keep a human check at every step.
| Job | Risk | What AI does | The human check |
|---|---|---|---|
| Reading incoming papers into the job record (AiMY, Proof's Autofill, or a general model) | Low to medium | Extracts parties, case number, court, hearing date, addresses and documents to serve | Compare each field with the paper, especially addresses and dates |
| Summarising the file for the dispatcher | Low | One-paragraph brief with special instructions and deadlines | Dispatcher reads the source for anything unusual |
| Routing and batching | Low to medium | Groups addresses by area and time window | Check windows and rush jobs on the map |
| Client status updates | Low to medium | Turns log entries into a plain message | Nothing added that the log does not say |
| Affidavit narrative drafted from the server's log | Medium to high | Restates logged facts in sentence form | The server reads every line against the log and signs |
| Address research and skip tracing | High | Nothing reliable; a model cannot verify where someone lives | Use authorised data sources and confirm on the ground |
Two of those jobs repay a worked example. The first is reading the papers, where the prompt below asks for a fixed set of fields and an honest "unreadable" flag.
Read the pasted text of a court document and fill these fields: case
number, court, party to be served (exact spelling), address for service,
documents to be served, hearing or response date, special instructions.
Use only the pasted text. If a field is missing or unclear, write
"unreadable" or "not stated". Do not guess. Quote the sentence you relied
on for each field.
[paste text, or the extracted text of the PDF]A reply might look like the next block, with the personal details left as placeholders.
Case number: [as printed] (quoted: "Case No. [as printed]")
Court: not stated
Party to be served: [name as printed] (quoted: "[sentence naming the party]")
Address for service: [address as printed] (quoted: "[sentence with the address]")
Documents to serve: Summons; Complaint (quoted: "Summons and Complaint")
Response period: "within [number] days after service" (as written; no date calculated)
Special instructions: unreadableWhat you would fix: the model correctly refused to guess the court, which is the behaviour you want. It also kept the response period as written instead of converting it to a date, so nobody should turn it into a deadline without the court's rules. The "unreadable" instruction field needs a person to look at the scan. Scanned papers are the weak spot for any tool, so test yours on your worst scans, not the clean ones. The sibling tutorials on AI extraction versus traditional OCR and AI sorting of incoming documents cover what to expect from scans.
The second example is the riskiest job on the list: drafting the narrative of an affidavit. The rule is that the model sees only the server's own log, never a rough draft, and is told to flag gaps rather than fill them.
You are helping a process server draft the narrative section of an
affidavit of service. Use ONLY the attempt log pasted below. Do not add,
infer or embellish any fact, time, description or conclusion. Write in the
first person, past tense, one numbered paragraph per attempt, in date
order, using the times and wording of the log. If something an affidavit
normally needs is missing from the log, list it under MISSING instead of
filling it in. Never say anyone was avoiding service.
ATTEMPT LOG: [paste]Run on the four-attempt log above, an acceptable reply would look like this.
1. On Tuesday at 6:40 p.m. I attended the address for service, knocked
three times and rang the bell. No one answered. A car was on the
driveway and lights were on downstairs. I photographed the front door.
2. On Thursday at 7:55 a.m. I attended again and rang twice. No one
answered. The same car was present. I photographed the door and car.
3. On Saturday at 11:20 a.m. an adult at the door said the recipient was
out and declined to give a name. I left my card.
4. On Monday at 5:50 p.m. the recipient answered to the name on the papers
and accepted the documents in hand.
MISSING: description of the person served; which documents were handed
over; how identity was confirmed.Check it line by line against the log. This reply dropped "bonnet cold" from attempt 2, so decide whether that detail belongs in the sworn account and put it back if so. The MISSING list is the valuable part: it tells the server what must come from memory or notes, and each item should be filled only with what the server actually did and saw. The server then reads the whole document, corrects it and signs. If the server would not feel comfortable swearing to a sentence, it comes out.
Six jobs that never go to AI
- Deciding whether service was valid, or which method of service to use.
- Adding an attempt, time, description or detail that is not in the log.
