The methodology behind the recovery.

The five things we do, in detail

On the home page we describe the product as five steps: pull bills, classify HS codes, reconcile RoDTEP and eBRC, find schemes, alert before deadlines. Here is each in depth.

One: Pull your shipping bills

In Phase 1, you upload shipping bills as a CSV (template at /audit-free) or PDF (we parse the standard ICEGATE shipping bill format and most CHA-generated formats). Your CA can also share files into your workspace directly.

In Phase 2 (Q3 FY 2025-26 target), we connect to ICEGATE directly via CSP credentials, after CBIC onboarding completes. Until then, the upload flow is the source of truth. We do not log in to ICEGATE for you, even when the CSP integration is live; we pull metadata only.

Two: Classify every HS code

For each shipping bill line item, the classifier runs the product description (and where present, the buyer-side purchase order description) through an LLM prompt tuned specifically to the Indian Customs Tariff Schedule, ITC (HS), Appendix 4R, and Appendix 4RE. The model returns up to five candidate HS-8 codes ranked by confidence.

Every candidate includes:

• The HS-8 code and the official tariff description.
• The current RoDTEP rate, with citation to the row in Appendix 4R or 4RE.
• The current Drawback All Industry Rate.
• Where applicable, GST refund eligibility flag.
• A reasoning trace (2 to 3 sentences explaining why the code fits).
• A "common confusion" note: which adjacent HS code this product is sometimes wrongly classified as, and why.

The classifier never invents a rate. If the rate cannot be verified against the current Appendix, the response reads "rate unverified, check Notification No. X dated DD.MM.YYYY" with the relevant notification reference.

Three: Reconcile RoDTEP and eBRC

For each shipping bill, we maintain a timeline: filed, ICEGATE confirmed, EGM filed, DGFT scrip issued, scrip value credited to e-scrip wallet, scrip utilised against customs duty (or sold), and bank realisation against export proceeds linked via eBRC.

The reconciliation engine matches your shipping bill data against (a) DGFT scrip ledger entries pulled from the public DGFT portal, (b) eBRC data shared by your AD bank (manual upload in Phase 1, API in Phase 2), and (c) EDPMS status if you grant access via your CA. Tolerances are configurable (we default to plus or minus 1 percent on FOB equivalence, matching RBI's EDPMS tolerance).

Gaps are highlighted in coral with the suggested next action. For example: "Shipping bill no. SB-247845 dated 14 Sep 2025, RoDTEP value of Rs 8,420 not credited within T+45 days. Possible cause: EGM filing delay by carrier. Suggested action: contact carrier and request EGM-XML re-upload, copy your CHA."

Four: Find every scheme you qualify for

We maintain a scheme eligibility engine that takes your IEC, your sector, your top HS codes, your turnover, your destinations, your district of manufacture, and your RCMC list. The engine cross-references against the current FTP 2023 scheme matrix and returns a ranked list of schemes you are eligible for, with estimated annual benefit and next application deadline.

The scheme list includes: RoDTEP, RoSCTL, Duty Drawback, EPCG, Advance Authorisation, NIRYAT PROTSAHAN, TMA for Specified Agriculture Products, Districts as Export Hubs flagship product programmes, EPC-specific schemes (APEDA, EEPC, MPEDA, GJEPC, Pharmexcil, EPCH, HEPC, CLE, etc.), FTA preference utilisation (UAE, Australia, EFTA, UK), and where active, sector PLI schemes.

Five: Alert before deadlines

Three deadline categories are tracked: (a) renewal deadlines (RCMC, IEC, EPCG block periods, AA closure periods), (b) scheme application deadlines (quarterly TMA, annual NIRYAT PROTSAHAN audit, etc.), and (c) reactive deadlines (DGFT trade notice publication, RoDTEP rate change effective dates that affect already-filed bills).

Alerts go via your preferred channel: email by default, WhatsApp on the Growth and Pro plans, SMS as a fallback. Languages: English, Hindi, Tamil, Telugu, Gujarati. The WhatsApp templates are approved through the Gupshup Business API.

