How would you price a free tier that is expensive to serve?

Cargent plans a truck's route: which stop next, in what order, reworked the second a stop changes or traffic breaks. Draymoor Systems gives one truck's worth of it away free, unlimited plans and replans, to pull small operators in before they ever see a Fleet invoice. Zolani Trentholme owns what that free tier actually costs to run, and she almost signed off on a campaign to promote it harder before pulling the real compute bill first. What she found: the free plan was quietly costing some accounts almost as much as a paying one.

How to answer this, stage by stage

Nobody is grading whether you can name a cost per API call. They're grading whether you know a flat free tier and two very different usage patterns are on a collision course from day one, and whether you can say, with real numbers, where that collision actually happens.

Let's learn

What does it actually cost to give away something a model has to think through, every single time somebody asks for it?

Cargent is Draymoor Systems' route planner for trucking and delivery fleets. Feed it a list of stops, and it hands back the order to drive them in, plus a live estimate of when each one gets hit. The moment a stop gets added, cancelled, or traffic breaks, the driver taps replan and Cargent works out the whole thing again.

Before Cargent, a small regional hauler planned an 18-stop morning route by hand, checking a paper manifest against known traffic patterns and gut feel. A dispatcher who knew the roads could do it in about 25 minutes. One who didn't could burn 40.

With Cargent, the same 18-stop plan comes back in under four seconds. Draymoor gives it away free for one truck, unlimited plans and replans, meant to pull small owner-operators in before they ever see a Fleet invoice. At launch, the free Solo tier had 400 accounts, all running it the way it was built for: one plan, once, each morning. The whole free tier cost Draymoor about 770 dollars a month to run. Nobody thought about it twice.

Three parts, every single run: read the stops in, write the new order out, check live traffic for each leg.

Read that number on its own and a free replan looks harmless either way. The turn: more replans was never the problem on its own, Cargent wants trucks calling it all day. What broke was who started calling it that often. Same-day couriers, running food and parcel drops with a new pickup landing every twenty minutes, found the free one-truck tier and started using it the way their day actually works: nine stops, replanned dozens of times, not eighteen stops planned once.

Same free tier, same one-truck rule. Two completely different amounts of Cargent underneath it.

A regional hauler on the free tier costs about two dollars a month, twenty-two mornings times nine cents. A courier running thirty replans a day, twenty-six days a month because gig work doesn't take weekends off, costs about thirty-five dollars a month on the exact same free plan. That's not a rounding difference. It's almost three quarters of what a whole paying Fleet truck brings in every month, for a free account bringing in zero.

At its worst: by month fourteen, Draymoor's free Solo tier had grown to 5,000 accounts. Ninety-three percent of them were still the cheap, design-case regional pattern. The other seven percent, 350 accounts running the courier pattern, were quietly responsible for 58 percent of the free tier's whole compute bill, about 12,285 of the roughly 21,285 dollars it cost every month. Nobody had ever split the number that way. The dashboard only ever showed one blended average, about four dollars and twenty-six cents an account, and that number looked completely fine.

A cap on trucks says nothing about a truck's day. A cap on replans does.

What I would leave alone: the flat, unlimited feel of the free plan for every account still under the real replan cap. Ninety-three percent of Solo accounts never come close to eight full replans a day. Metering only the accounts that do fixes the number without adding friction to the ones the free tier exists to attract.

The lesson: a blended average can hide a real cost the same way a blended accuracy number hides a failing segment. Free doesn't mean nobody's paying. It means the bill moved off the price tag and onto an invoice nobody was reading closely enough.

Now here is the same thing as a story

Read the short version above when you're in the room. Read this one when you want to feel why a healthy-looking average could still have been wrong the entire time, one truck at a time.

For fourteen months, Zolani Trentholme's monthly free-tier review took about ten minutes. Then one Thursday, it took the rest of her week.

Zolani had priced two other usage-based products before Draymoor hired her to run Cargent's economics, and she kept one habit from both jobs: pull the real infrastructure bill herself before trusting anyone's summary of it. For Cargent's first year, that habit was almost ceremonial. Free-tier spend was small, predictable, and boring, and boring was exactly what a good free tier was supposed to be.

The early Solo signups were almost all the same kind of account: a regional hauler with one truck, running one morning plan, checking it, driving it. Zolani's monthly number barely moved, month over month, a few hundred dollars for the whole free tier, easily smaller than what marketing spent on a single trade show booth.

It thinned in three quiet beats, and none of them looked like a mistake. Word got around among same-day and last-mile couriers that Cargent's free tier, built for regional haulers, would happily replan a single-van route as many times a day as anyone wanted, no charge, no rule against it. Those accounts signed up through the exact same form as everyone else, since nothing in Draymoor's system distinguished a truck doing one plan a day from a van doing forty. And because there were still far more regional haulers than couriers, the blended average per account barely moved, month after month, even as the mix underneath it quietly changed.

No single bad month. A number that held for over a year, until a marketing draft made Zolani look underneath it.

The trigger wasn't a cost alert. Draymoor's growth lead, Balthazar Yashiro, drafted a campaign built around the free tier's own biggest selling point: "Unlimited replanning, always free for your first truck." He sent Zolani the draft to sign off on the claim before it went out, purely as a formality. She almost approved it on sight. Then she decided to pull the real per-account numbers first, since "unlimited" was a word she was about to put in front of a lot of new signups.

