Scanned papers in, checked entries out, with people settling the doubts.
Purchase invoices, KYC files, claim papers and export documents reach you as scans, PDFs, photographs and e-forms. We would build an intake that identifies each document, picks out the fields you need, tests them against your rules and master data, and passes only the doubtful ones to a reviewer.
For Indian enterprises and public bodies that receive papers in many formats and languages and must be able to trace every value.
Every document enters one intake from its channel and is identified, read and checked by the system. A document with a weak field or a failed check is held in a reviewer’s queue. Checked data passes into the ERP or case system, and each correction a reviewer makes is added to the test set.
Before and after
How the working day changes
Today
With the system
Today
Today: Fields copied by hand
Staff open every scan, PDF or photograph and type GST numbers, amounts and references into the ERP or the loan system one at a time.
With the system
With the system: Structured data from the start
Each field is captured together with its position on the page, so a reviewer can compare the figure with the spot it was read from.
Today
Today: Mistakes surface at month-end
A wrong registration number or a missing page comes to light only during reconciliation, an audit query or a returned payment.
With the system
With the system: Checks at the point of entry
GST registration, purchase order, vendor master and duplicate checks run before any entry is created, while the sender can still correct the document.
Today
Today: Equal effort for every file
A clean e-invoice with its IRN takes as much handling time as a blurred photograph of a handwritten bill.
With the system
With the system: Effort spent on doubtful items
Clean documents flow through untouched. Reviewers work a queue of held items only, each tagged with the reason it was stopped.
Today
Today: No trail behind a figure
When an auditor asks where an amount came from, the only answer is the name of whoever typed it.
With the system
With the system: A trail for every field
For each field the system keeps the source page, the reading, its confidence score, the check results and any change a reviewer made.
Workflow
Step by step
Seven steps and one review path. Your process owners agree the thresholds and checks, and we tune them on your own documents before the system handles live volumes.
Illustrative workflow
7 steps · 1 human checkpoint
Legend
System
Person
Human checkpoint
Exception path
Document intelligence: illustrative flow with a review path
Read the diagram as text
Documents move from intake through separation, reading and checks to posting in the system of record, and finally to learning from corrections.
A document with any field under the confidence threshold, or one that fails a check, is diverted after the checks to the reviewer queue. A person accepts or corrects the values there, and the document then rejoins the main line at the posting step.
The reviewer queue is the point where a person must act: a doubtful value cannot reach the system of record on its own.
01
Step 1: One intake for all channelsSystem
Scans, email attachments, portal uploads, mobile photographs and e-forms enter a single queue with a reference number, the channel and a checksum. Where a GST e-invoice carries its IRN and QR code, the structured details are taken from the code rather than read off the printed page.
02
Step 2: Separate and identifySystem
Bundles such as a loan file or an export shipment set are split into individual documents. Each one is identified by type, language and script, which decides the field list and the checks that apply to it.
03
Step 3: Read the fieldsSystem
Layout-aware OCR reads each page and a language model fills the agreed fields into a fixed schema. Every value carries a confidence score and the coordinates where it was found.
04
Step 4: Check against rules and mastersSystem
Values are tested against your vendor and customer masters, GST registration numbers, bank details, purchase orders and duplicate rules. A document whose fields all clear the agreed threshold and every check moves on without a reviewer.
05
Step 5: Reviewer queuePerson
The reviewer sees the page and the extracted values next to each other, with the weak field or the failed check marked, and accepts or corrects each one.
Exception path: Weak field or failed check· Returns to step 6
Human checkpoint
No doubtful value enters the ERP or the case system until a person has looked at it. A reviewer may also reject the document and ask the sender for a fresh copy.
06
Step 6: Post to the system of recordSystem
Checked data creates or updates the entry in the ERP, loan system or case-management system through its API, with the original attached and a retention class assigned.
07
Step 7: Improve from correctionsSystem
Each correction is stored as a labelled example and added to the test set. A new threshold or prompt goes live only after it passes that set.
Components
Components we would deliver
01
Channel connectors and intake
Connectors for shared mailboxes, scanners, upload portals and e-forms that bring every document into one queue with its metadata and reference.
A labelled test set, release gates and dashboards that show the share processed without review, the reasons documents are held and any drift in field accuracy.
Salesforce for case and customer service recordsPlatformSalesforce
SAP S/4HANA accounts payable and procurementPlatformSAP
Odoo accounting, purchase and inventoryPlatformOdoo
Records management and archive
Vendor, customer and GST registration masters
Directory and single sign-on for reviewer roles
Controls
Controls built in
Access boundaries
Each reviewer sees only the document types and files that the role permits. The reading services hold no standing access to the ERP; every posting runs under a separate service account with a narrow scope per integration.
Human review
A weak confidence score or a failed check sends a document to a person. Thresholds are set field by field rather than per document, and lowered only when the test set shows it is safe. Sanction of loans, settlement of claims and acceptance of applications remain with your officers.
Auditability
Per field we record where on the page it was found, the value read, its confidence, the check results, the model and prompt version, and any reviewer change with a reason code.
