Introducing Validator: an AI-Native Platform for Export Letters of Credit

We just launched Validator, an AI-native platform for export letters of credit. We have run over $1B in live L/Cs through it so far in 2026, and we are finally ready to release it.

Built for corporates

Almost every L/C examination tool on the market was designed for a bank's trade operations desk, where the job is to check documents that already exist. Exporters live earlier in the timeline. They receive the credit, decide whether to accept it, produce the documents, and carry the payment risk when the presentation bounces. Validator is built for that side of the transaction.

End to end, in three stages

• Workability review the moment the L/C arrives. Validator reads the credit and flags the terms that cannot be met as drafted: a latest shipment date that is too tight, a document nobody in the corridor can issue, a clause that contradicts the sales contract. Amendments are cheap at this point; discrepancies later are not.
• Automated document generation and retrieval. Invoices, packing lists and beneficiary certificates are drafted from the credit's own data, so they start compliant. For third-party documents such as transport documents and inspection certificates, Validator produces the exact instructions the carrier or inspector needs.
• Full pre-check before bank submission. The complete set is examined against the credit, UCP 600, eUCP and ISBP 821E, field by field and across documents. Every finding names the document, the field and the rule behind it.

Validator is SOC 2 certified. Client logic learned from a company's own transactions is fenced to that client and never used to train the general model; we will write more about that architecture in a later post.

The result

Errors get caught before they happen, and exporters get paid faster. The ICC's long-standing estimate is that 60% to 80% of L/C presentations are refused on first submission. Every one of those refusals carries fees, a courier loop and a payment delay measured in weeks. A presentation that is clean on the first attempt collapses that cycle.

Why we built it this way

We spent 2026 running live letters of credit through the system before releasing it, because an examination tool earns trust on real presentations or not at all. Over a billion dollars of credits later, three design decisions have held up. The first is timing: catching a problem the day the credit arrives is worth more than catching it perfectly the day before presentation, so Validator starts at arrival. The second is provenance: every finding cites the document, the field and the rule or practice behind it, because a finding an examiner cannot verify in seconds is a finding they will redo by hand. The third is isolation: what Validator learns about one exporter's corridors, counterparties and documents stays with that exporter, is assigned to them as their IP, and is never used to train the general model.

The rest of the platform follows from those three. Generating documents from the credit is a timing decision. Explaining every decision is a provenance decision. Fencing client logic is an isolation decision. We will take each of them in turn over the coming weeks.

More on how each of these works over the coming weeks. If you present letters of credit today, the fastest way to see the difference is to run a free workability check on a credit you have in hand.