Auto Extraction for GST and Income Tax Notices: How LegalInk's Pipeline Works

By LegalInk Editorial ·

Every practising CA and tax advocate knows the routine. A notice arrives — sometimes a clean PDF from the GSTN portal, sometimes a scanned demand order with a smudged seal, occasionally a phone photograph forwarded by a panicked client at 11 pm. Before a single word of reply can be drafted, someone has to read the document for the section invoked, the quantum demanded, the response deadline, and the precise allegation. Multiply that across a compliance practice handling dozens of GSTR-3B mismatch notices, Section 148A inquiries, and Section 156 demands every month, and the administrative load becomes a genuine professional liability. This post walks through how LegalInk's auto extraction pipeline handles that load — from document upload and OCR processing through metadata extraction to the generation of a structured, jurisdiction-aware draft reply ready for review.

What "Auto Extraction" Actually Means in a Notice Context

The term gets used loosely across legal-tech marketing. In LegalInk's implementation it refers to a specific, layered process, not a generic document-reading feature. Understanding what the pipeline does — and what it explicitly does not do — is useful before you build it into your workflow.

The Three Extraction Targets

When a notice is uploaded to LegalInk's Reply to Notice tool, the system is not performing a keyword search. It is attempting to reliably extract three categories of structured metadata.

Statutory reference — the precise section, sub-section, or rule under which the notice is issued. For GST notices, this means distinguishing between a notice under Section 61 of the CGST Act, 2017 (scrutiny of returns), Section 73 (tax not paid, non-fraudulent cases), Section 74 (tax not paid due to fraud or wilful misstatement), or Section 148 of the Income Tax Act, 1961 (reassessment). These are not interchangeable in law, and the reply strategy differs materially across each.

Demand or allegation particulars — the tax period, the ITC amount disputed, the turnover discrepancy flagged, or the income figure alleged to have escaped assessment. This is the quantitative core of the notice, and it is the field most prone to misreading in manual workflows when figures are embedded in dense paragraphs or in tables that span page breaks.

Response deadline — the date by which a reply, documents, or appearance is required. Notice formats vary significantly between GST authorities and income tax jurisdictions. Some state GST authorities embed the deadline in a subordinate clause mid-paragraph; CBIC formats place it in a clearly demarcated row; certain ITAT and CIT(A) notices place deadlines on the cover page. The extraction layer is calibrated to locate deadline language across these variable positions.

Why OCR Quality Is the Foundational Constraint

Extraction accuracy is entirely dependent on the quality of OCR processing upstream. LegalInk applies pre-processing steps — deskewing, contrast normalisation, and resolution standardisation — before the extraction layer reads the document. In practice, a notice photographed on a phone in reasonable lighting will parse at substantially higher accuracy than one photographed in poor conditions or at an oblique angle. The system flags low-confidence extractions explicitly rather than silently inserting incorrect values, which matters more in a compliance context than a high confidence score on bad data ever could.

For notices that arrive as born-digital PDFs — as most CBIC portal notices, ITD-generated notices, and GSTN downloads now do — the OCR step is bypassed entirely and text is extracted directly, yielding near-complete metadata accuracy.

The GST Notice Landscape the Pipeline Is Built For

GST notice volumes have risen sharply since the GST Council's push for system-driven scrutiny of return discrepancies. GSTR-1 versus GSTR-3B mismatches, ITC reversal demands under Rule 42 and Rule 43 of the CGST Rules, 2017, and show-cause notices under Section 74 alleging suppression of turnover are among the highest-volume notice types in current practice.

Section-Level Differentiation in Draft Replies

The statutory consequences of Section 73 and Section 74 CGST Act notices are fundamentally different. Section 73 applies where non-payment or short payment is not attributable to fraud or wilful misstatement; the penalty exposure under Section 73(9) is capped at ten percent of the tax or ten thousand rupees, whichever is higher. Section 74, by contrast, addresses alleged fraud or wilful misstatement and carries a penalty equal to the tax payable under Section 74(9). A reply that is appropriate for a Section 73 notice — where voluntary payment before adjudication extinguishes the penalty — is not appropriate for a Section 74 notice, where the characterisation of intent is itself the central issue to be contested.

