What should you do when a single enterprise customer becomes unprofitable?

Chancery priced one customer like every other customer. The real choice was between having an uncomfortable pricing conversation out loud, or quietly making the product worse and hoping nobody on a live deal ever needed the depth it used to have.

How to answer this, stage by stage

Nobody's testing whether you'd notice a customer is losing money. They're testing whether you'll say the uncomfortable thing out loud, or quietly change the product instead and call that a fix.

Let's learn

What do you actually owe a customer who's losing you money: a quieter product, or an honest phone call?

Chancery is an AI tool built for enterprise legal teams. A lawyer uploads a vendor contract, and Chancery reads it against the company's own playbook, flags the risky clauses, and proposes the redline language before anyone signs off.

Before Chancery, a senior associate at a law firm or in-house legal team read every vendor contract by hand. A standard contract, an NDA or a simple services agreement, took about 90 minutes to review properly. A genuine deep due-diligence contract, the kind with cross-referenced schedules and buried liability clauses that only shows up during an acquisition, took a full working day, about 8 hours. At the volume of one modest bolt-on acquisition, 40 contracts, that was roughly 220 hours of associate time.

Denhurst Capital, a private equity firm, signed Chancery's enterprise tier two years ago: $15,000 a month, unlimited reviews. At signing, Denhurst ran about 40 contracts a month, almost all standard vendor paper. Chancery's own cost to serve them was about $800 a month in model calls. Nobody watched that line. There was no reason to.

The turn: then Denhurst went on an acquisition spree, fourteen bolt-on deals in five months. Volume climbed to 260 contracts a month, and a third of those now needed the deep due-diligence pass. Chancery's own cost to serve the account climbed from $800 a month to about $38,200 a month, while Denhurst's bill stayed exactly where it was signed: $15,000 a month, flat. The extra spend was never really the problem. The problem was that this account would keep losing more money every single month, because nothing about the flat price had anything to do with how Denhurst was actually using the product now.

The awkward call is small and over fast. The quiet cap is the one that grows in the dark, on someone else's live deal.

At its worst, quietly capping what Chancery does to control the bill doesn't fail loud. It fails on exactly the contract where the depth mattered most, the one buried liability clause a deep pass exists to catch, on a deal that's already closing. That's worse than if Denhurst had never used an AI tool at all, because a lawyer who trusts a capped review stops double-checking it themselves.

The choice I would take back: Thea's first move, before any renegotiation, was to quietly cap the deep due-diligence pass at two model passes per contract instead of the adaptive number a genuinely complex contract needed, sometimes five, to bring the compute bill down without an uncomfortable pricing call with a reference customer.

What I'd leave alone: Chancery's smaller enterprise accounts, the ones with steady, modest volume and no acquisition spikes, stay on the same flat unlimited pricing. The model only breaks when usage genuinely changes shape, not just when it grows a little.

The lesson: a flat price is a bet about how a customer will use your product, not a promise about what serving them will always cost. The bet holds right up until the customer's own business changes shape, and you don't find that out from the invoice you send them. You find it from the cost line you stopped watching.

Now here is the same thing as a story

Read the long version below when you want to feel why the quiet cap went wrong, not just be told that it did.

Thea Coyne could read Chancery's enterprise account list and tell you which customer was about to become a problem before finance finished building the quarterly report. She'd owned Denhurst's account since the contract was signed, back when "unlimited reviews" was a line nobody thought twice about.

The first eighteen months were quiet. Denhurst's legal team opened Chancery most weekday mornings around nine, ran a handful of standard vendor contracts through it before lunch, and the account barely showed up in any review of Chancery's margins. Then Denhurst's fund closed a new round and went shopping. Fourteen bolt-on acquisitions in five months, and Denhurst's legal team started running every target company's entire vendor contract book through Chancery's deep due-diligence mode, sometimes forty deep reviews in a single week.

Thea saw the burn climbing in her weekly cost dashboard: $800, then $6,000, then $19,000, then $38,200 a month, against a $15,000 flat contract. She didn't take it to Denhurst right away. It was the last week of the quarter, finance wanted trimmed spend across every account, and a pricing renegotiation with Chancery's single biggest reference logo felt like exactly the kind of conversation nobody wanted to be the one to start. So instead, she asked engineering to cap the deep due-diligence pass at two model passes per contract, down from the adaptive count the review normally used, quietly, in a config change nobody outside the team would see.

It came back on a Tuesday afternoon, six weeks later, not as an outage but as a phone call. A Denhurst paralegal, doing a final manual read before a deal closed the next morning out of old habit rather than any doubt in the tool, found a change-of-control clause buried in a cross-referenced schedule that Chancery's redline had missed entirely. The two-pass cap had stopped short of the cross-reference check that a genuine deep review does on pass four and five. It was caught, barely, the night before signing, not after.

Thea pulled Chancery's own eval set and ran it against the capped version, specifically on change-of-control and indemnification clauses, the rare, high-stakes kind. Accuracy on that clause family had fallen from about 93 percent to about 68 percent under the two-pass cap. The overall average across all clause types had barely moved, because those clause types are rare in the mix, which is exactly why nobody watching the average had caught it.

