Merchant recognition
Recognises merchant names across thousands of formats and aliases. McDonalds, McD's, and MCD-1247-OPS all roll up to the same vendor in your books. New merchants get tagged in days, not quarters.
DocPro · Receipt Agent (ADE)
Receipt capture that books the expense, not just reads the print.
The Receipt Agent reads paper receipts, phone photos, PDF receipts. It pulls merchant, total, tax components, tip, and line items, tags the expense against your policy, and returns JSON your expense tool can post. Most receipts clear without a human in the loop.
What it actually does
A receipt API built for the people who close the books.
Not just OCR on a phone photo. The Receipt Agent finishes the parts a finance team would otherwise hand to a person.
Totals, taxes, tip — split
Subtotal, tax, service charge, and tip returned as separate fields. Honours the math when the bill rounds awkwardly. Flags the receipt when the printed total disagrees with itself, instead of silently picking one.
Line items, not just totals
Pulls each line: item, quantity, rate, total. Useful when a policy cares whether the meal was for one person or eight, or when you need to claim input tax credit on specific SKUs.
Policy-aware categorisation
Tags each receipt against your category tree. Cabs, meals, accommodation, supplies, client entertainment. The policy lives in config, not in a reviewer's head, so the same receipt is tagged the same way every time.
Quality scoring at capture
Tells the user the photo is too dark, too blurry, or missing the total before the claim is filed. Saves a two-week back-and-forth and a closed expense report that has to be reopened.
How it works
Four steps, one REST call, JSON your expense tool can post.
No pre-processing on your side. No template setup per merchant. The agent does the reading and the reasoning.
-
Submit
POST a JPEG, PNG, HEIC, or PDF to
/v1/receipts. Phones, scanners, and email attachments all welcome. Multi-page receipts are fine. -
Read
The agent classifies the format, locates the relevant blocks, and extracts every field. Confidence is reported per field, not per document, so you know exactly what is sturdy and what is not.
-
Validate
Subtotal plus tax plus tip equals total. Currency matches country. Date is plausible. Tax rate fits the merchant category. Anything that fails routes to a reviewer.
-
Return
Structured JSON in your schema, plus a webhook fired when review is complete. Pre-built connectors for SAP Concur, Zoho Expense, NetSuite, and Workday Expense.
Feature matrix
Receipt Agent vs. legacy OCR vs. manual entry
What actually changes when a sales team uploads 6,000 receipts at month-end.
| Capability | Manual entry | Legacy OCR | Receipt Agent (ADE) |
|---|---|---|---|
| Reads photographed receipts at any angle | Slow, error-prone | Sometimes | Yes |
| Faded thermal paper | If readable | Often fails | Yes |
| Tax split (GST / VAT / service) | Manual | Returns one number | Per component |
| Merchant normalisation to vendor master | Per receipt | No | Built in |
| Line-item extraction | If asked | No | Yes |
| Expense category against policy | Yes | No | Yes |
| Confidence score per field | No | No | Yes |
| Multi-currency | Manual | Latin scripts | 40+ currencies |
| Per-receipt cost at scale | USD 2–5 (loaded) | USD 0.10–0.30 | USD 0.02–0.08 |
| Throughput per hour, single tenant | ~40 receipts | ~2,000 | ~50,000 |
Pricing is indicative. Final costs depend on document mix and volume commitments.
Where teams use it
Built for the workflows where receipts pile up.
Six places the Receipt Agent quietly removes a weekly bottleneck.
Corporate travel & expense
Sales, consulting, and field-ops teams uploading receipts from a phone. Most submissions clear within the hour, and the policy violations land on the right approver instead of bouncing through three people.
Reimbursement at scale
A 400-person sales force, 80,000 receipts a month. The agent posts the routine ones and routes the rest to a small review team. No more month-end backlog clearance weekends.
Restaurant & hotel chains
Vendor receipts, banquet bills, supplier slips. Pulls the GST credit accurately, splits taxes by state, and reconciles against the POS so the chef doesn't moonlight as the bookkeeper.