- Signing, e-signing or notarising for the server.
- Editing timestamps or GPS data.
- Interpreting court rules or calculating deadlines.
- Contacting the person to be served.
Write these into a one-page policy for your servers and office staff. The checklist for spotting AI-generated evidence shows what lawyers on the other side may look for in machine-written material, which is one more reason to keep the sworn account in the server's own words.
A demo script for the vendor call
Send this list ahead of the call and ask the vendor to do each item live, using a real order from your files.
- Create a job from an order PDF, and show how long it takes.
- Show the warning for a duplicate job.
- Assign a job to an outside contractor and show what that person can see.
- Log an attempt from a phone, with a photo and GPS, ideally somewhere with a weak signal.
- Edit that attempt and show the audit history.
- Produce an affidavit on your court's template, and show who can edit the text after it is signed.
- Show what a client sees, and prove one client cannot see another's jobs.
- Invoice a job with a rush fee and mileage, and calculate a contractor server's pay.
- Export all your data, meaning jobs, attempts, photos and affidavits, in a usable format.
- If the product reads documents with AI, show the extracted fields beside the source, say where the text is stored, for how long, and whether it is used to train a model.
The ninth item is the easiest to skip, and it decides how painful leaving a vendor will be. The tutorial on mobile notary software with AI uses a similar demo approach for a neighbouring trade.
A proof that said more than the log
A three-server agency let a server paste a rough draft affidavit into a general chatbot with the instruction "tidy this up". The reply added that the recipient "appeared to be avoiding service" and gave a time for one attempt that appeared nowhere in the log. The server, in a hurry, signed. The lawyer on the other side questioned the statement, the agency spent most of a day reconstructing the facts from the app's records, and the server had to correct the document.
The fix was procedural. Servers now paste the log, not a draft; the prompt forbids conclusions and lists gaps; and every paragraph is checked against a log entry before signing. This is an illustration, not a reported case, but it shows why the risk ranking above puts affidavit drafting near the top, and why the attempt log has to be good before any AI touches it.
Questions serving firms ask about software and AI
Will a court accept an affidavit drafted by AI?
That depends on the court's rules, which I cannot state for you. What stays constant is that the affidavit is the server's sworn statement. Treat AI as typing help working only from the server's own log, have the server read every line against that log, and check the court's template and signing rules before relying on any draft.
Should I pay per plan or per job?
Work out your own monthly volume and price both. In the example here, the flat plans were cheaper at 75 jobs a month and the per-job tiers were cheaper at 140 and 300, so the answer changes with volume. Confirm how tiers stack and what is not included before you commit.
Can AI find a defendant's new address?
Do not rely on it. A model cannot verify where someone lives, and a wrong address wastes attempts and weakens the record. Use data sources you are authorised to use for address research, log what you found and where it came from, and let the server confirm on the ground before the first attempt.
Do I need a client portal?
If clients ring to ask for status, yes. A portal where they can see job status, upload papers and download finished proofs removes most of those calls. Check that clients can only see their own jobs, and that you can switch off access when a relationship ends.
Further reads
- AI Handbook for Notaries: A Written AI Use Policy and Safe Prompts — A written AI policy and safe prompts for another signing-based trade.
- Signing Agent Software and AI: Scheduling, Quotes, Follow-Up — Scheduling, quotes and follow-up for mobile signing work.
- AI for Court Reporters: Draft Transcripts, Proofing and Turnaround — Draft, proof and turnaround in another court-facing role.
- AI Document Processing: Turn PDFs and Scans Into Usable Data — How AI turns PDFs and scans into usable data.
- AI Client Intake for Law Firms: Qualify Enquiries Out of Hours — The law-firm side of the same intake problem.
- NEMT Software for a Small Operator: A Checklist for Dispatch, Logs and AI — What to test in NEMT software before you sign: dispatch, trip logs, driver app, billing exports, plus which AI claims are real and which are labels.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ServeManager, Process Server's Toolbox, Process Organizer and Proof websites and pricing pages, checked on 4 November 2026.