Compare seat-based, usage-based and outcome-based pricing for an AI product.

Barrowmoor Legal Technologies built Clauseward so a contract could get the same careful read whether it was worth two thousand dollars or two million. Three ways existed to charge for that read: by the seat, by the document, by the result. For fourteen months Barrowmoor priced it by the document, and that is what let a twelve page vendor agreement slip through completely unread.

How to answer this, stage by stage

Nobody is testing whether you can rank three billing models on a slide. They are testing whether you can find the one moment a live price lands on somebody's judgment, and defend pulling it out.

Let's learn

Clauseward reads a contract someone uploads and writes a short note in the margin wherever a clause is worth a second look, the kind of note a senior associate leaves for a junior one before they've even asked for it.

Before Barrowmoor built it, a routine twelve page vendor agreement got about forty minutes of a real read, longer if the language was messy. Even then, on a busy week, the contracts that looked boilerplate got a much faster skim, sometimes none at all. A two million dollar acquisition got the careful read. A parking lot maintenance contract mostly didn't.

Clauseward changed that math. It reads that same twelve pager in under ninety seconds and flags anything unusual, whether the contract is worth two thousand dollars or two million. In its first year it caught something genuinely risky, an uncapped liability clause, an auto-renewal with no exit window, something like that, in about one contract in twelve.

Three real ways to charge for the same read. Each one bills a different person, at a different moment.

One pipeline. A usage price taps in at the second box, the exact moment a person decides whether the flag was worth paying for.

Here's the part that wasn't the real problem: Clauseward missing a clause now and then. It happens, every AI reviewer has a real miss rate, not zero. What actually hurt Barrowmoor was priced somewhere else entirely. Clauseward launched charging four dollars a contract, billed the moment someone clicked "check this." And once that price existed, attorneys started deciding, contract by contract, whether checking was worth it.

At its worst, that habit put the price tag on exactly the wrong contract. A twelve page vendor agreement that looked routine got skipped to save the four dollars, and it carried an auto-renewal clause with no exit window. Nobody found out until it renewed itself into a bad three-year term and cost real money to unwind.

Three worst cases, not two. Only one of them stays invisible until somebody sues.

Barrowmoor's own numbers show why a flat seat doesn't fit every buyer either. A light user checking twenty contracts a month barely dents a two-hundred-forty-nine-dollar seat. A firm running two hundred contracts through Clauseward during a due diligence sprint would badly undercharge Barrowmoor's own compute bill if that same seat stayed flat and unlimited forever.

That's the real shape of the problem: a flat seat is generous to the wrong user, a live per-document meter is dangerous to the wrong contract, and neither one, on its own, was ever going to be right.

The choice I would take back: the pricing launch review, fourteen months earlier, where charging per document felt like the fairest way to price a brand new AI tool. Nobody in that room decided to put a price on hesitation. It was just the natural shape of charging for what people use.

What I'd leave alone: Clauseward's separate renewal-tracking feature, which watches signed contracts for deadlines and is still priced per contract tracked. Nobody skips tracking a renewal date to save the fee, the way they skip an upfront risk check, because there's no moment of doubt sitting on top of a date.

The lesson: charging by use was never the mistake on its own. The mistake was landing that charge on the exact moment somebody had to decide if a risk check was worth it, so the cheapest-looking contracts stopped getting checked at all.

Now here is the same thing as a story

The short version sits above. Read below for why one offhand remark on a renewal call changed how Barrowmoor prices a contract it isn't sure is worth checking.

For six years before Clauseward existed, Loriane Fenimore built pricing models for legal software nobody used consistently, and she was good at it. Barrowmoor's board still credits an earlier seat-based model of hers with getting the company to break even eighteen months ahead of plan.

For Clauseward's first eight months, usage-based pricing looked like the right call. Three pilot firms loved it, paying four dollars a document felt fairer than a flat seat nobody could size correctly on day one, and everyone, Loriane included, called it the honest way to price a new kind of tool.

Then the habit thinned, in three beats nobody flagged at the time. For the first two months, associates ran everything through Clauseward, out of curiosity as much as anything. By month five, busier teams started batching contracts into one weekly review instead of checking each one the moment it landed. By month eight, at firms watching a tightening budget, skipping the "obviously fine" ones, the short, boilerplate-looking vendor contracts, had become the reflex, without anyone deciding it on purpose.

Three small beats, months apart. None of them looked like a crisis on their own.

The trigger was small. On a renewal call, Cassian Ilyenko, who ran customer success, mentioned almost as an aside: "one of our clients says their associates stopped running checks on anything under fifty thousand dollars in contract value, is that expected?" Loriane didn't have a real answer.

She pulled a quarter of usage telemetry that night. Twenty-three percent of contracts uploaded to Clauseward had never actually been run, not flagged, not reviewed, just sitting there. She had a fix live within two weeks: seat-based pricing, a bundled allowance of reviews, quiet metering on anything over it. She called it done.

