ITR and 26AS Analyser Software: Building a Verified Income View Automatically
An ITR and 26AS analyser parses the return and its schedules, pulls Form 26AS and the AIS/TIS, and reconciles TDS-visible receipts against declared income in both directions. What you are buying is not a field extractor. It is a defensible income view: every figure traced to a page, and every gap either explained by another document or flagged.
Key facts
- The volume is real and growing. Over 7.28 crore income tax returns were filed for AY 2024-25 up to 31 July 2024, up 7.5% on the 6.77 crore filed for AY 2023-24, with 58.57 lakh first-time filers (PIB, 2 August 2024). Almost every MSME credit file in India now starts with a return.
- Form 26AS is not what it was. From AY 2023-24 onwards, 26AS on TRACES shows only TDS/TCS information; the SFT, foreign remittance and GST turnover data moved to the Annual Information Statement (Income Tax Department, AIS FAQs). A tool that pulls only 26AS is pulling a slimmer document than it used to.
- AIS carries five information blocks — TDS/TCS, SFT, payment of taxes, demand and refund, and other information — and downloads in PDF, JSON and CSV. JSON is the format an analyser should be ingesting; PDF parsing is a fallback, not a design choice.
- TIS is the aggregated view, and it has two columns. The Taxpayer Information Summary shows both the system-processed value and the taxpayer-accepted value per category. Where those two differ, the taxpayer has filed feedback disputing a reported figure — which is underwriting signal, not noise.
- Extraction accuracy is the floor, not the product. YuSight extracts at 95.2% accuracy validated against a manual benchmark, with every figure carrying a citation back to its source document and page. Accuracy without traceability just moves the checking work rather than removing it.
What does an ITR and 26AS analyser actually have to do?
Seven jobs, in order. Most tools do the first two well and stop.
- Classify and route. Identify the form (ITR-1 through ITR-7), the assessment year, the PAN, the filing date and the acknowledgement number — and map each document to the right borrower entity when the file contains the firm, two partners and a guarantor.
- Parse the schedules, not just the summary page. The income figure on page one is the least interesting number in the return.
- Pull 26AS and AIS/TIS for the same PAN and the same year — the second half of the pair, which the borrower is far less likely to volunteer.
- Reconcile in both directions. Declared income not visible in 26AS, and 26AS receipts not declared in the return. The second direction is the one that finds problems.
- Cross-check against GST and bank credits. Three sources, one turnover.
- Compute the credit-relevant derivations — sustainable income, obligations from Schedule BP and the balance sheet schedules, depreciation add-backs, partner remuneration and interest.
- Produce a traceable output. Every figure clickable back to a document and a page, so a second reviewer or an auditor can verify it without reopening the PDF pack.
Which forms and schedules must it read?
Form | Who files it | Schedules that carry the credit-relevant numbers |
|---|---|---|
ITR-1 (Sahaj) | Salaried individuals, income up to ₹50 lakh | Salary detail, income from other sources, TDS schedules |
ITR-2 | Individuals/HUF without business income | Schedule CG, Schedule OS, Schedule AL where applicable |
ITR-3 | Individuals/HUF with business or professional income | Schedule BP, Schedule P&L, Schedule BS, Schedule DEP, Schedule OI |
ITR-4 (Sugam) | Presumptive income under 44AD/44ADA/44AE | Presumptive schedules — no balance sheet worth spreading, which is itself a finding |
ITR-5 | Firms, LLPs, AOPs, BOIs | Schedule P&L, Schedule BS, Schedule BP, partner remuneration and interest detail |
ITR-6 | Companies other than those claiming s.11 exemption | Schedule P&L, Schedule BS, Schedule BP, Schedule DPM/DOA, Schedule ICDS |
ITR-7 | Trusts, political parties, institutions | Rarely a commercial borrower; still needs classification so it is not mis-mapped |
Three schedule-level tests separate an analyser from a parser. Schedule OI (Other Information) carries disallowances and audit-related disclosures that never reach the P&L summary. Schedule DEP / DPM gives the depreciation add-back you need for cash accruals. And on ITR-4, the absence of a balance sheet means presumptive income under section 44AD is a declaration, not a computation — treat 8% or 6% of turnover as a tax election, never as evidence of margin. The form-by-form reading is set out in ITR analysis for loan underwriting: reading ITR-3, ITR-5 and ITR-6.
How does the reconciliation actually work?
Illustrative figures constructed to show the arithmetic. Not drawn from any actual borrower.
Suryodaya Fabrication LLP, ITR-5, FY 2024-25.
