The 150 Bills of Lading and the 49ers Game He Missed

A drayage operator in the San Francisco Bay Area invited a curious stranger to his office on a weekend. On the desk sat 150 bills of lading. He keyed them in, one by one by one — while the 49ers played in the playoffs.

"Susant, I'm doing this instead of watching the 49ers play the playoffs," the man said. "I hate doing this." That sentence — equal parts confession and indictment of an entire industry — is the closest thing Pallet has to an origin story. The curious stranger was Sushanth Raman, who had quit his job to find out why the most sophisticated logistics companies on earth still ran on manual labor and quiet resentment. He found his answer, founded a company, and this week sat across from the Journal of Commerce's Eric Johnson on TPM Today to argue something deceptively simple: the robots aren't the problem. The forgetting is.

Raman is not the usual freight-tech archetype. A Columbia computer science graduate, he was a product manager on Google's AI team before joining the low-code startup Retool, where he worked shoulder-to-shoulder with the likes of Airbnb, Flexport and C.H. Robinson. His grandparents were in shipping. So when he kept seeing commercial invoices, packing lists, ISFs and arrival notices processed by hand — container tracking done by hand — at companies that owned the best software money could buy, the question wouldn't leave him alone. "This problem was ubiquitous across logistics," he said. So he went looking for it, in person, in offices, on weekends.

Why 95% of AI agents quietly fail

If you've attended a logistics conference in the last year, you've heard the number. An MIT study found that roughly 95% of enterprise AI initiatives failed to do what they set out to do. Johnson noted it has surfaced "time and time again" in his travels — more people seem to know it than any other AI statistic. Raman has a thesis for why, and it isn't about model quality or compute.

Tribal rules. The phrase did a lot of work in this conversation. Consider: the Incoterm is rarely written on the document. The customer your email calls "Procter and Gamble" lives in your system as "PNG Incorporated" — and no agent knows they're the same entity. Freight class goes unstated. There are, by Raman's count, "thousands of tribal rules" inside every operation, most of them undocumented, many of them lodged in a single person's head. An AI agent without that context is exactly like a new hire who was handed a billing job and told nothing about which customers get leeway and which get chased. "Would they be successful at their job?" Raman asked. "The answer is probably no."

So Pallet's method is less science fiction than ethnography. The team goes on-site, compares what was on the input document with what an operator actually typed into the system, and infers the rules from the gap. Then they sit with the operators and check: did we get it right? The validated rules get chunked into individual "memories" the system knows when to retrieve. The result, Raman says, is a co-pilot running at 97–98% accuracy — and, crucially, customers who feel understood. One well-known Canadian 3PL had tried four other AI vendors. Every implementation failed. Pallet worked. "You guys are the only ones that captured our tribal knowledge," they told him. A Memphis freight forwarder, after auditing every option, made Pallet its exclusive automation partner because the others plainly didn't understand Cargo Wise.

Organizational memory, not magic widgets

The villain of Raman's story is "context scattering" — knowledge stranded in inboxes, spreadsheets and skulls. The hero is organizational memory: a specific, retrievable rule attached to a slice of the business. Work with Apple, and a memory might encode that every shipment defaults to full container load, with Ex Works as the default Incoterm. Capture enough of those and something useful happens to the lone overworked specialist who touches transportation, compliance and a dozen other things at once. If that person vanishes for three months, the institution usually loses everything in their head. With a digital workforce trained on their memories, it doesn't. The knowledge stays.

That's also Raman's defense against an obvious question from Johnson: why go broad? Some startups own one narrow process — customs entry, freight audit. Pallet spans document processing, container track-and-trace, ISFs, booking forms, letters of credit, even spot air-freight procurement. The answer is that once you've captured the organizational context, spinning up new agents is cheap. And customers would rather have one partner who automates the whole operation than juggle eight vendors, eight onboardings, eight contracts. The proof point Raman keeps returning to is C.H. Robinson — which stopped dancing around individual use cases, applied AI everywhere, and watched its financial performance turn around. "An example of what happens when you can do this automation successfully across your whole business," he said. Johnson predicted it'll become the Harvard Business Review case study for the category.

Shippers vs. 3PLs: two different machines

Asked by a viewer how shippers benefit, Raman drew a clean line. Shippers run lean teams with large transportation budgets, so Pallet attacks the spend: for one of Europe's largest manufacturers, it solicits air-freight quotes from a handful of forwarders, runs a negotiation inside a time window, and surfaces the cheapest options. Freight forwarders and 3PLs are the opposite animal — headcount-heavy — so the target metric is net revenue per full-time employee, and the goal is to shrink G&A as a share of revenue. For one forwarder, Pallet dissected the business, found roughly 900,000 unstructured documents sloshing around, and projects a couple million dollars of EBITDA improvement by automating invoices, packing lists, ISFs, bills of lading, booking forms, letters of credit and track-and-trace.

The AWS-ification of work

Johnson floated a framing Raman embraced: think of digital workers like AWS compute. Busy season, disruption, a surprise volume spike — scale up. Quiet stretch — scale down. No one's paying a quote-unquote worker for a 40-hour week during a lull. The sharper version came from a private-equity-owned forwarder told to grow revenue by a set percentage next year while adding zero headcount. "That is physically impossible," Raman recalled them saying. "The laws of physics do not apply." Agentic AI was the only door out of that room. Either you're augmenting capacity so the same people do more, or you're covering shipments no human roster could. Both beat the alternative, which is a mandate that can't be met.

How to shop for an AI vendor without getting fooled

The freight-tech market, Johnson admitted, looks "a little bit like the Wild West" — many players, all with persuasive pitches. Raman's buyer's guide is refreshingly blunt. Ask pointed domain questions: Do you have Cargo Wise experience? Can you describe the Cargo Wise XML structure? What's an ISF? An arrival notice? A bill of lading? If the answers are hand-wavy or live on a roadmap, disqualify the vendor — "they don't know what they're talking about." Then force a test: hand over real documents and emails under NDA and make them do the actual task. Demos dazzle; tests reveal. The field narrows to one or two fast.

Johnson added a field-tested tip of his own: bring a technical person to the meeting. AI has shoved non-technical buyers into technical decisions they never had to make when buying ordinary SaaS — where you didn't need to know how it was built. Now you do, at least enough to tell whether the pitch makes sense under the hood.

Still early — which is good news

For all the breathless coverage, Raman's verdict on adoption is sobering and, oddly, reassuring. "We're still very early," he said. "We're still in the phase of AI exploration. Very, very, very, very few companies have actually seen EBIT gains from implementing AI." Johnson seized on it as the antidote to FOMO: if you feel hopelessly behind, you aren't. Most people are barely past the starting line. "You're probably ahead of most people," he said, "if you're considering this." The only unacceptable posture is to ignore it.

The episode closed where good ones do — off-script. Raman's favorite musicians: Martin Garrix, for his early-age devotion to a craft, and John Summit, a fired accountant who lost his girlfriend, retreated to his parents' basement during the pandemic, and emerged one of electronic music's biggest names. "Pain creates the best art," Johnson offered. It's a fitting coda for an industry whose breakthroughs tend to begin, like Pallet itself, with someone hunched over a desk, missing the game, hating the work — and deciding it didn't have to be that way.