What killed Pandion wasn't the market. It was the model.

Scott Ruffin founded Amazon Air. That's not a throwaway credential. Building a cargo airline operation inside one of the most logistically complex companies on earth is a genuinely hard thing to do, and he did it.

So when Ruffin started Pandion in 2020, he had the founder story, the domain knowledge, and eventually the capital. $125M in equity over five years. The company was projecting $220M in 2024 revenue as recently as March of that year.

In January 2025, he sent a Friday afternoon memo to 63 employees. Company closing immediately. No severance. "We owe more than we have in the bank."

The default read on this: pandemic timing, capital crunch, tough logistics category. That's not wrong. It's just not the interesting part.

What's interesting is the structure of the bet Pandion was making, because that structure shows up constantly and doesn't end differently just because the category changes.

Attio is the AI CRM with conversational AI built directly into your workspace. Every morning, Ask Attio handles my prep:

All in seconds. No searching, no switching tabs, no manual updates.

Ready to scale faster?

Start now →

Pandion launched into a real gap. In 2020, pandemic e-commerce had pushed UPS and FedEx past capacity. Residential surcharges were climbing. Retailers were scrambling for alternatives. The middle mile, moving packages from fulfillment centers to last-mile carriers, was genuinely undersupplied.

Problem: that gap was cyclical, not structural.

E-commerce as a share of retail peaked in Q2 2020 and started reverting almost immediately. By 2022, the capacity crunch at incumbents had largely resolved. Amazon was building its own delivery network and subsidizing it through AWS margins. The competitive rationale for Pandion kept shrinking. And it's not like this was invisible — there were public data points on e-commerce normalization throughout 2022 and 2023. The model just kept requiring the original thesis to be true.

This happens all the time in venture and the pattern's pretty consistent. A temporary disruption gets read as a structural shift. Models get built on the elevated baseline. Infrastructure gets financed against projections that assume the new normal holds. By the time normalization is obvious, you've got two or three years of burn committed to a thesis that quietly stopped being true six quarters ago.

Peloton is the consumer version of this. Zoom's a milder version. Pandion is the logistics version. Category changes, structure of the mistake doesn't. And the reason it keeps happening is that early-stage data is genuinely ambiguous — a spike and a structural shift look identical for the first 18 months.

Part of why it's hard to catch in real time: your growth metrics are strong precisely because the macro is doing work your model is taking credit for. By the time normalization shows up in the numbers, you've already raised on the inflated baseline and committed capital to infrastructure that doesn't downsize easily.

The question worth asking early: does this business need the macro condition to persist, or does it work when things normalize? For Pandion, the whole unit economics model required volume to keep growing. Volume stopped growing like that. The loop never closed.

There's a version of this worth running on AI infrastructure right now. A lot of capital's being deployed against projections that assume GPU demand compounds the way e-commerce did in 2020. Some of those bets are right. The ones building owned physical infra on top of that thesis are making a structurally similar wager, with a similar timeline for finding out: probably two or three years, not immediately.

Every day, 4.5 million readers turn to 1440 for their factual news fix. We sift through 100+ sources to bring you a complete summary of politics, global events, business, and culture — all in a brief 5-minute email. No spin. No slant. Just clarity.

Join for free today!

The model: pick up from retailer fulfillment centers, route through five owned sortation hubs, push to a network of 1M+ gig last-mile drivers employed by staffing agencies. ML algorithms picked the optimal carrier and route. Pitch was 100% U.S. home coverage, no residential surcharges, one-to-five-day ground speeds.

The ML routing and the universal label system were both real. Neither's a moat.

Here's the unit economics problem. You need volume to hit sortation utilization, you need utilization to get contribution margin positive, you need positive contribution margin to justify the next raise. If volume stalls before you close that loop, you're carrying fixed costs you can't utilize while competing on price against incumbents who've had decades to optimize.

Five sortation centers aren't like five servers you spin down when traffic drops. They're leases, equipment, labor. The floor doesn't move. When the volume thesis broke, there was no fast way to adjust the cost structure. That's the thing people underestimate about owned physical infrastructure — optionality disappears when you sign the leases. Software businesses can pivot, reposition, cut headcount, change pricing in a quarter. A sortation network in five cities can't do any of that at speed.

Compare that to how the durable logistics tech companies are actually built. Samsara is fleet data and software on top of trucks other people own. Project44 is supply chain visibility, no physical assets. The companies that navigated 2022 to 2024 well tend to be coordination layers, not infrastructure owners. Same category label, completely different capital intensity, completely different risk profile when the market moves.

Convoy had $3.8B in GMV at peak and shut down in 2023. Pandion was moving 100K+ packages a day and shut down in 2025. GMV and throughput aren't the same thing as a business that works at the unit level. Revenue tells you about demand. Margin tells you whether that demand creates value. Pandion's revenue was real. The margin story never closed.

The March 2024 Series B looked like momentum. $41.5M raised. Revenue tracking toward $220M. Expansion plans.

Company shut down ten months later.

