Synthetic Onboarding — Direct Deposit Authorization Data

What this document is

The direct deposit authorization tells payroll where to send an employee's pay: bank name, routing number and account number for a primary account, an account type and a full-or-partial deposit election, and an optional second account for a split deposit. It is a short form with outsized consequences, because the numbers on it move money.

Why generate synthetically

This is the single most attacked document in HR. Payroll diversion — a fraudulent direct deposit change submitted in an employee's name — is one of the most common and most lucrative business email compromise variants, and the defence is a system that reads the form correctly and notices when something about it is off. Building that system needs volume, and volume is impossible: a real corpus is a set of live bank account numbers. Synthetic authorisations give correctly-shaped, entirely fictitious banking details at any scale.

What makes synthetic data useful

Each form is one simulated employee's real pay instruction. Account type is checking on 70.5% and savings on 29.5%, and the deposit election tracks it: 70.5% deposit the full net amount to one account, while 29.5% split, naming a second bank with its own routing and account number and a fixed dollar amount between $100 and $500. Bank names, routing numbers and account numbers are format-correct and belong to no real institution or accountholder, which is what makes the corpus safe to redistribute and still useful for fraud-detection training.

Training challenges

Routing and account numbers are long digit strings with no visible checksum, printed close together in a compact block, and a single transposed digit misdirects a payment without producing any downstream error a human notices until payday. The second account block exists on every form but is populated on only 29.5% of them, so nearly three quarters of documents present a fully labelled and completely empty block — the strongest hallucination bait in the onboarding packet. Two of the second-block checkboxes are never marked in the corpus at all, and the percentage-split alternative to a dollar amount is never exercised, so those targets supply negatives only.

Who uses this data

Payroll and HR-tech platforms automating pay-instruction capture, payroll-fraud and business-email-compromise detection teams training on the document the attack actually uses, employer-of-record and PEO providers onboarding at volume, and banking integrators validating routing and account capture from scanned forms.

Document complexity profile

21 fields on a single page: 13 text and 8 checkbox targets across two structurally identical account blocks, with one label-to-value relation per field. 1 arithmetic binding and 2 function calls derive the split amount, at maximum expression depth 2. The second account block is fully labelled on every form and populated on under a third of them.

Key stats from our synthetic corpus

Quantitative characteristics of the Onboarding — Direct Deposit Authorization documents our generator produces.

How this document co-occurs with others

Rates at which identities in our corpus that produce a Onboarding — Direct Deposit Authorization also produce other documents.

Prevalence rates above are corpus-derived: they were computed over the 641 employment-eligible identities inside a local synthetic corpus of 1,000 identities generated by SymageDocs' World Simulation Engine at seed 20260421 — the shipped definition gates onboarding forms on active employment. Field, type and relation counts, the never-marked checkboxes and the unused percentage field come from the shipped direct deposit authorization definition in the SymageDocs form library. No real employee or banking data was used at any stage.

Frequently asked questions

What data format do synthetic direct deposit forms include?

Each generated identity produces a rendered PDF plus a structured JSON annotation file with bounding boxes, field types, and ground-truth values for all 21 fields — 13 text and 8 checkbox targets — with one label-to-value relation per field. Annotations record which account block each value belongs to. COCO, YOLO, FUNSD, and BIO/NER exports come from the same job.

Are the bank account numbers real?

No. Bank names, routing numbers and account numbers are format-correct and entirely fictitious. No generated form points at a real institution's account, which is the property that makes a payroll-diversion training corpus possible at all.

How often is the deposit split across two accounts?

29.5% of forms split, naming a second bank with its own routing and account number and a fixed dollar amount between $100 and $500. The other 70.5% deposit the full net pay to a single account and leave the second block entirely empty.

Which targets are never exercised?

The second account's savings-type and partial-deposit checkboxes are never marked, and the percentage-split field is never populated — the corpus only ever uses a fixed dollar amount for the split. Treat those as negative-only rather than assuming coverage.

Can I use this data commercially?

Yes. Every employee, bank, routing number and account number is synthetic, contains no real personal or financial data, and is licensed for commercial use including model training, benchmarking, and redistribution inside your own products.