The decision Thea would take back isn't capping the review to protect margin, that's a real tradeoff worth considering some quarters. It's deciding to cap it quietly, without telling Denhurst, instead of picking up the phone and having the pricing conversation she was avoiding.

Run that quarter again with one change: the cap never ships without Denhurst knowing. Instead, in month two, once the burn crosses $18,000 for two months running, Thea brings Denhurst the real numbers and proposes a usage-based add-on: $250 for every deep review past the first 40 in a month. Denhurst's legal team agrees inside a week, mildly annoyed, genuinely fine. The two-pass cap never gets built. The change-of-control clause gets its full five-pass review like every other one. The near miss never happens, and the account is profitable again by month three instead of quietly losing $23,000 more every month it went unaddressed.

One design trusted a quiet config change to make an uncomfortable number go away. The other trusted an honest conversation to make the same number go away, in the open, with the customer still getting what they were told they were getting.

What I'd tell myself the week I shipped that cap: an awkward phone call you're avoiding doesn't stop being the answer just because a config change is faster to ship than a conversation.

PICK, four moves for an unprofitable enterprise account

This isn't a churn question wearing a pricing question's clothes. It's a real pick, and PICK is what keeps "renegotiate the contract" from staying a hunch instead of a decision you can defend under follow-up.

Three things worth stating directly, since this is where the real judgment sits. The alternative Thea's team seriously considered was walking away from the account immediately once the burn crossed $30,000 a month, rather than attempting a renegotiation first. It lost because Denhurst had never done anything wrong, they were using exactly the product they were sold, at exactly the depth they'd been promised, and Chancery's own pricing model was the thing that hadn't kept up. Churning a customer for using your product correctly is a sales problem dressed up as a customer problem. The AI-specific failure worth naming by name is silent quality degradation from a cost-saving change: capping the number of model passes on a deep review didn't throw an error or lower a visible score, it just stopped catching a specific, rare, expensive kind of clause. The guardrail is Chancery's own held-out eval set, run on the exact clause families the deep pass exists to protect, indemnification, change-of-control, liability caps, with a floor set per clause family rather than on the overall average, since the overall average is exactly what stayed steady while the real problem grew. And the tradeoff being accepted openly: renegotiating protects trust and margin but costs an uncomfortable conversation and a real chance the customer pushes back or leaves; capping quietly protects the relationship on the surface but trades away exactly the accuracy the customer is paying for, on exactly the contracts where it matters most.

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

Same four letters, a veterinary hospital chain instead of a private equity firm, and this time the cost that got out of hand wasn't a deeper review mode, it was sheer image volume.

Greavesend is a diagnostic imaging tool built for veterinary radiology. A vet uploads an X-ray or an ultrasound scan, and Greavesend flags likely fractures, masses, and soft-tissue abnormalities before the vet writes up the read. Hosea Callas is the finance partner who signed off on what it costs to run.

The build-up: Hollins Veterinary Group, a regional hospital chain, signed Greavesend's enterprise tier at a flat $22,000 a month for up to 3,000 scans, a limit nobody expected them to hit. Hollins acquired six more clinics inside a year and scan volume climbed to 11,000 a month. Greavesend's real cost to serve them climbed from about $4,100 a month to $31,000, while the contract stayed exactly where it was signed.

Same rank as before, different lever: for Chancery, the account went unprofitable because the mix of work shifted toward a genuinely more expensive task. For Hollins, nothing about the task changed at all, sheer volume just outgrew a hard cap nobody had priced past. When Greavesend's engineers throttled speed instead of raising the question of price, the customer felt it exactly once, at exactly the worst possible moment, with no warning that it was even a lever being pulled.

Swap the trigger and it still runs.
Speed: an interviewer caps you at ninety seconds. Skip straight to it: bring the customer the real numbers and renegotiate before you quietly change what the product does for them, and know your walk-away number before the call, not during it.
Cost: there's no budget this quarter for both a full renegotiation process and a faster engineering fix. The renegotiation wins for Chancery specifically, because a quietly degraded legal redline is a liability problem, not just a slower support ticket.
The model got better, for real: say the underlying model gets twice as cheap to run next year. That's not proof the account stops needing attention, since a customer's usage pattern can keep growing faster than any one model's price drops, and the habit of watching cost against a flat price still has to exist either way.

Where people run it wrong.
They treat "the customer is losing us money" as a reason to quietly change the product, instead of a reason to have an honest conversation with the person paying for it.
They set the walk-away number during the negotiation, under pressure, instead of before it, so it always gets talked back down by whoever wants to keep the logo.
They watch the overall accuracy average after a cost-saving change instead of the one rare, high-stakes case family the change actually touched.

How to use it live. Say the real question out loud before naming a number: "before I answer, is the account unprofitable because the customer's business genuinely changed shape, or because we priced it wrong from day one." That buys a beat to think, and it changes whether renegotiation or a pricing fix is really the first move.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0