How our HS code classifier actually works

The classifier uses an LLM (currently Claude Sonnet 4.5) with a system prompt that constrains the response to a structured JSON schema. The prompt includes:

• The full Indian Customs Tariff Schedule (ITC HS) chapter and heading directory as compressed context.
• The current Appendix 4R rate matrix for HS-8 codes the product is most likely to map to.
• Negative list flags for the relevant chapter.
• Last 12 months of rate-change notifications relevant to the chapter.
• Hard constraints: cite the row in Appendix 4R or 4RE for every rate stated. Do not invent rates. Reply only in valid JSON.

The model's response is parsed, validated against the schema, and the rate fields are cross-checked against our own copy of Appendix 4R before being shown to you. If the model cites a rate that does not match our table, the suggestion is downgraded to "rate requires verification" and surfaced to the human review queue.

Confidence calibration: we ran the classifier on 1,800 historical shipping bills with known-correct HS codes from beta partner CAs. The classifier's top-1 accuracy at 90 percent confidence threshold is 97.4 percent. The top-3 accuracy (correct code in any of top 3 suggestions) at any confidence is 99.1 percent. Below 80 percent confidence, we flag for manual review.

How we reconcile RoDTEP, in operational sequence

1. Shipping bill ingestion. Filed-with-customs status confirmed via uploaded shipping bill or ICEGATE pull.
2. RoDTEP declaration check. Field 22 (RoDTEP declaration) must be populated and set to "Y". If "N" or blank, claim is flagged immediately.
3. HS code rate lookup. Rate effective on shipping bill filing date is applied (not the current rate).
4. Eligibility validation. Negative list, scheme exclusion list (no AA on same line, etc.), value cap check.
5. EGM confirmation. Carrier-filed EGM tracked. T+7 day tolerance; T+15 day escalation.
6. Scrip generation tracking. DGFT scrip ledger checked daily. Expected T+30 to T+60.
7. Scrip value reconciliation. Expected RoDTEP versus actual scrip value. Variance > 5 percent flagged.
8. Scrip utilisation or sale. Tracked if you provide e-scrip wallet status; optional.

What we do not have access to

We are explicit about this because the alternative (vague promises of "integrated DGFT access") is the most common form of vendor mis-selling in this space.

• We do not log into your ICEGATE account. We do not file shipping bills.
• We do not log into your DGFT portal account. We do not apply for schemes on your behalf.
• We do not access your bank's net banking. eBRC ingestion is via bank-shared CSV or (later) AD bank API where you authorise it.
• We do not access EDPMS directly. EDPMS status is updated via your CA's RBI-provided extract.
• We do not have CSP certification yet for ICEGATE API. That is on the Phase 2 roadmap.

Where our data comes from

• Appendix 4R and 4RE. Pulled directly from DGFT publications. Updated within 24 hours of any DGFT notification.
• ITC HS schedule. CBIC-published. Refreshed annually with any chapter-level amendments.
• Customs Tariff Notifications. Tracked daily via CBIC website RSS and email alerts. Indexed by HS chapter for relevance routing.
• Sector-specific notifications. APEDA, MPEDA, EEPC, GJEPC, Pharmexcil, CLE, EPCH, HEPC. Tracked via council circular emails.

What happens when a rate changes

When DGFT issues a rate-change notification (typically by email Friday or Saturday for Monday effect), our DGFT scraper picks up the change within 30 minutes. We identify all users with shipping bills filed in the previous 45 days whose HS codes are affected, compute the financial delta, and push WhatsApp + email alerts the same day. The notification includes the new rate, the effective date, the affected shipping bill numbers, the rupee impact per shipping bill, and the suggested action (revision filing for already-claimed bills, etc.).

The audit trail, in case of customs query

Every action taken in the product is logged in an immutable audit log: classifications accepted, classifications overridden by your CA, scheme recommendations dismissed, reconciliation gaps acknowledged, alerts sent and acknowledged. The log is exportable as a tamper-evident PDF with hash chain. If customs queries a classification you accepted in our product, you can produce the exact suggestions surfaced, the citations referenced, the reviewer identity, and the timestamp. This is your defence.

What stops us from being a one-stop everything

We get asked this often. The answer is product discipline. The exporters we built this for are not looking for "more tools that promise more". They are looking for one tool that does one job very well, integrates with their existing CA and CHA, and does not get in the way. Every feature we add is tested against that constraint. The features that fail the test do not ship.