A $41.5M round burning through in ten months on a $200M+ revenue business is telling you something specific. Gross margins are thin enough that revenue scale doesn't fix the cash problem. You can be growing top line and still be structurally cash-consumptive if per-unit economics are negative.

In capital-intensive categories, late-stage rounds get read as health signals when they're actually time-extension signals. Investors still believe the model will work. The business hasn't proven it does. Those are different things, and the distinction matters a lot when the funding environment shifts. The round didn't change what was true about Pandion's unit economics. It just delayed when the truth became visible.

Ruffin's memo said the company "had runway through Q4 2024 based on our last funding rounds." That framing's the tell. A business generating real economic value describes burn relative to revenue. Describing your existence as contingent on the last round means capital is doing the job the unit economics should be doing.

Simple diagnostic worth running: take the last round size and divide by monthly gross profit. If the answer's more than 24 months out, the company isn't growing into its economics fast enough for the round to actually change the picture. It's buying time. Not always fatal, but it should be legible as exactly what it is — and priced accordingly.

Ruffin spent a month in daily conversations with potential acquirers including through the holidays. Close to a deal several times. Nothing closed.

That's not bad luck. The acquirer market for physical parcel infrastructure is structurally narrow, and it's knowable in advance. This is the version of due diligence that most cap tables skip because it feels defeatist to run it early. It isn't.

UPS and FedEx have no incentive to pay a control premium for assets in cities where they already operate. Amazon's building its own network and benefits from not strengthening competitors. Regional carriers aren't capitalized for acquisitions. PE looks at negative-margin logistics operations and passes quickly.

The tech assets, the ML routing and universal label system, are interesting separately. But technology stripped from an operating business doesn't trade at operating multiples. It goes for distressed prices in a wind-down, if it sells at all.

Worth asking early in any capital-intensive business: who buys this, and what's their actual economic incentive to pay a premium? Not optimistically — specifically. What does the acquirer's P&L look like post-deal? What synergies exist and are they worth more than the liabilities?

For Pandion, the honest answers: UPS and FedEx would be absorbing competing infrastructure with no clear synergy. Amazon would be acquiring a network that serves its competitors. Regional carriers don't have the capital. PE doesn't buy burning platforms. There's no scenario where any of them pay close to $125M for this.

If nobody has a clear incentive to overpay, that's not a question to save for the fundraising drought. It should inform how you build from day one. Either structure to be a natural bolt-on for an identified strategic buyer, build toward cash-flow positive on your own timeline, or accept you're dependent on continued VC cycles for liquidity. Pandion was building toward continued growth financing with no clear alternative. When that closed, there was nowhere to go.

The surface diagnosis: pandemic timing, capital crunch, tough logistics category. Accurate. Not useful the next time you're looking at a company in a similar structural spot.

More useful framing is conditional dependence: how many things have to simultaneously go right for the model to work?

For Pandion, the list: e-commerce demand had to stay elevated. Gross margins had to improve with volume before capital ran out. Funding environment had to stay open long enough to hit the efficiency threshold. And the acquirer market had to value the business near invested capital if the growth path didn't materialize. Each condition was plausible on its own. None of them held, and there's basically no scenario where the business works if two or three of them break at once.

This is different from a company that fails because of one obvious mistake. Pandion didn't make a single catastrophic call. Ruffin had the right credentials. The tech was real. The market insight was legitimate. The problem was the model needed too many non-overlapping things to go right, and when they didn't, there was no fallback. That's not a talent problem or a timing problem. It's a design problem, and it's fixable before you raise the next round.

Useful inversion: when you're evaluating any capital-intensive business, count the external conditions required for the model to work. One or two, probably defensible. Four or more, you're essentially betting on multiple simultaneous outcomes and should size the position accordingly. Doesn't mean don't build or don't invest — plenty of those bets pay off. But it should change what your milestones actually prove, how much capital you commit at each stage, and what the fallback looks like if two of those conditions break at the same time. Most people don't run that math until they need to.

The companies that have built durably through the last few years tend to share one thing: unit economics work at the volume they have today, not at some projected future volume. Sounds obvious. Rarely practiced. Most decks show you the model at scale. Very few show you the model right now, at current revenue and current margins, and ask whether that picture's good enough to survive without the next round. If the answer to that question requires a lot of assumptions, the model isn't proven yet — it's just funded.

Ruffin's memo ended with: "What each of you have built is unquestionably amazing and will leave its mark on the industry well beyond our years."

The routing tech and label system will probably show up somewhere else, absorbed into another company's product quietly, without credit to where it came from.

The 63 people in that memo got a last paycheck ten days later and no severance. The structure of the bet was wrong before the Friday afternoon it became obvious. Not the founder, not the market, not the timing in isolation. The model required things to go right that were outside the model's control, and there was no designed fallback if they didn't. That's a choice — one that gets made early, usually without ever being named as a choice at all. And it's the kind of choice that looks fine until the week it doesn't.

If you like what you read please do “add to address book” or reply (it helps with deliverability). If you don't you can unsubscribe below.