Sushant Ramen, photographed mid-sentence, explaining that the smartest robot in the room still can't read your mind — or your 900,000 unfiled documents.
▶ Watch on YouTubeThere is a man in the San Francisco Bay Area who once chose, on a glorious playoff Saturday, to sit at his desk and type. Not to type something interesting — not a novel, not a love letter — but to key in one hundred and fifty bills of lading, one at a time, by hand, while the 49ers played without him. "Susant, I'm doing this instead of watching the 49ers play the playoffs," he said. "I hate doing this." That man is not a cautionary tale. He is the entire logistics industry, condensed into a single, sad, fluorescent afternoon.
It was that scene — a real one, witnessed in person — that pulled Sushant Ramen out of a comfortable tech career and into the unglamorous, gloriously broken world of freight. And it's the scene that animates his appearance on TPM Today, the twice-monthly show hosted by Eric Johnson, senior technology editor at the Journal of Commerce. Johnson opened from S&P Global's New York headquarters with a confession that doubles as the thesis of the entire fall: he had spoken at eight events that season, and "virtually every conversation has been about AI."
So here is the question everyone is whispering and nobody is answering well: if AI is so transformative, why does it keep failing? Ramen has a number for that. A famous, slightly terrifying number.
The 95% Problem
"We have a fundamental belief that we understand why 95% of AI agents are not working," Ramen says, and you can almost hear the room lean in. The figure traces back to an MIT study — one that found roughly 95% of enterprise AI initiatives failed to do what they set out to do. It is not a logistics statistic. It is an everyone statistic. But it has detonated through supply chain circles with the force of gospel. Johnson noted it has come up "time and time again" in his travels — more people know this study, he says, "than they do about any other aspect of AI right now."
Most people hear "95% failure" and assume the machines aren't smart enough. Ramen's diagnosis is far more uncomfortable: the machines are plenty smart. The problem is that the people building them don't know what an ISF is.
It's a line delivered with a grin, but it lands like a thrown wrench. You cannot, Ramen argues, automate a freight forwarder's business if you have to be taught what a commercial invoice is, how a packing list ties to it, what an arrival notice means, or why an Incoterm matters. The domain experts, he says, simply aren't in the room during these onboardings. And when you pair clueless builders with the wrong use cases and zero change management, 19 out of 20 projects quietly die.
Tribal Knowledge: The Thing Nobody Wrote Down
Here is the heart of it — the idea that makes the whole conversation hum. Ramen calls it tribal knowledge, and once you see it, you cannot unsee it. Every decision in a logistics operation carries invisible context. The data on the document is the easy 10%. The other 90% lives in someone's head.
"Very rarely when you get these documents is the Incoterm explicitly specified," Ramen explains. The customer's email says "Procter and Gamble," but your system stores them as "PNG Incorporated." There is no field for that. Freight class? Not on the document. There are, he says, "thousands of tribal rules that exist inside every single operation" — and Johnson, the veteran, pushes back to say it's not thousands, it's limitless. A million unique circumstances. The relationship where you happily pay two dollars more. The supplier who's always a day late on one doc but flawless on everything else, so you forgive them.
Pallet's method is almost anthropological. They go on-site. They look at what was actually fed into the system versus what came out. They sit with operators and ask: did we get that right? Then they take those rules, chunk them into individual "memories," and teach the system when to reach for each one. Ramen's analogy is disarmingly human: onboard a new billing clerk and tell them nothing about which customers get leeway and which get pressured to pay faster — will they succeed? "The answer is probably no." AI, stripped of context, is just a very fast new hire who was never trained.
Organizational memory, made portable
This is where the idea turns genuinely useful for the lean shipper team — the one where a single human owns transportation and compliance and a dozen other things, and where, if they vanish for three months, the institutional knowledge vanishes too. Capture that person's tribal rules as organizational memory, and suddenly the contents of their head become transferable. "That's how you should think about it," Ramen confirms. The ghost in the machine, it turns out, is the knowledge your best person never bothered to document.