LegalInk's pipeline identifies the invoked section and routes the notice to the corresponding reply template and legal-argument framework. For a Section 74 notice, the draft reply generated through /reply-to-notice includes argument structure contesting the fraud characterisation — typically by reference to the requirement that the department establish positive evidence of suppression, not merely a discrepancy on the face of returns — rather than a quantum-only response. This section-level routing is what separates a structured extraction tool from a generic document-fill exercise.

GSTR-2A and GSTR-2B Mismatch Notices

A significant proportion of GST notices in current practice relate to ITC discrepancies between a taxpayer's GSTR-3B claims and the amounts reflected in GSTR-2A or GSTR-2B. The notice typically specifies the tax period, the ITC claimed, the ITC available as per GSTR-2B, and the difference. Extracting all three figures accurately — and mapping them to the correct tax period — is essential for generating a reply that engages with the figures the officer has actually used.

A practical example: a notice for FY 2019-20 alleging excess ITC of ₹4,82,300 may rely on GSTR-2A data that was itself updated after the relevant return period, in which case the assessee's reply will frame the response around the Supreme Court's reasoning in Union of India v. Bharti Airtel Ltd. and the CBIC's position in Circular 183/15/2022-GST. The extraction layer parses the disputed figure and tax period; the generation layer pulls the relevant circular into the draft. Where the notice uses a non-standard state GST format, the system flags the table for manual verification before the draft is finalised.

Section 61 Scrutiny and ASMT-10 Notices

Scrutiny under Section 61 read with Rule 99 typically begins with an ASMT-10 intimation listing discrepancies, followed by an ASMT-11 reply window. The pipeline extracts each enumerated discrepancy from the ASMT-10 and generates a paragraph-by-paragraph response in ASMT-11 format. Where the discrepancy relates to e-way bill data versus GSTR-1, the draft references the relevant tax period and invoice numbers extracted from the notice itself, rather than producing a generic explanatory paragraph.

The Income Tax Notice Pipeline

The Income Tax Act, 1961 generates a distinct set of notice types, each with its own procedural requirements and reply obligations. The three most common notice types in volume practice — Section 143(2) (scrutiny assessment), Section 148A (pre-reassessment inquiry), and Section 156 (demand notice) — are all handled within the automated reply workflow.

Section 148A and the Four-Step Sequence

The reassessment framework under Section 148A, as amended following the Finance Act 2021 and the Supreme Court's directions in Union of India v. Ashish Agarwal, creates a four-step procedural sequence: the Section 148A(a) inquiry, the Section 148A(b) notice requiring a response, the Section 148A(c) order, and finally the Section 148 notice if the authority proceeds. The reply to a Section 148A(b) notice is not merely a factual response. It is a structured legal submission that, if well-drafted, can prevent the authority from proceeding to issue a Section 148 notice at all.

When the pipeline receives a Section 148A(b) notice, it extracts the assessment year, the alleged income escaping assessment, the information or document that has triggered the proceeding, and any reference to the approval under Section 151. The draft frames the response around the requirement in Section 148A(b) itself: that the assessee be given an opportunity to show cause as to why a notice under Section 148 should not be issued. Where applicable, it raises threshold objections — limitation under the first proviso to Section 149, the adequacy of the "information suggesting income escaping assessment", and compliance with the sanction requirement under Section 151 — before engaging with the merits.

Section 143(2) Scrutiny Notices and Faceless Assessment

Under the faceless assessment framework standard under Section 143(3A) and the e-Assessment Scheme, scrutiny notices and replies are exchanged through the Income Tax portal. LegalInk's pipeline pulls the notice document from an uploaded file rather than requiring a portal integration, meaning the tool is portal-agnostic and works across multiple jurisdictions.

The extracted metadata from a Section 143(2) notice — assessment year, the specific income heads under examination, and the response date — feeds into the draft along with document-request checklists aligned with the issues flagged. If the notice raises questions on Section 56(2)(x) receipts, the checklist will include valuation reports, donor confirmations, and bank statements; if it raises Section 68 cash credit questions, the checklist shifts to creditor confirmations, ITRs, and source-of-source documentation.

Section 156 Demand Notices

Section 156 demand notices look procedurally simple but require careful handling because the thirty-day payment window under Section 220(1) starts from service of the notice, and a stay application under Section 220(6) must be filed before that window closes. The pipeline extracts the demand amount, the assessment year, and the date of service, and surfaces the Section 220(1) deadline alongside the reply deadline. The draft includes both a substantive response and, where appropriate, a parallel Section 220(6) stay application referencing the standard CBDT Office Memorandum dated 29 February 2016 on the twenty-percent pre-deposit norm.

