What documents does the AI process automatically?

KYC docs (PAN Card, Aadhaar, Driving License, Voter ID, Passport) — OCR extracts ID number + name + DOB + address. Income docs (Salary Slip, Form 16, ITR) — extracts employer + gross/net amounts. Bank Statement PDFs — the AI Bank Statement Analyzer parses every transaction, detects salary credits, EMIs, bounces, computes monthly inflow + FOIR. GST returns (GSTR-1, GSTR-3B) — extracts turnover + filing consistency. Property docs (Sale Agreement, Title Chain) — extracts property type + value + address. Business docs (GST cert, audited financials) — extracts turnover + tax data.

How accurate is the extraction?

Each extracted field shows a confidence score. PAN/Aadhaar OCR is typically 95%+ when the scan is clean. Bank statement transaction parsing is 98%+ for all major Indian banks (HDFC, ICICI, SBI, Axis, Kotak, Bajaj, IDFC First, etc.). For low-quality scans the AI flags fields below 70 % confidence so you can verify. Always review before assigning to the file.

What happens after I click Assign?

The Assign dropdown asks where the value should land — usually the applicant or one of the co-applicants. Pick the target; SigmaDSA writes the value to that entity's field (PAN, name, address, etc.) and shows the new value in the Personal Details / Co-Applicants card on the File Details tab. The original extraction stays for audit.

Can I correct an extracted value before assigning?

Yes — click the value cell, edit, then click Assign. The corrected value gets assigned. The original AI output stays in the raw-extracted log so you can see what the AI saw versus what you decided.

Does the AI cost extra?

AI processing is included in the SigmaDSA AI edition. Lower tiers (CRM-only) don't include AI processing — uploaded documents are stored but not parsed. Check your billing edition under Billing & Subscription.

AI Features — SigmaDSA Help

Tour of the SigmaDSA AI Employee — document upload triggers OCR + Bank Statement Analyzer + GST parser automatically; extracted fields land on each document with confidence scores, and a one-click Assign sends each value to the right applicant or co-applicant field.

The SigmaDSA AI Employee has two faces:

Document intelligence — every uploaded document is automatically OCR'd, bank statements run through the Bank Statement Analyzer (BSA), GST returns are parsed, ITRs are read. Extracted fields appear on the file with confidence scores and one-click Assign to map values to applicant or co-applicant slots.
CRM Assistant chatbot — a floating chat button (bottom-right of every page) that lets you ask anything about your data in natural language — "this week's leads", "pending files", "my tasks today". The assistant respects your role-based permissions.

This article tours both.

CRM Assistant — the bottom-right chat icon

Every screen in SigmaDSA has a purple AI button in the bottom-right corner. Click it to open the CRM Assistant chat panel.

Dashboard with the AI floating action button bottom-right and the CRM Assistant panel open above it highlighted — CRM Assistant: (1) the floating AI button on every page, (2) the chat panel — opens with suggestion chips for common queries.

What the assistant can do

The assistant has access to your tenant's data and respects your permissions — a sales rep can only ask about their own leads/files; a manager sees the whole team.

Type of question	Example	What the assistant does
Pipeline summaries	"How many leads converted this week?"	Aggregates leads in date range with status = Converted.
Lookup	"Show me Anand's file"	Searches leads + files by name, returns matches as clickable cards.
Filter + count	"How many files are stuck in Login Pending more than 7 days?"	Filters the file pipeline with aging rule.
Action	"Create a callback with Nitin for tomorrow 5pm"	Parses intent and opens the callback creation flow pre-filled.
Quick stats	"What's my disbursal volume this month?"	Pulls the metric from the Disbursals report.
Reports	"Top performing lenders"	Returns the same data as Reports → Dashboard, in plain text.

Suggestion chips

The chat panel opens with three suggestion chips so users don't start from a blank slate:

This week's leads — list of leads created in the last 7 days.
Pending files — files in active stages (not Disbursed / Rejected).
Today's tasks — your task queue for the day.

Tap a chip to run that query without typing.

When to use the assistant vs the UI

The assistant is fastest for lookups and quick aggregates. The full UI is better for multi-step workflows (filling out a long form, reviewing a sanction modal, browsing the pipeline). A common pattern — use the assistant to find a record, click through to open it, then use the UI to take action.

Same backend powers WhatsApp + Telegram bots

The CRM Assistant uses the same AI model that drives the WhatsApp bot and the Telegram bot. Conversation context isn't shared across channels — each channel is its own thread — but the capabilities are identical.

Document intelligence

The second face of the AI Employee — automatic extraction from every uploaded document.

How extraction is triggered

Extraction runs automatically the moment a document is uploaded. There's nothing to enable, no manual trigger needed.

Upload point	What gets triggered
File detail → Documents tab → Upload	Per-document pipeline (OCR / BSA / GST / financials parser depending on detected type)
File detail → Credit Evaluation editor → Documents section	Same pipeline; results feed Credit Evaluation sections directly
Customer uploads via the share link or WhatsApp bot	Same pipeline, attribution flag flips to "customer-uploaded"

For a 6-month bank statement, BSA finishes in 30–90 seconds; for OCR on KYC documents, ~5 seconds. The pipeline runs in the background — you can keep working.

Where the AI's output shows up

The output lands in three places, depending on the document type and use case:

1. AI Extracted Data (Raw) — `File Details` tab

Every document the AI has processed shows here with the structured fields it found and a per-row Assign button.

File Details tab with Extracted Entities and AI Extracted Data sections highlighted, plus an Assign button — File Details tab: (1) Extracted Entities — consolidated entity view; (2) AI Extracted Data (Raw) — per-document field list; (3) Assign button per field.

