What is the argument for pricing below cost during a land-grab phase, and the risk?

Sellthrough prices its retail forecasting tool below what it costs to run, on purpose, to win chains fast before a rival locks them in first. That is a real bet, not a mistake. The interviewer is not testing whether you would make it. They are testing whether you can say, out loud, who ends up holding the loss if the bet takes longer to pay off than planned, and whether that person ever had a way to see it coming.

How to answer this, stage by stage

Nobody is grading whether you know pricing below cost is common in a land grab. They are grading whether you can make the honest case for it, then name exactly who carries the risk if it does not clean itself up, in one breath, without hedging.

Let's learn

Here's what a price built to lose money on purpose usually forgets to say: when it's allowed to stop.

Sellthrough is the tool Anvergne Software sells to retail merchandising teams. Every night, it reads a store's real sales and tells the team how many of each item, each SKU (industry shorthand for one thing you stock), to reorder before shelves run empty or a stockroom fills up with things nobody's buying.

Anvergne priced Sellthrough at a flat $400 a store a month, unlimited SKU forecasting, no matter how big a store's catalog got or how often it asked for a new number. That price was set below what it actually cost to run, on purpose. At signing, forecasting a typical 3,200-SKU store once a week cost Anvergne about $520 a store a month. Against $400 of revenue, that's a $120 loss on every store, every month, approved with eyes open.

The argument for it was real, not just a sales line. Every store that signed fed real sales data into Sellthrough's shared model, how a heatwave moves cooler and grill sales, how markdown timing shifts by category, patterns a model only gets good at by seeing them happen, across many retailers, not from one account alone. Signing stores fast, even at a loss, meant that shared model improved faster than a fully-priced, slower-growing rival's could. And it meant one of Anvergne's earliest chain-scale customers, a 68-store apparel and outdoor gear chain, wasn't sitting on a competitor's shelf instead.

A formula alone will carry a below-cost price forever if nobody ever writes down where it's supposed to stop.

Eighteen months in, that chain was Anvergne's proof the whole product worked. Its merchandising director had added categories nobody had priced for at signing: footwear, camping hardware, a private label line. The store catalog had grown from 3,200 SKUs to 5,800. Sellthrough's shared model was also now committing to a real accuracy bar on the chain's top-selling SKUs, met by running each one through three models instead of one, a promise that the forecast clears a set error band most weeks, by design, not a promise every forecast is perfect. Combined with the extra work of predicting hundreds of brand-new SKUs with almost no sales history yet, cost to serve that one chain had climbed to about $740 a store a month. Same $400 price. The loss per store had gone from $120 to $340.

At its worst: the first anyone at Anvergne finds out is the quarter finance finally pulls this one account's numbers on their own, and by then the fix looks like a price increase sprung on the customer who staked her own department's reputation on the tool, thirty days before a routine renewal email. She either eats it with no warning, or her team spends a year unlearning a habit they'd fully handed to Sellthrough.

What I'd leave alone: the below-cost bet itself, for a brand-new account in its first few months. A store just getting started with 2,000 SKUs and a light catalog barely moves the shared model or the loss column. The strategy isn't the mistake. Leaving it open-ended is.

The lesson: a price that starts below cost on purpose is a real bet, not a mistake by itself. But a bet needs a written ending before the account signs, or you find out it never had one, on the one customer who grew the fastest.

Now here is the same thing as a story

The short version is above. Read this one when you want to feel why an account everyone pointed to as proof it worked came within one new hire's question of a very public argument.

Bethan Bruch had run merchandising for Fivepoint Outfitters, 68 stores of apparel and outdoor gear, for four years before Sellthrough existed. Every Sunday night she'd pull the week's sales and, category by category, work out what each store needed before Monday's truck. She was good at it. She could look at three weeks of footwear numbers and tell you which store was about to run out of size 10 boots.

Sellthrough arrived the way Anvergne sold it to her: one flat number, $400 a store, no matter what she threw at it. The good months were genuinely good. By the second quarter, her Sunday nights had turned into a fifteen-minute check-in instead of a six-hour build. Stockouts on her fastest movers dropped. She spoke about Sellthrough at Anvergne's own customer conference that spring.

It faded in three quiet beats, and none of them looked like a problem at the time. Beat one: she stopped double-checking the numbers against her own gut, because they kept being right, and second-guessing a tool that worked felt like wasted effort. Beat two: her team added footwear, then camping hardware, then a new private label line, onto Sellthrough without a second thought, because the bill never changed no matter what they added. Beat three: her whole Monday morning, the one she used to run herself, was now just reading whatever Sellthrough had already decided.

This was the whole Monday morning Sellthrough eventually replaced. Nobody replaces a habit like that for free.

None of these three beats looked risky by itself. Each one just made the next one easier.

The trigger wasn't a bad forecast. It was a question. A new hire on Fivepoint's own finance team, going through vendor contracts her first week, asked Bethan in passing why Sellthrough's bill hadn't moved in eighteen months when the team's own usage reports showed the catalog running through it had nearly doubled. Bethan didn't have an answer. She realized, standing there, that she had no idea what the number behind her own bill actually was, or whether she should be worried about it at all.

She raised it with her Anvergne account rep, carefully, the way you raise something you're not sure is even a real question. Nobody at Anvergne could tell her much either, not because they were hiding it, but because nobody had ever built a way to check one account's own number against what it actually cost to serve.

