In Today’s Issue:

⛽ How a Gulf disruption can reach a domestic grocer through fuel contracts

🧭 What a logistics team can change when rerouting is impossible

🕒 The first 24 hours after a critical input is delayed

📨 What changed at Knight-Swift when an agent took over the shipment inbox

🔐 Which actions an agent may take and which require approval

📦 Why many agents chasing the same alternative can worsen a bottleneck

🧪 How to test a contingency plan before relying on it

🇪🇺 Pallet's approach to GDPR, data residency and AI-provider choice

Dear Readers,

A disruption at the Strait of Hormuz can reach a business that never sends a shipment through the Gulf. In Sushanth Raman's example, a grocer buying produce domestically feels it through diesel prices, a carrier's fuel surcharge and the timing of its own customer contracts. The extra cost can arrive before the business is able to pass it on.

The decisions that follow concern real orders and limited stock: which customer gets supplied, which production run waits and who may approve an expensive alternative. Raman, co-founder and CEO of Pallet, argues that AI agents can connect those decisions to information scattered across emails, shipment systems and inventory records.

His concrete example is at Knight-Swift, where, by his account, a Pallet agent handles more than 20,000 shipment-status emails a day. In a disruption, that inbox is the first thing to flood. Customers get an accurate answer in seconds, and operators have time for the exceptions that need them.

We sent Raman seven questions. He answered in writing, using Hormuz as a planning scenario and drawing on Pallet's work in other logistics operations. We asked what happens when there is no viable detour, how an agent should handle conflicting information, and whether companies using the same AI tools could make a shortage worse. Pallet later added an eighth question and answer on GDPR, data residency and the choice of AI provider.

All the best,

Kim Isenberg

Pallet builds AI agents for logistics operations. In these answers, Raman describes systems that follow up with suppliers and carriers, reconcile information with enterprise software, draft customer responses and route exceptions to people.

The Hormuz questions are a planning scenario. The Knight-Swift and STG Logistics examples come from other customer workflows, and the figures in them are Raman's own.

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get The Free Guide

Q1. How can disruption at Hormuz reach a company that neither imports oil nor ships through the Gulf? Please trace one concrete chain of effects, from the initial disruption to an operational problem or a higher price for an ordinary customer, and explain which link businesses tend to overlook.

Take a grocer buying produce domestically. Disruption at Hormuz constrains global oil and refined-product supply, putting pressure on diesel prices. The carrier delivering that produce faces higher fuel costs and passes them through under its surcharge agreements. The grocer then absorbs the cost in its margins or eventually passes some of it to shoppers.

The overlooked link is often the contract. Fuel costs can rise immediately, while customer pricing adjusts on a different schedule. That difference matters even for shipments that stay within one country because domestic contracts can also delay or limit how quickly higher fuel costs are passed on to customers. AI can help connect the external event to the actual lanes, contracts, and billing schedules affected, so supply chain teams understand their exposure and respond sooner.

Q2. “Reroute the shipment” sounds straightforward, but Hormuz is not a chokepoint where every cargo has a viable detour. When alternative capacity or supply simply is not available, what can a logistics team still change? Where does AI help make a difficult trade-off, and where would it merely produce an unworkable plan?

Teams can still change how they allocate available inventory, sequence production, prioritize orders, and communicate delivery commitments. Those decisions can protect service even when they can’t restore supply.

AI helps assemble the information needed to compare those options. For example, which products depend on the missing input, which customers have inventory left, or what would delaying one order mean for another customer, or for margin? It produces an unworkable plan when it assumes a substitute is qualified, inventory is available, or a carrier has capacity without checking.

The physical economy involves independent companies with different incentives and constraints. AI needs to work within those realities. People should retain authority over consequential decisions, such as allocating scarce supply between customers.

Q3. As a planning exercise, imagine a manufacturer learns that a critical input linked to Gulf supply will be delayed by several weeks. What should happen in the first 24 hours? Walk us through the decisions, the data needed, and which steps an AI agent could realistically execute today.

The first 24 hours move through four steps. First, confirm and quantify which POs are affected and how firm the delay is.

Next, map exposure from the input through BOMs to open customer orders, and calculate days of cover by product and site. Then lay out the options: inventory at other sites, qualified alternate suppliers, partial expedites, approved substitutes.

Finally, by the end of day, make the commitments that can't wait: allocation, soft reservations, early customer notice, and quote requests. The data required is mundane but scattered: BOMs, inventory by location, POs with real ETAs, order priorities and penalties, and supplier contracts.

An agent can realistically chase suppliers and carriers for updates, reconcile them against the ERP, calculate cover, flag affected orders, draft notices and request quotes, turning a two-day scramble into a morning.

With Pallet’s Signals, which we recently launched, that work can start before a supplier even calls: it connects external events like a port slowdown, tariff change or material shortage to a company’s own orders, inventory and capacity, and traces which customers and sites are exposed as conditions change. Deciding which key account gets shorted, or approving a substitution, stays with named people.

Terms: PO = purchase order; BOM = bill of materials; ERP = enterprise resource planning; ETA = estimated time of arrival.

Q4. Could you walk us through one real Pallet customer workflow during a disruption, even if it was unrelated to Hormuz? What did the team do before, what did the AI actually change, and what measurable result followed? An anonymized example is fine; please distinguish faster administration from an actual improvement in delivery, cost, or service.

At Knight-Swift, every disruption shows up first as an inbox full of "where's my load?" emails. Operators used to look up each shipment in a legacy AS/400 system and write replies by hand, so ETA and late-load emails were about 17% of volume but 44% of handling cost. Today a Pallet agent classifies each email, runs the lookup, and drafts the reply for one-click review, handling over 20,000 emails a day in under 30 seconds.