It wasn't done. Two weeks after the new pricing shipped, a call came in from Quaystone Freight's in-house legal team. A twelve page vendor agreement, the kind nobody thought twice about, had auto-renewed into an uncapped, three-year liability term. It had never been run through Clauseward at all, back when the old pricing was still live, because it looked too routine to be worth four dollars. Unwinding it cost Quaystone close to $180,000 and four months of negotiating.

It was never really about the four dollars. Loriane had priced three real options, and for fourteen months the one she'd chosen billed the exact instant somebody had to decide whether doubt was worth paying for.

The decision Loriane would take back happened at that pricing launch review, fourteen months earlier. "We only charge for what people actually use, that's the fairest way to price an AI tool," she said, and with three pilot customers who loved it in the room, that was true enough that nobody pushed back. Nobody asked what happens once budgets tighten and "actually use" turns into a decision made contract by contract, four dollars at a time.

Run the audit again with the new pricing live, seat-based, forty reviews included, overage billed quietly at renewal. The following quarter, Loriane reran the same never-run check. The number that had sat at twenty-three percent for two straight quarters dropped to zero. Every contract uploaded got checked, boring-looking or not.

Same audit, rerun after the pricing changed. The tool didn't get smarter. It just stopped costing anything at the moment of doubt.

One design put a price tag on the moment of doubt. The other put a price tag on the month, and let the doubt go check itself.

What I'd tell myself in that first launch review: charging only for what people use isn't automatically fair. It's fair right up until the price lands on the one decision you actually needed somebody to make without thinking twice.

PICK, stretched to fit three options, not two

This isn't a two-sided argument dressed up with a third name. PICK is what keeps "charge for what people use, obviously" from becoming a decision nobody actually made.

One alternative Loriane's team considered and dropped early was a blended price, a small charge on all three axes at once, a bit per seat, a bit per document, a bit per outcome. It lost because splitting a small charge three ways just multiplies the number of moments somebody has to think about cost, instead of removing the one that mattered. The AI-specific risk worth naming by name is a quiet kind of drift: Clauseward's own eval set is built from real customer runs, so if a live price suppresses which contracts get checked, the eval set narrows right along with it, to whatever still gets run, and accuracy numbers can stay high while real-world coverage quietly shrinks underneath them. The guardrail is sampling a slice of the contracts nobody ran, checking them for free, and folding that back into the eval set every month, so the number on the dashboard can't outrun what the tool is actually seeing. And the trade being accepted, plainly: Barrowmoor gives up some revenue from its heaviest users, since a bundled allowance won't fully price what a due diligence sprint actually costs to run, in exchange for a price that never once asks an attorney to think twice before checking something.

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

Same four letters, a translated contract instead of a reviewed one, and this time the hidden cost isn't a missed clause, it's a wrong word nobody proofread.

Interline is a translation tool Milldrift Translation built for agencies that manage freelance linguists. A translator drops in a source document, a contract, a product manual, whatever it is, and Interline returns a full draft translation with flagged segments wherever it wasn't confident, an idiom, a legal term with no clean equivalent, that kind of thing. Benedek Redshaw runs vendor management for the freelance pool that piloted it. Interline's own cost per word works out to about six tenths of a cent, most of it spent on the flagged review pass, not the first draft.

Interline's pipeline has a box Clauseward never needed, a gate that only the delivery action has to pass through.

Same rank as before, a different lever: for Clauseward, the hidden cost was a contract nobody reviewed. For Interline, it's a document nobody proofread, and a wrong word sitting inside the wrong clause. The fix isn't a bundled seat allowance this time, translators are freelance, not fixed headcount. It's a gate: the AI draft and its flags can go out immediately, priced per word same as always, but the action of marking a job delivered without a human proofread is gated behind the document's own risk category. Legal, medical, and safety documents don't get to skip that pass, no matter what the per-word price tempted someone to save.

Swap the trigger and it still runs.
Speed: an interviewer caps you at ninety seconds. Skip straight to it, price the draft and the flags per word like everyone else in the industry does, but gate the one action that can't be undone, skipping human proofread, behind the document's own risk category, not behind what it costs to check.
Cost: budget exists this quarter for either a cheaper per-word rate or a mandatory proofread gate on high-risk documents, not both. The gate wins, because a per-word discount doesn't matter if the thing it saves someone from doing is the one check the document actually needed.
The model got better, for real: say Interline's flagged-segment accuracy improves enough that human proofread catches almost nothing new. Real progress, and it narrows how often the gate actually changes anything. It doesn't remove the gate, since "almost nothing" on a legal or medical document is still not nothing.

Where people run it wrong.
They let a price built for routine work quietly set the bar for whether a high-stakes document gets checked.
They assume a lower AI cost automatically means a safer shortcut, instead of asking what specific check the savings is skipping.
They price a new AI product to match the old human vendor's invoice out of habit, without asking whether that pricing shape was ever actually chosen on purpose.

How to use it live. Say the real question out loud before naming a side: "before I answer, is this a price on the work itself, or a price on the one moment somebody decides whether to double check it." That's what tells you whether a familiar per-unit price is safe to copy or not.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0