Declared gross receipts, Schedule P&L ₹8,45,00,000
TDS-visible receipts from Form 26AS
Sec 194C contract payments ₹4,62,00,000 TDS ₹9,24,000
Sec 194J professional/technical ₹36,00,000 TDS ₹3,60,000
Sec 194Q purchase of goods ₹1,20,00,000 TDS ₹12,000
Total 26AS-visible receipts ₹6,18,00,000 TDS ₹12,96,000
Gap (declared − 26AS-visible) ₹2,27,00,000
as a share of declared receipts 26.9%
A 26.9% gap is not a red flag on its own. Most of a manufacturer's customers are not required to deduct. The test is whether the gap can be closed with another document:
Explaining the gap
Exports — no Indian TDS applies, per GSTR-1 Table 6A ₹35,00,000
B2C and buyers below the ₹50 lakh s.194Q threshold,
per GSTR-1 B2C tables ₹1,92,00,000
Total explained ₹2,27,00,000
Residual unexplained ₹0
Now run it the other way — the direction that finds things:
26AS entries with no corresponding line in the return
Sec 194A interest on deposits, an NBFC ₹4,80,000 TDS ₹48,000
→ no matching entry in Schedule OS
Implied principal at ~7.5% p.a. (4,80,000 ÷ 0.075) ₹64,00,000
A ₹64 lakh deposit earning interest that the borrower did not declare is either an undisclosed asset or, far more often, a margin deposit or lien against a facility from another lender that is not on the application. Either way it changes the file. This reverse test is the single highest-yield check in the whole exercise, and it is the one most tools skip because it requires the 26AS to be read as a set of records rather than as a total. The method is worked through in Form 26AS analysis for lenders.
One more 26AS section to hard-code: any TDS under section 194N means cash withdrawals crossed the specified threshold. On a borrower who has told you the business is fully banked, that entry contradicts the story before you open a bank statement.
What has to be cross-checked outside the tax file?
An income view built only from tax documents is a view of what the borrower told the tax authority. Two more sources:
- GST returns. GSTR-1 outward supplies and GSTR-3B liability, reconciled to the ITR's declared turnover. The mismatch patterns and what each one means are covered in GSTR-1 vs GSTR-3B for credit assessment, and the wider framework in GST return analysis for lending. If you are evaluating the GST side separately, see GST analyser tools for lenders.
- Bank credits. Genuine business credits in the statement against declared turnover, net of inter-account transfers, loan disbursements and reversals.
Three sources, one number, three different definitions of "turnover" — accrual in the ITR, invoice-date in GSTR-1, cash-receipt in the bank. A tool that reports the three side by side without normalising the timing basis produces a mismatch that is arithmetic, not fraud. Three-way triangulation sets out how to do the normalisation properly.
The capability checklist, with a test for each
Take this into the POC. Score each line on the vendor's system with your own documents, not their demo set.
# | Capability | How to test it in an hour |
|---|---|---|
1 | Classifies ITR-1 to ITR-7, AY, PAN and acknowledgement number | Upload a mixed folder with two AYs of the same PAN and a wrongly-named file. Check nothing is silently mis-assigned |
2 | Reads schedules, not just the summary | Ask for Schedule OI disallowances and Schedule DEP depreciation. If they are not fields, the tool is a summary reader |
3 | Handles ITR-4 correctly | Confirm it flags presumptive filing rather than spreading 8% as a real margin |
4 | Pulls 26AS and AIS/TIS for the same PAN and AY | Ask which one it fetches. Many fetch only 26AS |
5 | Ingests AIS in JSON, not scraped PDF | Ask to see the ingestion format. JSON survives a portal layout change; a PDF scraper does not |
6 | Surfaces the TIS processed-value vs accepted-value difference | Give it a PAN where the taxpayer filed AIS feedback |
7 | Reconciles declared income to 26AS in both directions | Plant a 194A interest entry absent from the return and see whether it is reported |
8 | Section-wise 26AS breakdown, including 194N | Check 194N appears as its own flagged item, not folded into a TDS total |
9 | Maps documents to the right entity in a multi-entity file | Upload the LLP, both partners and a guarantor together — the failure mode is covered in multi-entity document mapping |
10 | Detects tampering — digital signature, XMP metadata, font substitution | Submit a lightly edited PDF. Silence here is disqualifying; see how to detect a fake or tampered document |
11 | Every output figure cites a document and a page | Click any number in the output. If it does not open the source page, your analyst still has to re-verify by hand |
12 | Analyst can override, with the override recorded | Change a figure. Check the original, the new value, the user and the timestamp are all retained |
13 | Field-level accuracy is measurable, not a marketing claim | Ask for accuracy per field on your 200-document set, not overall accuracy on theirs — the method is in extraction accuracy vs straight-through rate |
14 | Consent and audit trail for every fetch | Ask what consent artefact is stored and for how long |
Items 4, 7 and 11 are the ones that fail most often, and they are the three that decide whether the output is an income view or a pile of extracted fields.
Which tools do this in India?
Only own-site claims are stated as fact below. Anything a vendor does not publish on its own product page is shown as Not stated.