From Extracted Metadata to Drafted Reply

Extraction is necessary but not sufficient. The output of the extraction layer — section invoked, demand amount, deadline, assessment year or tax period, and allegation type — becomes the structured input to the reply generation layer. This is where the practical value for a tax practice is most directly realised.

What the Draft Contains

A generated reply from LegalInk's finance compliance framework contains the following components as standard:

  • Jurisdictional and procedural opening that identifies the notice by reference number, date, and section, and acknowledges receipt without making admissions.
  • Factual response to the allegation that maps the assessee's position against the specific figures or discrepancies the authority has identified.
  • Legal submissions citing the applicable provision, relevant circulars, and where applicable, GST Council clarifications or CBDT instructions that bear on the issue.
  • Document annexure list itemising the documents being submitted in support, formatted in line with the response requirements of the relevant authority.
  • Relief sought stating clearly what the assessee is requesting — dropping of the proceeding, deletion of the demand, or correction of the mismatch — without ambiguity.

The draft is a working document, not a final filing. Practitioners are expected to review and adapt the legal submissions to the specific facts of their client's position. The tool removes the blank-page problem and the metadata-reading workload; it does not substitute professional judgment on the merits.

Version Control and Deadline Tracking

One feature that matters in a high-volume practice: every notice processed receives a deadline entry in the compliance dashboard at /compliance-framework/finance. The extracted deadline is stored and surfaced as a reminder, which materially reduces the risk of a missed response date. That risk is not academic — under GST, a missed reply to a Section 73 or 74 show-cause notice can result in an ex-parte adjudication order under Section 75(5), and the appellate route under Section 107 then requires a ten-percent pre-deposit before the demand can even be contested on the merits.

The dashboard supports multiple noticees and multiple tax periods within the same view, which is the format a CA handling ten or twenty clients' compliance portfolios actually needs. Drafts are version-controlled, so iterations between the partner reviewing the matter and the associate preparing it remain auditable.

Limitations and Verification Obligations

An honest account of an extraction tool must address what it does not handle reliably, because misplaced confidence in automated output in a compliance context carries real risk.

Non-standard state GST notice formats remain the most common source of extraction error. States have significant latitude in how they format show-cause notices and demand orders, and some formats — particularly older pre-digitisation formats still in use in certain adjudicating authorities — yield lower-confidence extractions. The system flags these for manual review; practitioners should treat any flagged extraction as requiring line-by-line verification against the original document before the draft is used.

Multi-issue notices — notices that invoke multiple sections or contain demands across several tax periods in a single document — require careful review of the extraction output to confirm that every issue has been captured and that the draft addresses each one. The pipeline handles multi-issue notices, but the review step is non-negotiable. A common failure mode in manual practice is to respond comprehensively to the largest demand and inadvertently leave a smaller secondary issue unaddressed; the extraction output makes that omission visible by listing every extracted issue alongside the draft.

Penalty and interest calculations embedded in demand notices are extracted as figures but are not independently recalculated by the system. Verification of the arithmetic against Section 50 of the CGST Act (interest on delayed payment) or Sections 234A, 234B, and 234C of the Income Tax Act (interest for defaults in filing and payment) remains the practitioner's responsibility. The same applies to limitation computations under Section 73(10), Section 74(10), and the various provisos to Section 149 — these are surfaced as flags, not as conclusions.

LegalInk's position, stated plainly at legalink.co.in, is that these tools are practice-support infrastructure. The professional is the expert; the tool handles the volume and structural work that should not be consuming a skilled practitioner's time.

Why This Matters

The practical pressure on CAs and tax advocates handling notice-heavy practices is not primarily about drafting skill. It is about bandwidth. The ability to extract structured metadata from a notice accurately, route it to the correct legal framework, and produce a substantive first draft within the same workflow changes the economics of managing a high-volume compliance practice. It also reduces the specific risks that come with rushed or incomplete notice responses: missed deadlines, unaddressed sub-issues, replies that engage with the wrong version of the demand, and arithmetic that goes unverified. Done at the level of section-aware routing and compliance-integrated deadline tracking, auto extraction is not a convenience feature — it is a professional-risk management tool. Practitioners looking to integrate this into their workflow can begin with LegalInk's Reply to Notice tool and the broader finance compliance framework, both built precisely for this work.

Related posts

← Back to all posts