Ajwa Group case study: AI import duty landed-cost agent prepared 36 variance packets

What did the OPAG import duty landed-cost agent do for Ajwa Group?

Ajwa Group operates across product categories where import cost can move for different reasons: oil-related shipments, automotive parts, electronics, agriculture inputs, frozen foods, spices, confectionery packaging, and FMCG stock can each carry different duty, freight, tax, insurance, and receiving evidence.

OPAG narrowed the workflow to one agent capability: prepare a governed landed-cost variance packet whenever customs, freight, supplier invoice, exchange-rate, GRN, or allocation evidence suggests that stock cost, margin, recovery, or approval treatment needs review.

The answer-first summary is this: OPAG used governed AI to turn import cost review into a source-linked operating workflow with human approval, role-based access, override reasons, and audit trails instead of leaving landed-cost decisions spread across email, customs files, spreadsheets, ERP comments, and warehouse notes.

Why does landed-cost variance AI matter for import-heavy groups?

Import cost review is not only an accounting task. Procurement may know supplier terms, import teams may know customs treatment, finance may know exchange-rate policy, warehouse teams may know received quantities, and commercial leaders may care about margin impact. When those signals stay disconnected, avoidable cost leakage can enter inventory and customer pricing.

The agent helped reviewers separate normal landed-cost movement from exceptions such as unexpected duty variance, unsupported freight charge, wrong allocation base, invoice-to-GRN mismatch, exchange-rate policy gap, missing insurance evidence, supplier recovery opportunity, or ERP posting that needed manager approval.

• Finance teams needed invoice variance, duty evidence, exchange-rate treatment, allocation logic, stock-cost impact, and journal approval context.
• Procurement teams needed supplier terms, PO references, freight responsibility, claim ownership, and supplier recovery context.
• Import and customs teams needed customs entries, HS codes, duty tables, insurance evidence, shipment files, and broker notes.
• Warehouse teams needed GRNs, received quantity, damaged or short shipment evidence, lot status, and stock-hold context.
• Leadership needed margin impact, category exposure, repeated supplier patterns, approval thresholds, and audit-ready decisions.

How did the agent prepare 36 landed-cost variance packets?

The workflow started with approved source boundaries and role-based access. Finance saw cost, allocation, and posting evidence. Procurement saw supplier terms and recovery ownership. Import teams saw customs and shipment files. Warehouse saw receiving evidence. Managers saw approval packets when cost impact, ERP posting, recovery, or policy exception crossed a threshold.

Each packet included supplier, shipment, SKU or lot, PO, invoice, customs reference, HS code, duty line, freight line, insurance charge, exchange-rate source, GRN reference, allocation basis, expected cost impact, recommended owner, approval requirement, and audit history.

• Scan: review POs, import invoices, customs entries, HS codes, duty tables, freight bills, insurance charges, exchange rates, GRNs, inventory records, supplier terms, and finance policy.
• Score: rank packets by cost variance, margin impact, stock value exposure, shipment urgency, recovery potential, evidence completeness, exchange-rate sensitivity, and approval threshold.
• Draft: prepare a source-linked landed-cost packet with variance driver, missing evidence, allowed actions, recommended owner, and ERP posting status.
• Route: send invoice issues to finance, freight recovery to procurement, customs questions to import owners, receiving mismatches to warehouse, and high-value changes to management approval.
• Audit: record source retrieval, generated packet, reviewer edits, approval decision, ERP action, supplier or broker follow-up, override reason, and final variance status.

What governance kept customs and finance decisions under control?

A landed-cost agent should not quietly change duty treatment, update inventory values, post journals, create supplier claims, dispute freight charges, accept broker explanations, contact customs partners, or approve write-offs. Those actions affect inventory valuation, supplier relationships, tax exposure, margins, and audit evidence.

OPAG separated evidence preparation from decision authority. The agent could explain which invoice, customs entry, HS code, duty table, freight bill, exchange rate, GRN, allocation rule, or approval policy created a variance, but authorized reviewers retained control over postings, claims, partner messages, and final treatment.

• Role-based access separated finance, procurement, import, customs, warehouse, operations, and leadership context.
• Source evidence showed whether a variance was driven by invoice price, duty line, HS code, freight bill, insurance charge, exchange rate, GRN quantity, or allocation rule.
• Approval gates protected customs treatment, inventory valuation, ERP postings, supplier claims, freight disputes, broker messages, write-offs, and policy exceptions.
• Segregation-of-duties rules prevented the same user from preparing evidence, approving cost treatment, posting ERP changes, and closing exceptions without oversight.
• Audit trails preserved the packet, sources, reviewer comments, approval route, final treatment, ERP action, and override reason.

Which OPAG services connect to import duty landed-cost AI?

The landed-cost agent shows how OPAG connects import evidence to controlled finance action. Conversational AI answers source-linked questions, Predictive AI ranks variance and margin impact, and Agentic AI routes each packet through accountable approval gates.

The same pattern can support FMCG groups, distributors, food businesses, oil-related operations, automotive parts suppliers, electronics distributors, procurement teams, warehouse teams, import teams, and finance shared services.

• Conversational AI: source-linked answers about import invoices, customs entries, freight bills, GRNs, allocation rules, and approval status.
• Predictive AI: variance, margin, recovery, posting, and evidence-quality scoring.
• Agentic AI: routed review packets, approval gates, exception ownership, override capture, and audit trails.
• Indirect tax exception AI: tax-sensitive invoice and treatment review before filing or posting.
• Accounts payable exception AI: invoice matching, approval routing, and source evidence for finance teams.

What can another import-heavy finance team copy from this case study?

The practical first workflow is not broad autonomous import automation. It is one high-friction decision where cost evidence is fragmented: duty variance, freight variance, invoice-to-GRN mismatch, exchange-rate treatment, or stock-cost allocation approval.

OPAG usually recommends starting read-only. Once reviewers trust the packet quality, the workflow can expand into supplier recovery, freight dispute preparation, ERP posting requests, category margin reporting, and management approvals.

• Start with one repeated landed-cost decision that already creates manual follow-up.
• Connect only the approved sources needed for that decision: PO, invoice, customs, freight, exchange rate, GRN, inventory, and policy.
• Define protected actions: duty treatment, ERP posting, supplier claim, freight dispute, inventory value update, and write-off require human approval.
• Measure variance aging, reviewer time, recovered value, stock-cost accuracy, override rate, and audit completeness.
• Expand after finance, import, procurement, and warehouse teams trust the evidence and routing model.