Platform | What its own site states | ITR + 26AS reconciliation |
|---|---|---|
Perfios | Analyses "all ITR Forms 1-7 and can plot financial spreads out of ITR 3-7 in bespoke formats"; captures via "statement upload (pdf/scanned) or direct fetch modes"; retrieves historical annual income "for the last 10 years"; verifies tax filing status, address and PAN from ITR-V (product page) | The ITR document-analysis page does not mention Form 26AS or ITR-to-26AS reconciliation |
Surepass | Form 26AS API "fetch the Form 26AS of a client for a given financial period in JSON, PDF, or XML formats" from the Income Tax Department (product page) | A retrieval API rather than an analyser. The consent and authentication mechanism is not described on the page |
Signzy | Markets ITR and 26AS retrieval APIs | Could not verify the capability on signzy.com itself at the time of writing |
FinBox | Markets lending-infrastructure data products including income verification | Could not verify ITR/26AS specifics on finbox.in itself |
YuSight | Document Intelligence classifies and validates tax returns, bank statements, financial statements and bureau reports, maps them to the right borrower entity in multi-entity structures, and spreads them with 95.2% extraction accuracy and every figure cited to its source document and page | Reconciliation runs as part of spreading; the output feeds the CAM with citations intact |
How to read this table. Retrieval and analysis are different products, and most Indian lenders end up buying both — an API layer that fetches the document and an analysis layer that reads it. The question to put to each vendor is which of the two they are, because a "26AS API" and a "26AS analyser" solve different halves of the problem and the pricing models are not comparable.
How does the document reach you in the first place?
Three routes, and the choice constrains everything downstream.
Route | What you get | Constraint |
|---|---|---|
Borrower-supplied PDF | ITR-V, full return, 26AS printout | Highest tamper risk; needs digital-signature and metadata validation |
Direct fetch with borrower credentials | Machine-readable ITR JSON, 26AS, AIS | Depends on the borrower sharing e-filing portal access — get the consent artefact right |
Account Aggregator | Bank and GST data | AA does not currently carry income-tax data, so ITR and 26AS still arrive by another route |
The AA framework is worth understanding before you design the collection flow, because the parts of the file it does cover change what you need from the borrower — see Account Aggregator framework explained.
FAQ
Which software analyses ITR and Form 26AS together?
Several Indian platforms cover parts of it. Perfios publishes ITR analysis across forms 1 to 7 on its own site; Surepass publishes a Form 26AS retrieval API. YuSight handles both inside Document Intelligence and carries the reconciliation into the spread and the memo. Ask each vendor whether they retrieve, analyse, or both.
Can ITR PDFs be parsed automatically?
Yes, and reliably, because the forms are structured and the schedule layouts are stable within an assessment year. The harder problems are scanned or photographed copies, older assessment years with different layouts, and correctly identifying which entity in a multi-entity file the return belongs to.
How accurate is automated ITR extraction?
Ask for a field-level number measured on your documents. YuSight benchmarks at 95.2% accuracy against a manual baseline. Overall document-level accuracy figures are close to meaningless — a return can be 99% correct and still have the wrong turnover.
What is the difference between Form 26AS and AIS?
From AY 2023-24, Form 26AS carries only TDS and TCS information. The Annual Information Statement is broader, covering TDS/TCS, SFT reported transactions, tax payments, demands and refunds, and other information including GST turnover. If your process only pulls 26AS, you are missing the SFT data.
What is TIS and why should an underwriter care?
The Taxpayer Information Summary aggregates AIS information by category and shows two values — the system-processed figure and the value the taxpayer accepted. A difference between them means the taxpayer filed feedback disputing something a reporting entity said about them, which is worth reading before you accept the declared figure.
What does a gap between declared income and 26AS receipts mean?
Usually nothing, on its own. Most customers are not required to deduct TDS, and exports attract none. The test is whether the gap closes against GSTR-1 and bank credits. An unexplained residual is the finding — not the gap itself.
Which 26AS entries should always be flagged?
Section 194N, because it means cash withdrawals crossed the threshold. Section 194A interest that does not appear in the return, because it implies a deposit you were not told about. And any deductor name that is a lender, because it suggests a facility outside the declared obligations.
Can an ITR analyser replace the bank statement analyser?
No. The ITR tells you what was declared for a completed year; the bank statement tells you what actually moved and when, including the last three months. They answer different questions and the reconciliation between them is the point — neither substitutes for the other.
Do I need borrower consent to pull 26AS or AIS?
Yes, and you need to store the artefact. These are taxpayer records accessed through the e-filing portal, so the consent record, its scope and its retention period should be part of the credit file, not the vendor's system alone.
Key takeaways
- Buy the reconciliation, not the extraction. Field extraction is table stakes; a two-way ITR-to-26AS reconciliation with an explained residual is the product.
- Run the reverse test. Receipts in 26AS with no matching line in the return find undisclosed deposits, liens and facilities that the forward test never will.
- Confirm the tool pulls AIS and TIS, not just 26AS. Since AY 2023-24, 26AS alone is TDS/TCS only.
- Insist on citations. A figure you cannot click back to a page is a figure your analyst has to re-verify, which removes most of the saving.
- Test on your own documents, with a planted error, and measure accuracy per field.
Upload a messy document set and see it classified — [book a live demo](https://yuverse.ai/yusight).