To be precise, that's faster administration: it doesn't make a truck arrive sooner. It does measurably improve service, since customers get accurate answers in seconds and handling cost no longer spikes with volume during a surge, and it frees operators for the exceptions that need a human.

Editor’s note: Pallet clarified that “under 30 seconds” refers to the automated processing of a single email: classifying the request, looking up the shipment and generating a draft reply for operator review. Human review and sending are excluded.

Q5. During a crisis, a carrier email, a shipment-tracking feed, and an internal inventory system may disagree. How should an agent decide what to trust, and when to stop and ask a person? Please give a concrete example of an action it may take autonomously versus a costly commitment that requires approval.

No single source is always right, so the agent should trust different sources for different facts, following rules written in advance. It needs to distinguish confirmed physical events from predicted ETAs and check timestamps when information conflicts.

That’s where agentic governance matters: the business defines what the agent can access, which actions it can take autonomously, and when it must seek approval. Actions should be traceable, with people able to intervene.

For example, if a carrier emails that a container missed its sailing but tracking still shows the old ETA, the agent can request confirmation, flag the discrepancy, and alert the planner. Once verified, it can update the internal ETA within its permissions, much as our agents at STG Logistics process routine orders automatically and route exceptions to operators. Booking expedited air freight or releasing inventory promised to another customer requires approval.

Q6. If many companies use AI to identify the same alternative supplier or transport capacity, could their responses make the bottleneck worse? How should planning account for competitors chasing the same options, rather than assuming that available capacity will still be there when a booking is made?

Yes, AI could amplify that risk. If many companies identify the same alternative supplier or lane, their responses can concentrate demand and make availability change faster. Duplicate orders and speculative bookings can further distort the picture.

Planning needs to distinguish quoted availability, reserved capacity, and confirmed commitments, while continuously rechecking availability before action. AI should surface feasible alternatives, account for rising prices and tighter capacity, and prevent duplicate bookings.

This is why company-specific context matters. The right response depends on your inventory, qualified suppliers, contracts, and customer commitments. At Pallet, we focus on connecting that context to execution, so agents can help teams act on options they can realistically secure. Preparation and supplier relationships remain essential, while AI helps businesses reassess and coordinate faster as conditions change.

Q7. What is one practical change a business should make before the next disruption, and one common contingency measure that creates false confidence? How would you decide whether the cost of extra inventory or backup capacity is justified, and what would tell you the plan is actually ready?

The practical change is to make exposure computable: connect your data so you can answer "which orders depend on this input, and when do we run out?" in hours, and assign decision rights in advance.

The contingency that creates false confidence is dual sourcing on paper, where both suppliers share the same upstream source, or the backup has never shipped real volume.

To justify a buffer, compare how long it would take to recover supply with how long you can survive without it, price that gap in lost margin, penalties, and customers, and weigh it against the holding cost, which is essentially an insurance premium.

You know the plan is ready when you've run it: simulate a four-week delay with real data, time how long the team takes to reach a decision, and confirm the backup supplier has shipped a real order recently.

Q8. How is Pallet addressing European companies’ concerns around GDPR, data residency, and dependence on American AI providers?

Pallet supports GDPR requirements through contractual safeguards, including Data Processing Agreements (DPAs) and EU Standard Contractual Clauses. For customers with specific data residency requirements, we can also support dedicated, on-prem deployments.

Our platform is model-agnostic, supporting multiple AI providers, open-weight models, and custom models. This gives enterprises greater flexibility and control over their AI infrastructure, rather than locking them into a single provider.

This interview was conducted in writing. The original seven answers were received on October 26, 2026. Pallet supplied Q8 and the timing clarification in a follow-up received on October 28, 2026. Answers are reproduced in full, with paragraph breaks and emphasis added; the note after Q4 is editorial.

Gmail tracks you. Proton doesn’t. Get private email that puts your data — and your privacy — first.

Ditch the Gmail data grab

Raman's sharpest point comes first: a disruption's real cost sits in the contract. A fuel-price increase can hit the carrier before the customer is able to adjust its own prices. An agent that links the external event to lanes, surcharge agreements and billing schedules shows a business what the disruption will cost, and when.

His first-day plan turns that into a routine. Agents chase suppliers and carriers, reconcile the answers with the ERP, calculate days of cover and draft customer notices, turning what used to be a two-day scramble into a morning. Named people make the calls on scarce supply and expensive alternatives.

AI can shorten the scramble to establish what has happened and who is affected.

Raman describes agents chasing updates, matching them to orders, calculating how long stock will last and preparing responses. His first-day plan runs on information most businesses already hold, only spread across different systems: purchase orders, bills of materials, inventory, contracts and customer priorities. Connecting those records is the work Pallet's agents are built for.

Knight-Swift shows what that looks like at volume. By Raman's account, a Pallet agent classifies more than 20,000 shipment emails a day, runs the lookup in a legacy system and prepares a draft for one-click review, so handling cost no longer spikes when a disruption floods the inbox.

Raman also explains why speed alone is not enough when several companies discover the same alternative supplier at once. Quoted availability, reserved capacity and a confirmed commitment are different things, and an agent that rechecks that difference before each booking keeps a team from planning on capacity that is already gone.

His readiness test is the part to take away: simulate a four-week delay with real data, time how long the team needs to reach a decision, and confirm the backup supplier has shipped a real order recently. A company that can answer "which orders depend on this input, and when do we run out?" within hours can start acting while others are still collecting information.

Our thanks to Sushanth Raman and the Pallet team for the written interview.

Sushanth Raman is the co-founder and CEO of Pallet, which builds AI agents for logistics and supply-chain operations. Before Pallet, he was an early engineer at the software company Retool. He started Pallet with his co-founder and fellow Retool engineer Andrew Spencer.

Sourcs: