Drawback Advisor

Deterministic orchestrator. Surfaces CBP duty drawback opportunities on imports that have been (or could be) re-exported, destroyed, or used in manufacturing of exported goods. Up to 99% of duties + fees paid can be recovered under 19 USC 1313 / 19 CFR Part 190 (the TFTEA-modernized drawback regulation — Part 191 is the legacy rule, retained only for pre-2019 claims).

Most mid-market importers leave drawback on the table because the recordkeeping is punishing and broker contingency fees (15–25% of recovered amount) eat small claims. An automated agent flips that math — every dollar drawback-eligible becomes visible, classifiable, and one click from filing an electronic type-47 drawback entry in ACE (the modern equivalent of the legacy paper Form 7551).

Live now: a guided import wizard (paste / CSV / XLSX); import-side claim management + status workflow; export-record ingestion + an automatic import↔export matcher over the org's real cleared entries (unused-merchandise AND manufacturing/BOM); a §1313 provision-determination engine that derives the correct drawback provision from the data (never silently defaulting); a document-completeness checker per 19 C.F.R. § 190.51; and generation of a field-accurate ACE Drawback Entry Summary (CBP CATAIR/TFTEA v27, 80-char records) — plus a structured Form 7551 CSV / draft worksheet for provisions a broker files manually. A companion recoveries surface handles IEEPA-refund claims on the same book. Roadmap: actual ACE transmission to CBP (needs the broker's certified ABI connection).

What it does

Reads the importer's persisted entries (we already have these from the customs clearance flow) and identifies claims that fit one of four CBP-recognized drawback categories:

1. Direct identification — same article imported and then exported (often re-export of unsold inventory).
2. Substitution — similar commercial fungible goods substituted (8- digit HTS match works for most commodities).
3. Manufacturing — imported merchandise consumed in production of a subsequently exported good (e.g. components into finished goods).
4. Unused merchandise — imported and destroyed under CBP supervision without entering commerce.

For each match, computes the recoverable amount as 99% × (duty + fees) and tracks the claim through its lifecycle:

eligible → claimed (type-47 drawback entry filed in ACE) → paid (refund received)
                                                ↘ denied (CBP rejected)
                                                ↘ expired (5-year window closed)

Critical date: drawback must be filed within a uniform 5 years of the date of import (TFTEA standardized this; the old 3-year-from-export window survives only for the niche 1313(d) flavoring/spirits category). The agent surfaces every claim within 180 days of that 5-year window closing as "at risk of forfeiture."

Inputs

{
  // Per claim — derived at seed time + updated as broker takes actions
  shipmentId: string;          // the original import
  claimType: 'direct_id' | 'substitution' | 'manufacturing' | 'unused';
  dutyPaidCents: number;       // sum of HTSUS duty + Section 301 + IEEPA + Fentanyl
  feesPaidCents: number;       // MPF + HMF
  importDate: string;          // entry date — 5-year clock starts here
  // Optional, set as broker progresses
  exportDate?: string;         // export proof; the 5-year clock runs from import (TFTEA)
  cbpForm?: '7551' | '7552' | 'TFTEA';
  cbpClaimNumber?: string;
}

It also accepts (shipped):

• Export records — the proof-of-export list (ingest-exports), used for both 1313(j) unused-merchandise matching (Phase 1) and as the proof-of-export for finished manufactured goods (Phase 2).
• Products + bills of materials (ingest-products) — finished goods and their component parts, enabling manufacturing-drawback matching at the component level (19 USC 1313(a)/(b)).

Outputs

{
  kpis: {
    totalRecoverableCents: number;   // eligible + claimed
    paidYtdCents: number;            // refunds received this calendar year
    filedYtdCount: number;           // claims filed this year
    expiringSoonCount: number;       // < 180 days to expiry
    expiringSoonCents: number;
    byStatus: Record<'eligible'|'claimed'|'paid'|'expired'|'denied', { count, cents }>;
  };
  claims: DrawbackClaim[];           // full list with status, dates, recoverable, CBP refs
}

Per-claim shape includes daysToExpiry, so the UI can sort/filter by urgency at a glance.

