Customer Health

With twenty accounts, nobody needs a number for this. One person can hold twenty companies in their head — which one has gone quiet lately, how the last call sounded, who on the customer side actually opens the product. At three hundred accounts that approach falls apart. A single person can have maybe six or seven real conversations in a day, which leaves the other two hundred and ninety unexamined for a month at a stretch.

Customers keep leaving during the months nobody is looking. And the ones who leave rarely announce it. An angry customer calls. A customer who has quietly stopped caring says nothing at all, and on a dashboard a silent account looks exactly like a healthy one.

At Pagedeck the problem surfaced in a single phone call three weeks before a renewal. Sofia called to talk about the next year's contract and heard this: "We moved to another tool last month. Everything's already migrated." That company had not filed a single complaint in twelve months. Because it was quiet, it never landed on any list, and the quiet was the strongest signal available — there was just nothing set up to read it.

Looking back, the record was all there. Weekly active users had slid from 50 to 12 over six months, no new member had been invited in four months, and the person who signed the original contract had left the company in March. The trouble was that those three facts lived in three different places: usage counts in the product data, invitations in the admin console, the staffing change in a note inside the sales tool. Knowing the state of one account meant opening three systems and lining them up by hand, and nobody can do that for three hundred accounts every week.

Customer health is the attempt to collapse those three systems into one line per account. Pull the scattered signals into a single view, attach a stage to each company, and the morning routine becomes reading the twelve accounts marked red instead of skimming three hundred. The concept gets much easier to handle once you hold on to the fact that its job is ordering, not forecasting. Nobody needs to defend why an account scored 63 rather than 68. The question it has to answer is which customer gets called first today.

Customer success as a field splits into seven strands, and health sits in the one that judges present state. It stands in front of the work of keeping contracts alive and in front of the numbers that grade the team, picking out where attention should go. Where churn counts something that already happened, health aims at the weeks before it happens.

Six terms sit under customer health. They fall into four strands, and the dividing line is the sequence by which a signal becomes a judgment and a judgment becomes an action: what the raw material is (inputs), how that material becomes one value (the metric), how that value sorts customers (classification), and what gets done once they are sorted (activities).

Inputs cover the raw material of the judgment.

1. Usage signal — login frequency, whether the core feature gets touched, whether new members are invited, how many support requests came in recently. Which behaviors actually carry information differs by product, so teams find them rather than invent them. The candidates are the behaviors that customers who did leave shared in the weeks before leaving; the filter is whether customers who stayed also did those things. In a shared document product like Pagedeck, five different people editing the same file is a far stronger signal than one person logging in every day. Any single signal read alone will mislead. Watch login counts only, and the finance team that genuinely uses the product in a burst at quarter close turns red every quarter. Most teams end up with five or six signals rather than one, and they check each candidate against a year of history before letting it into the set.

The metric strand folds that material into one value.

1. Health score — a number from 0 to 100 built by weighting several signals. Pagedeck assigns 40 points to the share of the account's seats that were active this week, 30 to whether the core feature is in use, 20 to the support picture over the last 30 days, and 10 to whether the main contact has changed. The reason for collapsing signals into a number at all is sorting. Left side by side, an account with falling logins but lively support traffic and an account with the opposite pattern give no answer about which to open first. One number produces an order. The cost is that folding always discards information, so anyone who reads the score has to be able to unfold it back into the signals that produced it.

Classification cuts that value into pieces a person can act on.

1. Red account — a company flagged as likely to leave. Pagedeck drops anything under 45 points into that bucket automatically and lets the owner move accounts in or out by hand. The manual override exists because there are always facts the score cannot see. An acquisition, or next year's budget cut in half, appears in no usage log — but it does come up in a review meeting. The override runs the other way too: a low-scoring account with company-wide training booked for next month gets moved back to green.

Activities are what happens after the sorting, and there are three of them. They divide again by whether the account is doing well or badly.

1. Quarterly business review (QBR) — a standing meeting where the two sides go over what the last quarter produced and agree on what the next one should. It is not a summons reserved for troubled accounts; green accounts get one too. Its value is that facts the score cannot reach come out in conversation. A decision-maker has changed, or the team next door wants access — no amount of staring at usage data surfaces either.
2. Success plan — a document holding what the customer bought the product to achieve and the route to getting there. It sets the baseline against which health should be measured. If a customer's goal is to cut document review time in half, a healthy state for that account is faster review, not a larger headcount logging in. Without the document every customer gets judged against the same yardstick, and the ones succeeding in an unusual way come up red.
3. Escalation path — the route agreed in advance for handing a problem upward when the account owner cannot solve it alone, along with the timing. Skip it and the owner wrestles with the account for three weeks and raises it to the chief executive two weeks before renewal. At Pagedeck, two straight weeks of red brings in the team lead and four weeks brings in Grace.

The four strands run in one direction. Signals build a score, the score builds a classification, the classification sets the order of the work. There is a return path as well. What comes out of a review changes which behaviors get treated as signals, and the goal written in a success plan decides which signals deserve more weight. When that loop breaks, the score stays frozen at the assumptions of the day it was built. Six months later an account sitting at 82 leaves without a word, and that moment calls for a look at the weights rather than at the customer.

"This account dropped from yellow to red last week. Active users halved and we haven't had a single support request."

Silence is not reassurance here — a quiet account can be in worse shape than a noisy one.

"Let's work the reds this week and cover the green accounts in one pass at next month's review."

There aren't enough hands, so the stage decides who gets called first.

"They're sitting at 82 and they're not renewing. The score never picked up that their billing contact changed."

The thing to fix is the health model, not the customer.

At Pagedeck the phrase comes up most often in the Tuesday morning account check. Sofia puts up the list of accounts whose stage dropped last week and says in one line why each one moved. Fewer active users, a changed contact, a sudden spike in requests — each points to a different next step. An account with a new contact gets fresh training booked from scratch; an account with falling usage gets a look at which feature stopped being opened.

Mia reads the same list differently. She checks whether accounts using one particular feature are overrepresented on it. If they cluster around a feature, the cause is the feature rather than each customer's own circumstances, which makes it something to fix in the product instead of something to call about. Last quarter, nine of the seventeen red accounts were ones that had been using an outside integration, and when Theo checked, that integration had been quietly broken for a month. None of the nine reported it. They simply used the product less.

The most common mistake is reading green as safe. The traffic-light shape invites it. But this value summarizes what has accumulated so far; it does not predict. A customer who was using the product happily yesterday can get a budget cut today, and the score will still read 88 when the contract ends. Treating green as safety means nobody checks the side the score cannot see — the facts that only come out when someone asks. The gap widens with the size of the account, because the largest customers are the ones whose decisions get made in rooms the product never touches. A renewal can be killed by a new head of operations who has never logged in, and the seats that person inherited will keep showing healthy usage right up to the cancellation notice. That is exactly why review meetings stay on the calendar for green accounts too.

The second is making the score itself the goal. The number sits on a screen and moves every week, so wanting to raise it feels natural. But it is a sum of weights the team chose, which means the average can climb without anyone doing anything: cut the weight on support volume and next week's average rises eight points. Once the number feeds evaluations, teams do exactly that. Six months on, the average has gone from 71 to 84, the renewal rate has not moved, and nobody who looks at the number believes it any more. What deserves to rise is what customers actually get, and whether that rose is a question for renewal rates and expansion figures.