Cab, ride, and delivery aggregators
Driver fuel receipts, toll slips, maintenance invoices. Categorised by vehicle, tagged for input credit, and reconciled with the per-trip records. The driver app does the capture; the agent does the rest.
Patient reimbursements
Pharmacy bills, diagnostic receipts, hospital invoices. Extracted, validated, and matched against insurance claim rules. Cuts the back-and-forth between patient, hospital, and TPA from days to minutes.
Tax and statutory audits
Bulk-process a year of receipts before the auditor walks in. Categorised, totalled, GST-mapped, and matched against the ledger. The exception list is short and the audit closes faster.
Receipt Agent in numbers
~95%
Straight-through processing
< 6 s
Median time per receipt
99.95%
API uptime, last 12 months
Security & data residency
Receipts contain PII. We treat them like it.
The same posture you would expect from a banking vendor, applied to a humble expense receipt.
Encryption end-to-end
TLS 1.3 in flight. AES-256 at rest. Customer-managed keys (CMK) available on enterprise plans. Image and JSON are encrypted with separate envelopes.
Region-pinned residency
Choose AWS, Azure, or Google Cloud, in the region your contract names. On-premise and air-gapped deployments are available for banks, regulators, and government workloads.
Audit trail and replay
Every receipt has a signed, time-stamped audit record from upload to post. Replay the agent's decision later, with the exact model version and rule pack used.
FAQ
Questions teams ask in the first 30 minutes.
If yours is not here, a short call with an engineer is usually faster than email.
What kinds of receipts can the Receipt Agent read?
Paper receipts photographed on a phone, scanned receipts, PDF receipts emailed by the merchant, HTML receipts forwarded from Gmail, and screenshots of mobile-payment receipts. Faded thermal paper and crumpled paper both work, with reduced confidence scores when the print is genuinely unreadable. If the photo is unusable, the API says so instead of guessing.
How accurate is the Receipt Agent?
Around 95 to 99 percent field-level accuracy on totals, tax components, and merchant name once the agent is tuned to your usual receipt mix. Every field carries a confidence score, so the receipts that look risky route to a reviewer instead of posting silently. The interesting number is not raw accuracy — it is what you do with the 1 to 5 percent that needs a second look.
Can it categorise expenses against our policy?
Yes. You ship us your category tree and policy rules at onboarding. The agent tags each receipt against that tree (cabs, meals, accommodation, supplies, client entertainment) and flags receipts that violate policy — over per-meal cap, alcohol on a no-alcohol policy, missing GSTIN, duplicate of a receipt already filed — before they hit a reviewer.
How does it integrate with our expense tool?
Pre-built connectors for SAP Concur, Zoho Expense, Expensify, NetSuite, Workday Expense, and Oracle Fusion. Anything else is a REST endpoint and a webhook. You send a file, you get JSON back, and your tool posts the expense. SDKs ship for Python, Node.js, Java, and .NET.
How is the Receipt Agent priced?
Per receipt, with volume tiers. Indicative pricing starts around USD 0.04 per receipt at low volumes and falls sharply above a million receipts a year. Enterprise plans bundle committed throughput, custom categorisation, CMK encryption, and priority support. Final pricing depends on your document mix — happy to share a real quote once we see a sample batch.
Where is our receipt data stored?
In the cloud region you choose, on AWS, Azure, or Google Cloud. On-premise and air-gapped deployments are available for regulated environments. We sign a DPA before the first receipt is processed, and data is encrypted at rest and in transit. PII can be redacted from logs on request.
Can we try it on our own receipts before we sign anything?
Yes. Send us a batch — 100 to 1,000 receipts is ideal — and we will run them through the sandbox and return the JSON within a working day. No credit card, no sales call needed for the first batch. If it works, we talk pricing. If it does not, you have output to compare against whatever else you are evaluating.
Send us a thousand of your worst receipts.
Crumpled, faded, photographed under a streetlight, in three currencies, with handwritten tip amounts. We run them through the Receipt Agent and send back the JSON. No deck, no sales call. Just output you can compare against what you have now.