What came back wasn't a bell curve. Sorted by replans a day, the free account list looked like a cliff: 4,650 accounts running one to three replans, then a short, steep step up to 350 accounts running eighteen to forty-six. One of them, logged the week before, had replanned 1,340 times in a single month.

The whole answer to this question in one picture. Zero on the sticker was never the same thing as zero on the invoice.

The decision that opened the door traced back to the tier's very first pricing meeting, eighteen months earlier. Someone asked whether it should be capped by usage instead of by trucks. The honest answer at the time was that there wasn't a usage pattern yet to cap against, the beta was sixty regional haulers doing exactly what was expected. One truck, unlimited plans, was one clean line in a pricing doc instead of a whole metering system nobody had a reason to build yet. It was the sensible call, for the product that existed then.

Run that meeting again with one change: an eight-replan-a-day soft cap, tracked from real percentile data instead of guessed at launch, with overflow downgraded to a cheaper partial replan instead of a full one. Same 350 courier-pattern accounts, same real demand. Instead of 35 dollars a month, the average heavy account now costs Draymoor about 12, most of it eight full replans and the rest the cheaper partial kind, still enough replanning to get a courier through a real shift. Free-tier spend drops by roughly a third, and not one regional hauler, the 93 percent who were never the problem, ever notices the cap exists.

One design let "unlimited" mean whatever an account's day happened to need. The other let the account's own real usage decide where free stopped being flat.

What I'd tell myself, back in that first pricing meeting: "unlimited" isn't a number, it's a bet that every account will use the product roughly the same way. The bet held for a year because the only accounts who'd found us yet made it hold. Nobody asked what happens the day a completely different kind of driver finds the same free button.

BOUND, priced against a free account instead of a paid one

Not a story wearing a framework's clothes. This is an estimation problem with two usage patterns hiding under one price tag, and BOUND is what turns "the model's cheap per call" into a number a free tier can actually survive.

Even the worst case sits well under a paying truck's price. Doing the same work by hand sits far above it.

Three things worth stating directly, since this is where the real judgment sits. The alternative Draymoor's team seriously discussed, billing a small per-replan fee on the free tier itself, lost because it protects the cost by adding checkout friction to the exact small operators a free tier exists to convert, and does nothing for the regional haulers already comfortably under the real cost line. The AI-specific failure worth naming is silent cost drift as trust grows: the more a courier account uses Cargent and finds it reliable, the more it replans, so the product succeeding at its own job is what quietly inflates the bill, the same shape as an accuracy number that looks fine until a segment is split out. The guardrail is tracking replans per account as a standing weekly metric, segmented by pattern, with an alert when the heavy segment's share of total spend crosses a set line, not a once-a-year blended number pulled together for a board slide. And the trade-off is real: overflow replans past the cap get downgraded to a partial re-optimization that only reconsiders stops not yet visited, at roughly 40 percent of a full replan's cost, which means a courier deep into a route with a pile of new pickups can't get quite as globally optimal a route as a full replan would find. Draymoor is trading a little route quality on the busiest accounts for a real cut in what the free tier costs to run.

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

Same five letters, a pet-health photo triage app instead of a truck router, and this time the lever isn't replans a day, it's photos per checkup.

Muzzlecheck is Alderfen Health's free consumer app: a pet owner uploads a photo of a rash, a limp, or a cut, and the app reads it, asks a couple of follow-up questions, and writes a plain-language triage note, see a vet today, this week, or keep an eye on it. It's free and unlimited, meant to build word of mouth before pitching a 19-dollar-a-month telehealth add-on called Alderfen Plus.

A casual pet owner's checkup, two photos and a short write-up, costs Alderfen about a cent and a half. But a handful of accounts started using the free consumer app very differently: a partner clinic's own front desk, triaging real walk-in intakes on the free tier instead of paying for Alderfen's clinic product, submitting six photos a checkup from every angle and running about forty checkups a day. That pattern costs about four cents a checkup, and at 1,040 checkups a month, one free account was running up about 42 dollars a month in AI cost, more than double what a paying Plus subscriber brings in.

The account driving Muzzlecheck's cost was never the pet owner checking often. It was the one submitting the most photos, the most often, at once.

Same rank, different lever: Grigori's team held the free tier's price at zero for every account and metered per photo past a real per-checkup threshold instead, keeping the same six-photo read for genuine clinic-style triage, because a triage note is only trustworthy if it actually looked at every angle of the injury.

Swap the trigger and it still runs.
Speed: an interviewer caps you at ninety seconds. Skip straight to it: cost each action by what it actually contains, find where real usage breaks a flat free rule, meter only past that line.
Cost: there's no engineering time this quarter to build per-photo metering. Ship the cheap version first, a monthly report flagging any free account whose photo count per checkup sits far above the pilot's own baseline.
The model got better, for real: say the vision model gets twice as accurate for the same price. That doesn't change the verdict. A cheaper, better model moves the line, it doesn't remove the fact that a flat free rule needs to know the shape of usage underneath it before it can be trusted.

Where people run it wrong.
They watch one blended cost-per-account number and never once split it by how an account actually uses the product.
They fix it by adding a fee for everyone, which punishes the light accounts who never caused the problem.
They ban or hard-cap the heavy pattern outright instead of pricing it, and lose a real, valuable segment to whoever prices it correctly first.

How to use it live. Ask the shape question before quoting a number: "does every account on this free tier use it about the same amount, or does a small slice of them drive most of the cost?" That decides whether a flat free rule is even the right shape, before any number gets named.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0