Why Go Broad When Everyone Else Goes Narrow
The logistics-AI landscape is a carnival of specialists: one company obsessed with customs entry generation, another with freight audit and billing, a TMS picking off five pain points it already knows. Pallet zags. It automates commercial invoices, packing lists, ISFs, bills of lading, booking forms, letters of credit, container track-and-trace, even air freight procurement. Why so sprawling?
Because the hard part, Ramen says, isn't any single agent — it's context scattering. Capture the organizational context once, and spinning up new agents becomes easy. And from the customer's chair, the logic is brutal arithmetic: would you rather manage eight vendors for eight processes, or one partner who looks at your whole operation? "That's like so much work to manage," he says of the multi-vendor maze. The one-stop shop wins not because it's flashier, but because it's less exhausting.
Shippers, Forwarders, and the Math of Survival
A viewer, Eduardo Ramirez, asked how agents actually help shippers. Ramen's answer reveals a clean split. Shippers run small teams on big transportation budgets, so Pallet helps them spend smarter — reaching out to a handful of freight forwarders, running a negotiation, and procuring air freight on the spot market at cheaper rates for one of Europe's largest manufacturers.
Freight forwarders and 3PLs are the opposite: headcount-heavy machines where the metric that matters is net revenue per full-time employee. The goal is to shrink G&A as a function of revenue. One customer — "one of our mutual friends," Ramen tells Johnson — is on track for "a couple million dollars of EBITDA improvement" as Pallet automates invoice processing, ISFs, BLs, booking forms, letters of credit, and track-and-trace across a business buried under roughly 900,000 unstructured documents.
The AWSification of Work
Johnson offered the metaphor of the episode, and it's a good one: the AWSification of logistics labor. Just as software companies scale cloud capacity up and down with demand, agents let operators flex their "workforce" for peak seasons and lulls. Ramen had a sharper, real-world version. A private-equity-owned freight forwarder is told to grow revenue by a fixed percentage next year — with zero new headcount. "The laws of physics do not apply," Ramen says of the impossibility. Without agents, there is no way to cover the extra shipments. With them, suddenly there is.
How to Spot a Fraud
If the market feels like the Wild West — and Johnson admits it does, with persuasive pitches everywhere — Ramen offers a buyer's survival kit that is refreshingly blunt. First, ask pointed domain questions. Do you have CargoWise experience? Can you point to the CargoWise XML structure? What's an ISF? An arrival notice? If the answers are hand-wavy or "it's on the roadmap," disqualify them. "They don't know what they're talking about."
Second, force a real test. Hand over your documents, your emails, make them sign an NDA, and make them do the actual task. The field of contenders collapses to one or two. Johnson added a piece of hard-won advice from shippers: bring a technical person to the meeting. AI has dragged non-technical buyers into deeply technical decisions, and someone who understands what's under the hood is the cheapest insurance you can buy.
The Punchline: You're Not Behind
For all the breathless urgency, Ramen ends on a note of calm. Are we at the transformative moment? "We're still very early. We're still in the phase of AI exploration. Very, very few companies have actually seen EBIT gains from implementing AI." Johnson seized on it as reassurance for the anxious: "You're not that far behind anybody else, and you may be ahead of people if you're considering this."
So the marching order isn't panic. It's curiosity. Stay in "constant research mode," Ramen says — you can't ignore it, but you don't have to sprint, either.
And then, because TPM Today ends every episode with the same question, Johnson asked Ramen to name a favorite musician. He picked two EDM artists: Martin Garrix, for rising from Europe to global fame young, and John Summit — the fired accountant whose girlfriend left him and who retreated to his parents' basement during the pandemic, where everyone wrote him off, and who emerged a star. "Pain creates the best art," Johnson offered. It's a fitting coda for an industry full of people keying in bills of lading on playoff Saturdays — proof that the most painful, manual, soul-crushing work is exactly the raw material the next great breakthrough is built from.
As for the panda ghost? It eats bamboo. You'll have to watch for the delivery.