The double-edged sword: why mediocre support costs more than good support

From the engine room – bigger-picture thinking on support and CX, beyond Zendesk too.

At many companies support counts as a cost center you keep small. Behind that sits an assumption: that support acts linearly. A bit more quality, a bit more satisfaction, a bit less churn. The assumption is wrong. In our customer surveys, support acts with a threshold and a sign, not as a sliding scale.

No neutral middle ground

Mediocre support pulls referral sharply down. Good to very good support pulls it clearly up. Between the two there’s no quiet zone. The middle is the dangerous one: customers who are neither annoyed nor bound, and who leave at the next opportunity.

That’s an observation from one operation to start with. Loyalty research has known the pattern for thirty years.

What the research says

The threshold is one of the best-documented effects in CX research. Jones and Sasser called it the «Zone of Indifference» in the Harvard Business Review in 1995: only fully satisfied customers become loyal, while merely satisfied ones churn easily. Anderson and Mittal showed in the Journal of Service Research in 2000 that the chain from satisfaction to loyalty runs nonlinearly and asymmetrically; linear models systematically underestimate it. The Kano model splits the same thing into basic factors, which punish hard at the bottom, and delight factors, which bind further at the top.

Support acts with a threshold and a sign. That’s not a gut feeling, it’s textbook.

The lower edge cuts sharper

The sword isn’t symmetric. The negative side weighs more than the positive one, and Kahneman and Tversky’s Prospect Theory (1979) gives the reason: losses count psychologically about twice as much as equal-sized gains. For support that means: bad support hurts more than good support helps, at the same distance from the middle.

So the statement isn’t «good is as strongly positive as mediocre is negative.» It’s: the downside is bigger than the upside. Both signs are real, the lower one is sharper.

The objection

The best-known counter-voice comes from CEB, today Gartner: service drives disloyalty four times more often than loyalty, and delight in support barely pays off («Stop Trying to Delight Your Customers», HBR 2010). That sounds like a rebuttal to the positive side. It isn’t one. First, the finding confirms the negative half. Second, CEB itself stands on thin ice: De Haan, Verhoef and Wiesel (2015) found that CEB’s favorite metric, the Customer Effort Score, predicts retention worst of all metrics.

Both hold, depending on context. Routine support is a hygiene factor: the upside saturates fast, and here CEB is right. Relationship-driven support becomes a delight factor: here the upside applies. Whoever handles mass routine optimizes for effort. Whoever advises leaves value on the table the moment they settle for «error-free».

Measure the danger zone, not the cut

Under asymmetric effects, the average lies. A rising CSAT or NPS mean can mean the top is gaining, or that the middle is growing and a few outliers pull the value up. The mean hides exactly the shape that matters.

So measure a second number next to the average: the share in the middle. The percentage of ratings that sit neither low nor high, but in the churn-ready zone. Not «how satisfied on average», but «how many sit in the expensive middle», and whether that share is shrinking.

What this means for the operation

The point is uncomfortable: mediocre is the most expensive tier. It costs twice, in the operation and in lost referral. Whoever stays just below the threshold pays for support and loses customers anyway.

That shifts the investment question. Perfection everywhere buys little, because the upside saturates. What counts is lifting the cases that count safely over the threshold and never falling below it anywhere. No study tells you where the threshold sits. Your numbers do.

This is where what the bot can do splits from what the human can do. The bot is consistent: it keeps support away from the catastrophic edge, the same in every ticket, and secures the lower edge that way. It lifts nothing over the threshold at the top. Only the human does that, the one who has time for the conversation. The bot prevents the bad, the human creates the good. What you do with the capacity the bot frees up decides whether you ever get over the threshold. That’s the next question; how to read resolution rates alongside it is in the note on resolution rates.

Support isn’t a cost item with linear effect, but a lever with a sign. The companies that grasp this stop keeping their support small and start lifting it over the threshold. The most expensive thing you can do is leave it standing in the middle.