How would you handle a customer whose usage makes them unprofitable?

Pactline Systems sells ScreenArc to staffing agencies: upload a stack of resumes, get back a ranked shortlist for the job you're trying to fill. Every account also gets a second option, a wide search across the agency's entire resume archive, meant for the handful of roles that are genuinely hard to fill. Nobody is grading whether you can say the word "margin." They are grading whether you can find the number that would have caught this while it was still a five-minute fix, weeks before a customer's invoice ever said the word "unprofitable" out loud.

How to answer this, stage by stage

Nobody is grading whether you can name margin as the metric that matters. They are grading whether you can build the whole chain live: the real number, the signal that beats it, how someone could dodge either one, and what you'd actually do, in that order, out loud.

Let's learn

Here's a mean trick a metric can play: it can hold dead still for weeks while the thing underneath it quietly goes broke.

ScreenArc is the tool Pactline Systems sells to staffing agencies. A recruiter uploads a stack of resumes, and ScreenArc reads them and ranks the field against whatever job they're trying to fill.

Before ScreenArc, a Nash Talent Group recruiter spent close to six hours screening 150 resumes against one job opening, reading each one, checking it by hand. With ScreenArc, the same screen takes about twelve minutes: upload the stack, and a ranked shortlist comes back with a plain reason attached to every name.

That's the ordinary story on almost every one of Nash Talent Group's 40 seats, every month. What changed wasn't the time it saved. It was a second button next to the normal one.

ScreenArc's normal rerun checks a job's own shortlist, the 60 to 90 resumes a recruiter actually sourced for that role. It costs Pactline about $3.20 a run. Next to it sits a second option: search the entire archive, all 40,000 resumes Nash Talent Group has collected over three years, instead of just the shortlist. It costs about $38 a run, twelve times as much, and it was built for the rare, genuinely hard-to-fill role.

Same size button, same one click. Twelve times the cost, and nothing on the screen ever said so.

At its worst: over ten weeks, the wide-net button went from a rare tool to the default first move on most of Nash Talent Group's reqs. The account that used to run at about 72 percent margin ended the month about 23 percent underwater, roughly $3,200 lost that month alone, on a customer paying $14,000 for its seats.

What I'd leave alone: the flat monthly seat subscription, and the normal scoped reruns on 39 of Nash Talent Group's 40 seats, and on nearly every other Pactline account. None of that ever changed. No review needed there.

The lesson: a metric that only watches what a customer bought, seats, job postings, will hold steady right up until the invoice tells you the truth. Watch what a customer actually does with what they bought instead.

Now here is the same thing as a story

Read the longer version below when you want to feel why ten quiet weeks with nothing turning red on a dashboard still ended in a real loss.

Liora Winslet has run recruiting operations for staffing agencies for eleven years, three different firms, and she can tell in the first paragraph of a job order whether it's going to fill itself or fight her the whole way. Nash Talent Group hired her two years ago mostly for that instinct.

ScreenArc arrived in her first spring there, and for months it did exactly what it was sold to do. A recruiter would source forty to ninety resumes for a role, upload them, and get back a ranked shortlist with a plain reason attached to every name: strong match, five years of similar scheduling work, or weak match, no direct customer-facing experience. Twelve minutes instead of most of a morning. Liora liked it enough to put it in the onboarding deck for new hires.

Every number Pactline actually watched held completely flat, the whole ten weeks.

There was a second button on that same screen, smaller, labeled search entire archive. It ran the same kind of check, but against every resume Nash Talent Group had ever collected, three years and forty thousand of them, instead of just the shortlist a recruiter had actually sourced. Liora's team used it maybe twice a quarter, on the roles that felt hopeless: a bilingual quality inspector for a plant two hours outside anywhere, a warehouse supervisor with a very specific forklift certification nobody local seemed to have. Once, it found someone. A dispatcher role that had sat open for six weeks filled inside two days, because the archive turned up a candidate who'd applied to a completely different Nash Talent Group posting a year earlier and never heard back.

That single win is where the habit really started, and it thinned out in three beats, none of which felt like a decision at the time. Beat one: Liora told her team, informally, in a Tuesday standup, that if a role felt stuck, it was fine to just hit search everything, worst case it doesn't help. Beat two: a couple of recruiters stopped waiting for stuck. They'd run the archive search on a fresh job order right alongside the normal one, just to see. Beat three: by late summer it wasn't a fallback anymore. It was reflex, the second click after uploading a new stack, on almost every requisition, not just the ones actually fighting them.

Nobody at Nash Talent Group ever saw a bill broken out by run type. It wasn't hidden on purpose. It was just never shown.

There was no single Tuesday where it broke. That's the part that still gets me. No catastrophe, no angry email, no obvious mistake anyone could point at. Just ten quiet weeks where the button people reached for first kept changing, while every number Pactline Systems actually watched, forty seats, about a hundred and twenty open jobs a month, held completely flat the entire time.

Nobody decided, on any one of these weeks, to let it keep climbing. It just never had a line drawn under it.

I found it the boring way, on a routine monthly margin review, the kind I run across forty accounts without expecting to find anything. One line, Nash Talent Group, was negative. I checked it twice assuming a billing error. It wasn't a billing error. It was a real customer, paying us fourteen thousand dollars a month, that we were actually losing money to serve.

Pulling the usage log back ten weeks told the real story. Archive-wide search had gone from five percent of the account's reruns to sixty-one percent, climbing every single week, while seats and job count never so much as flickered. By the time the invoice caught it, the account had already been bleeding for over a month.

