An agent that sends each client the right checklist, reads what comes back, files it against the right client and year, asks only for what is still missing, and keeps one live board of who is complete.
Reference build — a working agent and a worked example, not a client deployment
Fifteen days before a due date, a firm with 400 individual and small-business clients still has something missing from most of them: a bank statement, Form 16, a capital gains statement, a sales register. An article assistant messages each one from a personal WhatsApp, several times. Files come back as photos of single pages and get saved into folders named after the client, mostly. Nobody can answer "who is complete?" without opening 400 folders.
Three touches per client at about five minutes each is roughly a hundred hours a season on chasing alone, done by people who should be learning accounting. And the missing capital gains statement is still found by the partner at review, two days before the deadline.
Why an agent, and not a chatbot or a Zap: A reminder sequence sends the same message to everyone on a timer and cannot tell that the photo a client sent is page two of a three-page statement. The agent reads what came back, decides what is still missing for this client, and asks for only that. That decision, made 400 times a day, is the whole job.
The tool list is the permission list. If a capability is not here, the agent does not have it, however it is asked.
| Tool | What it does | What it may touch |
|---|---|---|
get_client_profile | Loads the client type and last year’s document set | Reads the client master |
send_checklist | Sends the approved checklist message for this client | Writes to WhatsApp or email via the firm’s templates |
classify_document | Identifies a received file, its owner, period and completeness | Reads the uploaded file only |
check_completeness | Compares what has arrived against the checklist | Reads the tracking rows |
request_missing | Asks for exactly the missing items, within the frequency cap | Writes an approved message |
update_board | Updates the completion board and flags contradictions | Writes to the board and the staff queue |
| Hours chasing and filing documents, per season | ~100 → ~15 | Worked example |
|---|---|---|
| Knowing who is complete | 400 folders → one board | Worked example |
| Classification accuracy on common document types | Measured on the firm’s own files in week one of a pilot | Target |
| Missing items found at partner review | Tracked per season as the pilot’s headline | Target |
Assumptions
What a pilot measures that this page cannot
A pilot runs it on your data in two to three weeks for a fixed fee, measured against your own baseline. Thirty minutes to scope it, free.