For many companies, tariff exposure is hidden inside thousands of customs entry numbers, spread across multiple freight forwarders, and trapped in disconnected ERP systems. As a result, finance and operations leaders often lack a clear, consolidated view of what they have paid, what they may be owed, and where compliance risks may be building. Justin M. Sherlock, Co-Founder of Caspian, argues that artificial intelligence (AI) is finally turning that opacity into clarity.
“Most of the reasons that companies aren’t able to capture duty mitigation is because they don’t have the record storage or they haven’t centralized these data sets,” Sherlock says. “The refund might be there, but because the data chase is so painful, its hard to observe and capture.” Sherlock’s perspective is shaped by eight years spent working alongside finance and operations leaders to identify total tariff spend and reduce it. A former private equity investor and leader within Flexport’s capital arm, he became a licensed customs broker to understand the mechanics of trade compliance firsthand.
Why Trade Data Becomes a Black Box
His approach — that AI does not magically create refunds but instead makes fragmented trade data usable — comes into sharper focus when examining how trade data is handled inside most organizations. The modern supply chain is vendor-dependent by design, with companies relying on multiple customs brokers and freight forwarders that each maintain separate systems, documents, and data formats. Import data often sits with procurement or operations teams, while export data may live inside sales or compliance functions. These silos make it difficult to calculate total tariff exposure across countries, products, and shipments, reinforcing the very opacity AI is meant to resolve.
“You have a fragmented set of supply chain vendors who each maintain their own data and documents,” he says. “Centralizing all that information across different logistics movements is difficult for organizations to execute.” Even when a significant refund opportunity exists, companies cannot substantiate a claim without linking purchase orders to entry numbers, shipments to products, and product classifications to duties paid. The result is a persistent black box that obscures both overpayments and compliance risk.
Structuring Imports and Exports for AI Insight
Sherlock frames the solution as a data matrix. Companies must address three categories of information: imports, exports, and product data. Each category exists across both document-based and electronic formats, and without alignment across that grid, AI cannot deliver meaningful output.
On the import side, the foundational task is linking purchase orders sent to suppliers with actual shipment entries. Many organizations possess both sets of documents but lack the connective tissue between them. “Being able to tie orders that you’ve sent to suppliers with actual shipments and entry numbers is something that a lot of organizations haven’t really built into their ERP,” Sherlock says. AI can match those data sets, but only if they are collected and structured consistently.
Companies also often struggle to connect shipment identifiers, such as bills of lading, to sales order reports and product detail, revealing a parallel issue in export data. Without those identifiers stored in a systematic way, tracing an exported product back to an imported component becomes nearly impossible. A reliable refund program depends on moving fluidly between shipment-level documentation and product-level detail.
The Product Database as a Financial Asset
If import and export data provide the skeleton, product classification data supplies the muscle. Sherlock emphasizes the importance of a living product database that includes codes, descriptions, component quantities, values, and approved tariff classifications for both finished goods and their inputs.
That database enables precise mapping between imports and exports at the product level. It also exposes misclassifications where companies may be paying the wrong duty rate. “Those opportunities are the things that AI does a really good job of finding,” Sherlock says. “But it all depends on having that structured data in the first place.” Product data is not merely an operational record. It is a financial asset that determines whether a company can defend a refund claim or identify overpayment patterns. Without it, even the most sophisticated analytics tools fall short.
From Reactive Refunds to Proactive Avoidance
Over the next three to five years, Sherlock expects AI to move beyond data structuring into agent-driven trade compliance. Historically, companies have relied on external consultants charging several thousand dollars per hour for complex analysis. As AI systems mature, internal teams will be able to query intelligent agents for guidance and scenario modeling, reducing dependence on brokers for routine mitigation work.
More importantly, the best-run teams will integrate refund insights upstream. Instead of reacting to policy changes, they will embed tariff mitigation into sourcing decisions, contract negotiations, and even product design. Sherlock points to tariff engineering and free trade agreement optimization as areas where structured data can inform strategic choices before duties are incurred.
Regulatory volatility adds urgency, and as trade regulations grow more complex, Sherlock sees AI evolving into an early warning system that surfaces product-specific risks before they materialize in headlines or policy announcements. “How can you anticipate the future better?” he asks, framing predictive insight as the next frontier for trade technology.
The black box of trade data is not disappearing. Supply chains will remain fragmented, and vendors will continue to operate disparate systems. What is changing is the ability to centralize, structure, and interrogate that data continuously rather than annually. For organizations willing to treat trade information as a strategic dataset rather than a compliance afterthought, AI offers visibility, defensibility, and cash returned to the balance sheet.
Follow Justin M. Sherlock on LinkedIn or his website for more insights.