The decision I would take back happened much earlier than any of that, in a short product meeting almost a year before Liora ever joined. Someone asked whether the archive search needed its own visual treatment, a cost badge, a confirmation step, anything that marked it as different from the normal rerun. The answer was no, because almost nobody used it, and marking it up felt like clutter on a clean screen. That was true when it was true.

Run that meeting again with one change: the archive button gets its own weight from day one, a small line underneath it, about 40,000 resumes, costs more, best for roles open three weeks or longer. Recruiters still get to use it exactly as often as the job genuinely needs it. They just can't reach for it without seeing, every time, that it isn't the same button.

Replayed with the real fix in place: the nudge lands the moment wide-net share crosses fifteen percent, in-app and from Nash Talent Group's account rep, the week it happens, not the month after. The hard cap, twenty archive runs included, a real cost confirmation past that, catches anything the nudge doesn't. Four weeks later, wide-net share is back down near eight percent. The account's margin, which bottomed out at twenty-three percent negative, is back above sixty-seven percent, and it never has the chance to fall further than that first month's loss.

One design let a habit run for ten weeks with nothing to slow it down. The other one puts a small, honest cost label right in the way of the exact click that caused all of it.

What I'd tell myself, back in that meeting nobody remembers: making the expensive option look identical to the cheap one was never really about a clean screen. It was a bet that nobody would ever lean on it out of habit. Nobody ever checked whether that bet was still true a year later.

LEAD, and the two numbers that stayed perfectly still

Not a checklist to recite. This is what stops "usage went up" from quietly standing in for "we know why this account is bleeding money," which are not the same sentence.

A per-seat number can be diluted by adding seats. A per-requisition number can't, because a requisition still costs what it costs to serve.

Three things worth stating directly, since the real judgment sits here. The alternative Pactline actually considered, and rejected, was folding the growing archive-search cost into next year's list price for every account, spreading Nash Talent Group's own pattern across customers whose usage was completely normal. It lost because it would have made every healthy account subsidize one customer's habit, and it wouldn't have fixed the actual cause, a recruiter clicking an expensive button with no idea it was expensive.

The extra spend doesn't even reliably buy a better shortlist if the pool it's searching has gone stale.

The AI-specific failure worth naming by name is silent corpus decay: the archive itself gets noisier every year, full of duplicates and candidates long since placed elsewhere, so the expensive search doesn't even reliably return better matches, it just returns more of them from a pool that's quietly gone stale. The guardrail is a routine dedup and freshness pass on the archive, with a plain freshness note next to the search button, so the extra spend at least buys what it claims to. None of this is free: the cap and the cost confirmation add a real click of friction to a workflow that used to be one click, and Pactline is trading a little recruiter convenience for keeping the account's math sane and the real cost of that click visible instead of hidden.

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

Same four letters, a crop-disease scanner instead of a resume screen, and this time the lever isn't how wide a search runs. It's how often the expensive check runs when nothing's actually wrong.

LeafGauge is Kepler Field Data's tool for farm co-ops: a phone photo of a few leaves runs through a light model and flags anything that looks like early blight or rust, for about six cents a scan. A second mode, a full multispectral drone pass processed by a heavier model, catches disease before it's visible to the eye, at about fourteen dollars a field per pass. It's meant for fields with a recent flag or a known disease history, maybe one field in ten in a normal week.

Odalie Cadmus runs field operations for Trentham Growers Co-op. A bad blight season two years before LeafGauge arrived cost the co-op a third of one variety's yield, and she still thinks about it. Once LeafGauge gave her a way to check more often, she started running the deep multispectral pass on every field, every week, flagged or not, just to be safe. Nobody at Kepler's end noticed, because the number they watched, total acres under contract, never moved. Deep-pass share of all field-weeks climbed from about 10 percent to 70 percent over one growing season, and Trentham Growers' account margin followed almost the same shape as Nash Talent Group's.

The field that earned the expensive pass and the field that got it were often two different fields.

Same rank, different lever: the segment that needed the guarded slice is still the one to find first, but the axis isn't how far a search reaches, it's whether the expensive check is running because a field actually earned it or because someone's still spooked from two seasons back. Kepler's team held price, put a plain flag, last check clear, three weeks ago, next to the deep-pass button so Odalie could see when she was paying for reassurance instead of information, and only capped weekly deep passes on fields with no flag and no history.

Swap the trigger and it still runs.
Speed: an interviewer gives you ninety seconds. Skip straight to the order: link to margin, watch the mix, name how it's gamed, act in stages.
Cost: there's no budget this quarter for both a usage-mix dashboard and a friction-adding confirmation screen. Build the dashboard first. You can't act on a threshold you can't see, and a confirmation screen with nothing behind it is just an annoyance.
The model got better, for real: say the archive search actually got more accurate, not just more expensive. That changes what the nudge says, better matches are available now if the role's genuinely stuck, but the order barely moves. You'd still want the mix watched and the cost visible, because better and free-to-use-constantly are not the same fact.

Where people run it wrong.
They watch what a customer bought instead of what a customer does with it.
They define the metric per seat, which is exactly the number a customer can quietly water down.
They cap everything account-wide because it's simpler to build, and end up throttling the 95 percent of usage that was never the problem.

How to use it live. Say the real question out loud before naming a metric: "am I about to measure what they paid for, or what they're actually doing with it." That buys a beat to find the real signal instead of reaching for the easiest number already sitting in the contract.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0