Extraction that understands layout
Modern document models read tables, stamps, handwriting and regional scripts, without a template per vendor.
Build
Stop retyping PDFs.
We build systems that read invoices, KYC documents, claims, contracts and bank statements, check what they read, and put clean data into Tally, SAP or your core system. A person reviews only what the system is unsure of.
- Typical length
- 4 to 10 weeks
- Indicative price
- ₹6 lakh to ₹30 lakh, excluding GST
- Engagement path
- Discover→Prove→Build→Run
Sound familiar?
Typical situations, not quotes from clients.
- We have twelve people keying invoices into Tally. Every vendor uses a different format.
- KYC files come in as phone photos, sideways, in three languages. Our OCR tool gives up.
- Claims take nine days to process. Six of those are someone reading a PDF.
Templates break the moment a supplier changes its invoice layout. Indian documents add their own trouble: handwritten notes, stamps over text, regional scripts, mixed-language forms, and GST details that must match exactly.
What we build
Checks, not just reading
GSTIN format and checksum, totals that add up, PAN and IFSC validation, dates in range, purchase order match. A field that fails a check never reaches your ledger unreviewed.
A review screen for your team
Uncertain fields are highlighted on the original document. Your reviewer confirms or corrects in seconds, and each correction becomes a test case.
Classification and routing
Mixed bundles split into documents: invoice, delivery challan, credit note, KYC proof, discharge summary. Each one goes to the right queue.
Posting into your systems
Approved data is posted to Tally, SAP, Zoho Books, your loan origination system or claims platform, with a link back to the source page.
An audit trail
Who approved what, when, and against which document version.
How it fits together
Where it fits
- Tally Prime
- SAP S/4HANA and SAP Business One
- Zoho Books
- Oracle NetSuite
- Loan origination and management systems
- Core banking
- Claims management systems
- DigiLocker-sourced documents
- Email inboxes and shared drives
- SFTP drops from scanners
How the engagement runs
Typical length: 4 to 10 weeks from pilot start to production.
Discover
1 weekWe collect a sample of a few hundred real documents, count the formats and volumes, and agree which fields matter and what “correct” means for each.
Prove
2 to 5 weeksA fixed-price pilot on one document type, scored field by field against an evaluation set your team has labelled with us.
Build
2 to 5 weeksThe review screen, posting to your systems, more document types, security review and training for the operations team.
Run
monthlyWeekly accuracy reports by field and document type, new formats added, reviewer corrections fed back into tests.
How we measure it
We agree with you in writing, before we build:
Field-level accuracy
Per field, on a labelled evaluation set of your real documents.
Straight-through rate
The share of documents posted with no human touch.
Review time
Minutes per document for the ones a person checks.
Error escape rate
Wrong values that reached your system after review. The target here is close to zero.
Cost per document
Including model and cloud costs.
What you get
- A production extraction and validation pipeline
- Reviewer screen with keyboard shortcuts
- Integrations to your accounting or core systems
- Labelled evaluation set and accuracy reports, yours to keep
- Validation rules document
- Runbook and reviewer training
- Source code and configuration in your repository
What we need from you
- 200 to 500 sample documents per type, with the correct values where you have them
- Two or three people from operations for labelling and review sessions
- Access to the target system in a test environment
- A process owner who decides what is acceptable accuracy
- Your retention rules for source documents
For small businesses: Invoice and Document Automation, from ₹1.5 lakh. A fixed-scope package for one to three document types, such as purchase bills, delivery challans or bank statements. Fields are read and checked, then posted into Tally or Zoho Books as draft entries that wait for your approval. Ask for it on the discovery call.
Indicative price
₹6 lakh to ₹30 lakh
excluding GST, from pilot to production.
The range depends on the number of document types, languages, validation rules and target systems. The pilot is a fixed price quoted after Discover. 40% advance, the rest on milestones.
Model, API and cloud costs are passed through at cost plus a 10% handling margin when we hold the accounts (no margin if you pay the providers directly). We estimate cost per document before you sign.
Frequently asked questions
We already have an OCR tool. Why change?
You may not need to. If your OCR works for some formats, we keep it there and use document models only where it fails. The pilot shows where the line is.
Can it read handwriting and regional scripts?
Often, yes, but quality varies by document. We test on your own samples in Discover and tell you before the pilot what to expect.
What about Aadhaar numbers in KYC files?
We mask Aadhaar numbers as required, restrict who can see full values, and keep processing in your environment. The design is reviewed against RBI KYC directions and the DPDP Act with your compliance team.
How fast is it?
Most single documents take seconds. Large batches run in parallel overnight if you prefer.
Does accuracy drop over time?
It can, when suppliers change formats. That is why we measure weekly under Managed AI and add failing documents to the tests.
Send us ten documents your team hates.
On a 30-minute call we will look at what you process, how much, and what a pilot would need to prove.