When it runs

• Drawback recovery screen — primary surface, accessible from the sidebar (DATA section between Suppliers and Post-entry audit).
• GET /api/customs-clearance/drawback — KPI + list endpoint, used by the screen and any importer-facing portal.
• POST /api/customs-clearance/drawback/:id/status — broker advances the claim (eligible → claimed → paid). Writes filed_at / paid_at timestamps.

Planned — the agent will also fire automatically (today the matcher is broker-initiated):

• On entry release/cleared — assess if drawback-eligible (origin, commodity, importer's typical export pattern).
• On export-record ingestion — match against eligible imports.
• Nightly cron — re-check expiry windows on open claims, flag any crossing the 180-day boundary.

Dependencies

• Cleared shipment data (from the customs-clearance module) with duty + fees breakdown.
• CBP ADCVD case database — to handle a special case where AD/CVD cash deposits ARE drawback-eligible. (Roadmap, Q4.)
• HTS Classifier (Agent #1) — for substitution-drawback matching at the 8-digit subheading.
• Document Extractor (Agent #2) — for richer export-record ingestion (commercial invoice + BoL extraction); today ingestion is CSV, the document-driven path is planned (Q4).

This is an orchestrator in the same family as Agent #7 (Audit Responder) and Agent #8 (Landed Cost Simulator) — it composes the output of other agents into a user-facing artifact.

What success looks like

The industry-wide proxy: estimated $1–2B+ left on the table annually by US importers who don't file drawback they're entitled to. The agent's job is to make that money visible.

What it does NOT do

• Does NOT auto-file with CBP. Filing always requires broker review
  • sign-off. The UI ships with "File" / "Mark paid" buttons that record the state change but don't transmit.
• Automatically matches imports to exports over the org's real cleared entries — this is live. You ingest export records and the matcher pairs them against eligible imports (unused-merchandise and manufacturing/BOM), computes the lesser-of recovery, and surfaces unmatched high-duty imports as leads. What's still planned is auto-triggering the match without a broker kicking it off (on entry release, and a nightly expiry sweep) — see the roadmap.
• Uses the TFTEA substitution standard: same 8-digit HTS — except when the import's 8-digit subheading is described as "other," in which case substitution requires a matching 10-digit classification (and neither described as "other"). The matching service enforces this.
• Applies the lesser-of rule: recovery is capped at 99% of the lesser of (a) duties/fees paid on the import, or (b) duties/fees that would apply to the exported article if it were imported. For same-HTS substitution these usually coincide; they diverge when trade-program stacking differs between the two.
• Does NOT handle complex substitution rules at the commercial-interchangeability standard (the old pre-TFTEA rule). We use the TFTEA "same 8-digit HTS" standard, which is the post-2018 default.
• Does NOT apply to all duty types. Some are non-drawback-eligible (e.g. some MPF amounts, certain Section 232 components). The recoverable formula uses 99% × (duty + fees) as a conservative ceiling; broker validates per-claim.
• Does NOT yet provide accelerated-payment financing (refund paid up front against a bond) — that's a future surety-carrier partnership. The generated entry already records and structures the AP / WPN / OTW privileges themselves.
• Manufacturing bill-of-materials matching is now shipped (Phase 2 — matchManufacturingDrawback). What it does NOT yet do: multi-level (nested) BOMs, yield/scrap factors, or specific-manufacturing-ruling constraints — v1 is a single-level BOM with value-apportioned per-line duty; a broker validates each claim.

Why this is a high-leverage agent

Most agents we ship are cost-aware — they help the broker file correctly the first time. This is the only agent that's revenue-aware — it surfaces cash that's already been paid to CBP and can be recovered.

