The Most Loyal-Looking Customers

In August I showed you the enterprise paradox: the typical enterprise-focused company keeps 98% of its customers through year one and bonds essentially none of them. A cage, not a bond.

Most people filed that under “enterprise problem.” It isn’t one. Enterprise is just the most expensive example of something the data shows everywhere: the customers who look most loyal are the ones you know the least about.

Here’s the background fact that makes this urgent. In the behavior of 1.3 million customers across roughly 70 B2B SaaS companies, the average company has bonded about 6% of its customers — with them for good. Roughly 92% are still leaving: paying now, gone eventually. So when you point at an account and call it loyal, the odds are overwhelming that you’re pointing at a customer in the second group. The question is whether any of the surface signals you use — size, tenure, usage, scores — can pick out the bonded few.

The data says no. Take them one at a time.

The biggest account. The instinct is that big customers are attached customers — they did the diligence, they integrated deeply, they have the most to lose. But decompose all the variation in who bonds, and 88% is which company they bought from — the product — and only 12% is how big the customer is. Split almost any customer base into its largest and smallest customers and the bonded fraction barely moves. One case from the study: a company selling to small businesses bonded about 75% of them, while a competitor selling to much larger, better-funded buyers bonded almost none — its biggest accounts no stickier than its smallest. Put company size and deal size head to head and only one signal survives: commitment. Serious customers stay longer. Bigger customers just look like they do.

The longest tenure. This is the one that fools the most careful people, because tenure is real behavior, not a survey. But a customer can be retained and still be on the way out — paying through a contract, coasting on inertia, quietly shopping alternatives. A company can keep its customers three years or more on average and still have a bonded core of zero: every customer leaving eventually, just slowly enough to make the average look reassuring. Two companies can share the identical average tenure where one has built a durable core and the other a slow conveyor belt to the door. Your ten-year logo tells you the customer hasn’t left yet. It does not tell you they’ll be there when leaving gets easy.

A customer base can average three years of tenure and have a bonded core of zero. Long tenure is not loyalty. It can be a long, slow exit.

The heaviest user. Utilization is the most seductive false signal, because it feels like progress. It isn’t. Usage does not cleanly predict retention — at one company in the study, the heaviest users churned as fast as the lightest. Activity was never the goal. The result is. Adoption that doesn’t produce a result bonds no one; teams keep mistaking the means for the end.

The happiest customer. Test each customer’s score against that customer’s lifespan and there is no statistically significant difference between promoters, passives, and detractors — and what direction there is runs against the promoters. I’ve been saying this since 2018: satisfaction fails to explain retention because there are happy customers who leave and angry customers who stay. The worst customer I ever had taught it to me in one sentence. He was demanding, never satisfied, and I finally asked why he didn’t just cancel. His answer: “Why would we cancel when you get us such good results?”

The green account. Same story for the CSM’s gut call. At a company that scored account health, the accounts rated “Great” churned no better than the ones rated “Poor.” A health score measures how your success team feels about an account — not whether the account got a result.

The heaviest users churn as fast as the lightest. Promoters retain no longer than detractors. Accounts rated Great churn like accounts rated Poor. Every signal that feels like loyalty fails the same test against actual behavior.

Notice what these signals have in common. Size, tenure, usage, a survey score, a health color — every one of them is free. The customer gives up nothing to produce it. A big account costs the customer nothing to be. Tenure accrues on its own. Clicking around the product costs minutes. A 9 on a survey costs one second of goodwill.

Now look at the signals that do predict who stays. They all cost the customer something.

Commitment at the sale. Customers who buy without a discount go on to last 2.6 times longer than discount buyers. A discount used to close a wavering buyer tracks with shorter loyalty; a discount traded for a longer commitment tracks with longer. A customer who commits is telling you they intend to get a result. The mechanism is skin in the game: a customer who paid little to nothing for it will do little to nothing with it.

The effort of complaining. Negative experiences correlate with higher retention — customers who complain actually tend to stay much longer. A complaint is work. Nobody fights with a vendor they’ve already decided to leave.

The work of reaching a result. Customers who reached a measured result formed a ~99% bonded core and essentially stopped leaving. Customers who didn’t had no core at all.

The loyalty signals that mean nothing cost the customer nothing. The ones that mean something all cost the customer something: a serious commitment, the effort of complaining, the work of reaching a result.

So how do you find your actual loyal customers — the bonded 6%, or whatever your number is? Not by inspecting accounts one at a time. You read the shape of retention over time. Follow your customers from signup and plot the share still with you. The curve falls as the still-leaving customers churn — then, if a durable core exists, it stops falling and flattens onto a floor. The height of that floor is the share of your customers who are bonded. It is the one retention number that can’t be faked by a good quarter or a long contract, because it counts only the customers who are still there long after both are gone.

Every loyal-looking signal on your dashboard counts customers who haven’t left yet. The floor counts the ones who never will. You can read yours from your own data.