The task
Onboarding goes wrong quietly: one missing document nobody chased, discovered on the first morning. Configure an agent to check each submitted file against your requirement list and return what is outstanding, citing the document it read. The output is a chase list with a date on it, not a compliance opinion.
What you provide
The submitted documents
What the joiner uploaded, usually as photographs. Identity documents, qualifications, signed policy acknowledgements and any role-specific certification.
Your requirement checklist
The documents required for this role and location, the fields you need from each and the validity you accept.
The start date
So the output can distinguish what is missing from what expires shortly after the joiner begins.
Give the agent a clear brief
Use these instructions as a starting point, alongside your approved sources and permissions.
Identify each submitted document and map it to the requirement it satisfies. Extract only the named fields, citing the file and page. Report a requirement as outstanding when nothing covers it, and as expiring when it lapses near the start date. Report an unreadable photograph as unreadable. Do not make an eligibility determination, confirm a right to work or approve a joiner to start.
How the work comes together
Match documents to requirements
Classify each upload against the checklist. Joiners routinely submit the right document under the wrong label, and a misfiled document is not a missing one.
Read only the fields you need
Take the name, the reference and the expiry date. A personnel file should not accumulate fields nobody asked for because they happened to be legible.
Produce the chase list
Return what is outstanding, what is unreadable and what expires soon, ordered against the start date. That list is the only part anyone acts on.
Illustrative example · fictional data
What still needs chasing
An illustrative check of a fictional joiner’s uploads against a requirement checklist, two weeks before the start date.
| Requirement | What was submitted | Status |
|---|---|---|
| Photo identity document | Passport photo page, readable, valid | Received |
| Professional certification | Photograph too blurred to read the reference | Resubmission needed |
| Signed policy acknowledgement | Nothing submitted for this requirement | Outstanding |
Legal eligibility to work is a determination for your people team under the rules of the jurisdiction. This playbook reports what the documents say and stops there.
The human review
An HR specialist confirms every finding and makes any eligibility determination. The agent produces a chase list; it does not decide that someone may start work.
Set it up for your team
Set the checklist per role and location
Requirements differ by jurisdiction and role. One global checklist produces confident findings that are wrong in half your locations.
Expect poor photographs
Most joiner uploads are phone photographs taken in bad light. Test on that quality, not on clean scans, and make unreadable a first-class result.
Restrict what is extracted
Name the fields to read and nothing else. Identity documents contain far more than your process needs.
Check the result before scaling
Run it over completed onboarding files, including ones where a document was initially missing or unreadable. Count requirements wrongly marked as received, which is the failure that reaches the first morning.
Keep the data in the right place
Onboarding documents are among the most sensitive personal data an employer holds. A fully local deployment keeps identity documents inside your network, and the field restriction above is what keeps the personnel file proportionate.
Explore deployment optionsQuestions & answers
Does this decide whether someone can legally work?
No. It reports what the submitted documents say. Eligibility determinations belong to your people team under the rules of the jurisdiction.
What about blurred photographs?
They are reported as unreadable so they can be resubmitted. That is a deliberate result, because a guessed reference number is worse than a missing one.
Will it store the identity documents?
Storage is your deployment decision, not the playbook’s. Configure where documents live and how long they are retained before you process a real joiner.