Data protection
Only the personal data the stated purpose requires is captured, in line with the purpose limitation of the DPDP Act. Redaction is available, logs are kept for at least one year as rule 8(3) of the DPDP Rules requires from May 2027, and deletion follows your retention schedule. Processing takes place in the region you agree.
AI transparency
Each model and prompt version is documented with its test results. Where extracted data feeds decisions about people, the use of AI is entered in your AI register and explained in the notice given to applicants or customers.
Measures
How success would be measured
We fix these measures with you in discovery and first record a baseline from the way documents are handled today.
How success would be measured
Metric
Why it matters
How we would measure it
01Share processed without review
Why it mattersTells you, document type by document type, how much handling now happens without anyone touching it.
How we would measure itProportion of documents clearing every threshold and check with no review, taken from the workflow logs.
02Held documents and their reasons
Why it mattersSeparates bad scans and incomplete files from checks that are too strict and need revisiting.
How we would measure itVolume in the reviewer queue by reason code, split by channel and document type.
03Accuracy per field
Why it mattersThe figure that matters for the payments, sanctions and filings that rely on the data.
How we would measure itPrecision and recall for each field on a labelled test set agreed with you, repeated at every release.
04Turnaround and review effort
Why it mattersShows the time saved overall and the time that has shifted into review.
How we would measure itTime from receipt to posting, and active handling time per held document, from queue timestamps.
No targets are set before a baseline exists.
Rollout
Phased rollout
Phase 01
Discovery and baseline
Two or three document types, such as purchase invoices and delivery challans, on one channel, with current volumes and error patterns. We label a test set from real documents with personal data masked.
Exit criteria
Fields, checks and acceptance criteria signed off
Baseline recorded
Data protection assessment under way
Phase 02
Shadow-mode pilot
The system reads and checks documents alongside the current process; staff continue the work and compare outcomes.
Exit criteria
Accuracy per field within the agreed criteria
Thresholds fixed for each field
Reviewer team trained on the workbench
Phase 03
Controlled go-live
Documents of the piloted types that clear every check post directly, and people inspect a sample of them.
Exit criteria
Sampling finds no unacceptable error
Runbook and dashboards with your team
Rollback rehearsed
Phase 04
Scale-up
Further document types, channels and languages, each clearing the same test gate before it goes live.
Delivery challans, test certificates and export shipping documents matched against purchase orders and sales orders.
Illustrative example
Purchase invoices at a multi-State distributor
Situation
A distributor of electrical goods with warehouses in several States receives supplier invoices as GST e-invoices with an IRN, as PDFs and as photographs of paper bills from small vendors, each arriving in a different finance mailbox.
System
One intake would read the invoices that carry an IRN as structured data and extract the rest. Each value would be matched in the ERP to the purchase order, the goods receipt and the vendor master, including the supplier’s GST registration.
Human control
Invoices with weak fields, a changed bank account or a price difference would wait in a reviewer queue for the accounts payable team. Payment would be released only from the ERP, after the maker-checker approval.
What we would measure
Share processed without review per warehouse and channel, the reasons invoices are held, and accuracy per field on a labelled test set.
India
GST invoices at home, structured invoices in Europe
An Indian business receives supplier invoices in several forms: GST e-invoices registered on the Invoice Registration Portal (IRP) with an Invoice Reference Number (IRN), PDFs, and scans of paper bills from smaller suppliers below the e-invoicing threshold. We design one intake in which invoices carrying an IRN are read as structured data, while the rest pass through extraction and the same checks.
Groups with European subsidiaries meet a second format, because EU e-invoicing is moving to structured invoices under the VAT in the Digital Age package, and those invoices skip extraction too. Where extracted data feeds decisions about individuals, the DPDP Act governs it in India, and the GDPR and the EU AI Act come in where people in the EU are concerned. Either way we design for documented processing, review by a person and logs your auditors can follow.
That depends on your documents, so we measure it instead of promising it. We label a test set from your own documents with personal data masked and report precision and recall for each field. Models do make mistakes, which is why a doubtful field goes to a person and never straight into your ERP.
02Can it read handwriting, faint scans and documents in Indian languages?
Handwritten bills and faint scans can be read at lower confidence, so a larger share of them reaches the reviewer queue. Documents in English, Hindi or another Indian language are first identified by language and script; a language would go into production only after its own test set shows acceptable accuracy.
03We already run an OCR or capture tool. Must it be replaced?
Usually not. If your capture layer performs well, we add checks, review and posting around it, and bring in model-based reading only for the document types where it improves results.
04Will the system sanction loans or settle claims?
No. It prepares and checks data. Sanctions, settlements and other decisions with legal or financial effect stay with your officers, as your own policies require. Where the GDPR applies to the processing, its Article 22 points the same way.
05Where are documents processed and stored?
In the cloud region or data centre you select, written into the contract. We document every processing step, including any external model service, so your privacy and legal teams can assess it before go-live.
06Where should a first release start?
With one high-volume document type that has clear checks, such as purchase invoices or a single loan application form. It proves the flow, the review screen and the measures before more types are added. Our delivery model explains the phases.
Share your document flows
Pick two or three document types and the systems where their data ends up. We will show what a first release could cover, from GST invoices to the papers of your European entities.