For an importer doing $50M+ in dutiable imports per year:

• Conservative recovery rate: 5–10% of duty paid
• On $5M annual duty exposure: $250K–$500K recovered per year
• Broker contingency fee at industry standard 15–25%: $40K–$125K
• Importer net: $200K–$400K/yr

That's a renewable line item, year-over-year, that didn't exist before the agent. It changes the broker-customer conversation from "fee for service" to "revenue share on recovered duty."

Roadmap

• Phase 1 — Claims workflow. Live. Persisted claims, broker workflow (file → paid), KPI screen, manual broker review.
• Phase 1 — Unused-merchandise matcher. Live. Real export-record ingestion (CSV) plus a runtime import↔export matcher over the org's actual cleared entries: direct-ID (10-digit) vs substitution (8-digit, guarding .00 "other" baskets), lesser-of recovery, org-scoped writes, unmatched high-duty imports surfaced as leads. ingest-exports + match-drawback CLIs, POST /drawback/match.
• Phase 2 — Manufacturing / BOM matcher. Live. Products + BOMs (ingest-products) explode against the org's finished-good exports; imported components are attributed to exported goods with value-apportioned per-line duty, never over-claiming the imported quantity. match-manufacturing-drawback CLI, POST /drawback/match-manufacturing. Next: multi-level BOMs, yield/scrap factors, specific-manufacturing-ruling constraints.
• Phase 3 — ACE Drawback Entry Summary generation. Live. Filable claims are bundled by 19 USC 1313 provision into a field-accurate ACE Drawback Entry Summary — the CBP CATAIR/TFTEA v27 fixed-width format (80-char records, Appendix-B check digit, B…Y block), ready for the broker to upload into their own ABI software. generate-drawback-entry CLI, POST /drawback/entries, Entries tab. Honest boundary: this GENERATES the entry file; it does not transmit to CBP, so the CBP entry + claim numbers stay null until the broker transmits through their certified ABI connection.
• Provision-determination engine. Live. The correct §1313 provision is derived from the data (disposition → specialized → rejected → manufacturing → unused), never declared per importer; fact-based provisions are proposed for broker review rather than silently defaulted. A broker confirms the provision, which drives the generated entry — a confirm-to-file loop.
• Document-completeness checker. Live. For each claim, the required document set (per 19 C.F.R. § 190.51 / § 190.72) is computed from provision × disposition × waiver-of-prior-notice, and checked against the documents actually attached — so a broker sees exactly what's missing before filing.
• Structured Form 7551 export. Live. A CSV / draft worksheet for provisions a broker files manually, alongside the CATAIR file.
• Guided import wizard. Live. Import the drawback book by paste, CSV, or XLSX with column mapping, validation, and dedup — no rigid template.
• IEEPA-refund recoveries. Live. A companion recovery surface re-derives the vacated IEEPA / fentanyl duty on pre-halt China-origin entries (Learning Resources v. Trump) into per-entry refund claims (estimate; CBP sets final + interest).

Planned, after the above:

• Q4 2026 — Richer export-record ingestion via the Document Extractor (Agent #2) — auto-extract BoLs and AES filings, beyond the current wizard.
• Q4 2026 — ACE transmission so entries actually file with CBP (the remaining Phase 3 piece — needs the broker's certified ABI connection).
• Q1 2027 — TFTEA accelerated-payment financing (surety-carrier partnership) — refund up front against a bond. (The entry already records/structures the AP/WPN/OTW privileges; this adds the financing.)
• Multi-level (nested) BOMs, yield/scrap factors, and specific-manufacturing- ruling constraints for manufacturing drawback.

Strategic positioning

The Drawback Advisor turns Aduaria from "AI-native customs broker software" into "AI-native trade financial optimization." Same data, same agents, new surface — but the conversation with the importer shifts from "we save you time" to "we recover money you'd otherwise forfeit."

For a Port of Miami audience: their importer base running cleared entries through Miami → Latin America re-exports is exactly the profile where drawback matters most. Hub-and-spoke trade flows are drawback-heavy by structure. Pitching this as "your importers can now see and recover the drawback they're entitled to on Miami-routed cargo" is a real differentiator.