The decision that opened the door traced back to a short pricing meeting at Anvergne, more than two years earlier, when Priya Nandan, who owned Sellthrough's pricing, pushed for one flat, unlimited number to win Fivepoint over a slower-moving rival. No usage line. No written review date. It was the fastest number to sell, and at the SKU counts Fivepoint carried then, the math genuinely worked. Nobody in that meeting asked what the formula should do once a favorite customer started growing.

Three of these four were decided on purpose. Nobody in the room ever answered the fourth.

Run that meeting again with one change: the rate is still $400, still below cost, still the number that wins the account fast. But underneath it sits a usage floor, priced low, that grows with catalog size and how often a store gets forecast, and a written line that says the rate gets reviewed at month eighteen against real usage. Replay the same eighteen months: the catalog still grows to 5,800 SKUs. Cost to serve still climbs to $740 a store. But now the bill climbs too, to about $610 a store, because the floor is doing its job. The loss holds near $130 a store, close to what it was at signing, instead of climbing to $340. And the review conversation happens on schedule, sixty days before Bethan's own budget meeting, with real numbers in hand, not thirty days before a renewal notice that reads like a surprise.

One design hands the account a number that never moves no matter what they do to it. The other hands them a number that grows the way they do.

One design let Bethan find out from a new hire that the price she'd built a year of budgets around was never really fixed. The other lets her walk into her own budget meeting already knowing the number, because it moved a little every quarter instead of never moving until it had to move a lot.

What I'd tell myself, back in that first pricing meeting: a flat number wasn't really about keeping the deal simple. It was about assuming the account would stay roughly the size it signed at. Nobody ever promised us that, and Fivepoint was never going to be the account that proved us right by staying small.

GUARD: who actually carries a price built to lose money

Not a policy lecture. This is a live risk question with a real dollar loss in it, and GUARD is what stops "we approved the loss once" from quietly standing in for "we're still fine with it two years later."

One side of this deal can watch the gauge. The other side only ever sees the bill.

Three things worth stating directly, since the real judgment sits here. The alternative Anvergne's team actually considered, and rejected, was dropping the account's three-model ensemble back to a single model to cut cost immediately. It lost because that ensemble was the only reason Sellthrough could hit the accuracy bar Bethan's team was already planning inventory around, on their top sellers specifically, and breaking that promise would have cost more trust than the margin it saved. The AI-specific failure mode worth naming by name is cold start: hundreds of brand-new SKUs with almost no sales history, each one needing the model to borrow patterns from similar items instead, which is real compute work, not a free lookup. The guardrail is flagging cold-start SKUs as their own line on the cohort dashboard, so a catalog growing mostly through brand-new items shows up differently than one just adding proven ones. That guardrail isn't free either. Anvergne accepted a wider error band on any SKU in its first eight weeks of history, a real quality tradeoff, in exchange for not letting cold-start cost balloon the ensemble even further.

And if you want to be sure it really works, try it somewhere else

Same five letters, a veterinary triage tool instead of a forecasting one, and this time the old decision wasn't a missing usage floor. It was a promise stated out loud that never should have been.

Denwatch is Penhallow Health Systems' triage tool for independent veterinary clinics. It reads phone-in symptoms and photos and predicts how many urgent, routine, and complex cases a clinic should expect that day, so a two-doctor practice can staff correctly instead of guessing.

The build-up: Denwatch signed Hollow Pines Veterinary Clinic, owned by Yara Voss, at $150 a month flat, when a typical day's case mix cost about $95 a month to predict. As Hollow Pines became known locally for handling harder referral cases, Denwatch had to run far more multi-photo, longer-history calls per case. Cost climbed to $210 a month within a year. The price never moved, because Yara had been told, explicitly, this rate was hers for good.

The clinic that needed the guarded slice was never the biggest one. It was the one whose case mix got harder the fastest.

Same rank, different lever: here the fix isn't a usage floor on volume, it's tiering by case complexity, since one routine call and one multi-photo complex case cost very differently to predict, not by count alone. And Penhallow's real change was never phrasing a founding rate as permanent again. Every new clinic now gets a written eighteen-month review point in the contract itself, stated at signing, not left to a renewal email later.

Swap the trigger and it still runs.
Speed: an interviewer caps you at ninety seconds. Skip straight to it: below cost is fine during a land grab, but write down when you'll look at it again, and never call it permanent.
Cost: there's no budget this quarter for both a usage floor build and cohort-margin dashboards. Build the dashboard first. You can't fix a gap you can't see forming.
The model got better, for real: say Sellthrough's per-forecast cost actually fell 30 percent from a newer, cheaper model generation. The review point still matters, because a customer's own catalog growth can outpace even a real efficiency gain, the same trap running in reverse.

Where people run it wrong.
They let "we approved this loss once" stand in for "we're still fine with it two years later."
They watch one blended number and never split it by which cohort actually signed at the deepest discount.
They fix it with a sudden price jump instead of a floor that should have been written into the deal from the start.

How to use it live. Say the real question before naming a fix: "below cost is fine, is it below cost forever, or below cost until a date we already wrote down?" That buys a beat to actually answer, instead of defending a number nobody ever bounded.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0