Each row shows:

Column	What it shows
Field	The semantic name of the extracted item (e.g., "Full name of the person", "PAN", "Account Number").
Value	What the AI read from the document.
Confidence	How sure the AI is — 95 %+ is reliable, 70–95 % is usable with verification, < 70 % needs human review.
Assign	Click to write the value into the file's structured fields.

2. Credit Evaluation sections — Bank Statement, GST, ITRs

For documents that feed the credit decision, output also lands on the Credit Evaluation editor:

Bank Statement uploads → output populates the Bank Statement section (inflows, outflows, salary detection, EMI detection, bounces).
GST returns → populate the GST section (turnover, filing consistency).
ITR / Form 16 → populate income inputs of the DSCR section.
CIBIL PDF → populates the CIBIL Score section.

3. Extracted Entities card — consolidated entity view

A single card showing the entities the AI has identified across all documents — applicant, co-applicants, business entity, bank accounts. Each entity's fields are populated from the highest-confidence values across all extractions.

Click Re-extract All to re-run the pipeline if you've added new documents.

Assigning extracted values to file fields

This is the core action that saves typing.

Assign dropdown showing Applicant and Co-Applicant options — Click Assign → pick the entity to write the value into (Applicant / Co-Applicant).

Review the extracted value + confidence

In the AI Extracted Data card, find the field you want to populate. Eyeball the value against the source document — for high-confidence values (≥ 95 %) you can trust it; for lower confidence, verify by clicking the document filename to open the source.

Click Assign

The Assign button opens a dropdown listing the entities the value can map to:

Applicant — the primary borrower's structured fields.
Co-Applicant — each co-applicant entry (if multiple, each shows separately).
Business — if the file has a business entity, you can assign to its fields.

Pick the target.

SigmaDSA writes the value to the entity's field

The value lands on the relevant card on the File Details tab (Personal Details, Co-Applicants, Customer Bank Details, etc.). The corresponding field is updated, the file's other sections (Credit Evaluation, lender matching) refresh.

The original AI extraction stays in the raw log for audit — you can always see what the AI saw vs what you assigned.

What the AI extracts per document type

Document Type	Fields the AI returns
PAN Card	PAN number, full name, father's name, DOB
Aadhaar Card	Aadhaar number, name, DOB, gender, address
Driving License / Voter ID	ID number, name, DOB, address
Salary Slip	Employer name, employee name, month/year, gross + net amounts, deductions
Form 16	Employer, PAN of deductor, assessment year, annual gross + TDS
ITR / Computation	Assessment year, total income, total tax, refund/payable, PAN
Bank Statement	Account holder name, account number, IFSC, bank name, statement period, every transaction (date, narration, amount, type), monthly summary, salary credits, EMI debits, bounces
GSTR-1 / GSTR-3B	GSTIN, period, total taxable turnover, total tax paid, filing date
Property Sale Agreement	Seller, buyer, property type, area, purchase price, address
Audited Financials	Year, total turnover, PAT, total assets, total liabilities
Utility Bill	Consumer name, billing address, period, amount
Rent Agreement	Tenant, landlord, monthly rent, address, agreement period

Bank Statement Analyzer (BSA) — the heaviest lifter

For bank statement PDFs, the AI doesn't just OCR — it understands every transaction.

Within 30–90 seconds of upload, BSA produces:

Average monthly inflow (gross credits / months).
Average monthly outflow (gross debits / months).
Average closing balance.
Salary credit detection — date, amount, source bank per credit.
Existing EMI detection — auto-populates the Existing Loans section of Credit Evaluation.
Bounce events — cheque/ECS/auto-debit failures (each one is a lender red flag).
Salary Stability Score (1–10) for salaried applicants.
Cash Flow Trend (improving / flat / declining).
Suspicious Activity Flags — round-tripping, large cash deposits, etc.

This output is what drives the FOIR / DSCR computations and saves the most underwriting effort.

→ Full guide: Bank Statement section (Credit Eval)

Workflow shortcuts using extracted data

Upload a folder of customer documents → AI categorises them by type (PAN, Aadhaar, Bank Statement, etc.) → review the raw-extracted card → Assign each value to the applicant. Typical workflow: 5 minutes for a complete file vs 20 minutes of manual typing.
Self-employed file — upload ITRs + GST returns + Bank Statements → BSA + GST parser + ITR parser run in parallel → the entire DSCR section auto-populates within 2 minutes → Credit Summary auto-generates.
Co-applicant onboarding — upload PAN + Aadhaar for each co-applicant → AI detects multiple entities → Assign dropdown lets you map each set of fields to the right co-applicant slot.

What to do when confidence is low

70–95 % — verify visually against the source document (click filename to open). Edit the value if needed, then Assign.
Below 70 % — usually a scan quality problem. Either re-scan the document at higher quality and re-upload, or enter the value manually on the File Details card.
AI didn't detect a field at all — fall back to manual entry. The most common cause is non-standard document layouts (small regional banks, hand-written income statements).

Re-extraction

To re-run the AI pipeline on existing documents (e.g., after the AI engine is upgraded):

For all documents — Extracted Entities card → Re-extract All.
For one document — click its row → Re-extract OCR in the actions menu.

Useful when you've corrected the document type or when the AI engine releases an improvement.

Permission gating

Upload + Extract — anyone with File Update permission can upload and trigger extraction.
Assign extracted values — same permission.
Re-extract All — admins / managers by default; configurable per tenant.

AI Features