Why Operational Data Is the Real Moat for Supply Chain AI

According to FreightWaves, freight-tech startup Nauta just closed a strategic round backed by BMW i Ventures, Bosch Ventures, Hitachi Ventures, and Yamaha Motor Ventures — and the most interesting thing about the announcement isn't the check, it's what the CEO said about why anyone should care. Valentina Jordan, Nauta's co-founder, is making a deliberately unfashionable argument in a hype-soaked market: the AI agents themselves are about to be commodities, and the actual moat lives underneath them, buried in messy, unglamorous operational data.

The "operational brain" and why the wiring matters

Jordan calls Nauta's platform an "operational brain" — and yes, the name makes me flinch a little, but the architecture underneath is worth taking seriously. The system plugs into the usual enterprise alphabet soup — ERP, TMS, WMS, OMS — but the part that actually matters is that it also ingests the stuff where operators do 80 to 90% of their work: email, Microsoft Teams, spreadsheets, contracts, arrival notices, payment terms. It layers in external signals too — weather, port disruptions, a recent earthquake affecting Colombian ports — and builds a continuous model of how each individual company actually runs.

On top of that data foundation, Nauta drops task-specific agents. The highest-value one, per Jordan, is three- and four-way invoice matching for accounts payable reconciliation — and given that the system reportedly flags somewhere between $300,000 and $500,000 in incorrectly invoiced amounts per client, that's not a rounding-error line item. It's also the seam where the conversation starts brushing up against adjacent infrastructure: a proper smart contracts development deployment checklist becomes directly relevant the moment you're automating payment terms at scale — because probabilistic agents flagging invoice errors is fine, but you eventually want the settlement logic itself to be verifiable, not just plausible.

Where AI actually pays back — and where it flatlines

Inbound Logistics, leaning on McKinsey's distribution-operations research, puts numbers on what's working: reductions of 20 to 30% in inventory, 5 to 20% in logistics costs, and 5 to 15% in procurement spend. One last-mile operator running more than 10,000 vehicles reportedly booked $30 million to $35 million in savings from AI-powered virtual dispatcher agents — a 15x return on a $2 million investment. The common thread across these wins isn't magic — it's structure. High-volume, rules-based, repeatable workflows: demand forecasting, warehouse slotting, freight matching, shipment visibility.

Where it still flatlines — and where you want a human firmly in the loop — is judgment-based, probabilistic work. Exception handling on damaged or misdirected freight. Complex customs brokerage. Supplier relationship negotiations. AI still chokes on ambiguity, novel scenarios, and frontline workers who simply don't trust tools that don't match how they actually make decisions.

Matt Huckeba, Chief Strategy Officer at Evans Transportation, makes the operational case from the trenches: order entry used to mean every team member manually keying 100 to 120 orders per day. Now integrations and AI agents pull shipment details from emails and PDFs and drop clean data straight into the TMS — staff touch one or two orders a day, and only when something genuinely needs a human call. Faster order-to-tender cycles, far fewer billing disputes. Not flashy, just compounding.

The blocker nobody wants to put on a slide

According to Proxima's Global Supply Chain Resilience Outlook report — a Proxima survey of more than 500 CEOs at businesses generating over $500 million in annual revenue across the US, UK, Australia, Singapore, and Germany — data quality (38%), lack of skills (30%), and clarity around ROI (29%) are the three named barriers to scaling AI further in the supply chain.

That's the unglamorous truth Jordan is openly betting on. Nauta already operates in more than eight countries with 8,000 suppliers and reach across 60 countries through its clients, spanning food and retail distribution, manufacturing, and consumer brands including New Balance, L'Oréal, and Moët Chandon. Her thesis — and it's a contrarian one in a market obsessed with model benchmarks — is that there's more similarity between a food distributor and a furniture business than between two food distributors, because every company operates differently. The data layer is the only thing that can capture that — and right now, it's still the bottleneck.

The agents are getting cheap. The data underneath them is not. Watch who actually owns it.