The Number That Goes Up When You Give Up

The last time I couldn't reach a person, I assumed someone had decided I wasn't worth the time. Turns out that's the flattering explanation. Nobody had decided anything.

Someone told the US consumer finance regulator about days spent trying to pay a credit card bill due that day — account closed without warning, phone line down. "I tried the chat feature and the virtual assistant kept sending me in circles… I am being sent in an endless loop with no way out."

HMRC, which published its own homework

The UK tax authority had a £149m customer-service savings target. Not someone's bonus — a budget line. Still money. The plan: move people online. In 2023–24, by the audit office's count, it stopped or restricted calls on four helplines. Its own evaluation records what happened to the customers it redirected. 113,000 callers with non-priority queries were directed to self-service — and 61% of them phoned the same helpline again within five working days, most of those calls within five minutes.

Notice where this bites hardest: places you can't leave. HMRC is the only tax office there is.

The chatbot that was performing help

I once ran a heuristic evaluation for a logistics company, cataloguing what was broken on the live site. I ended up watching a real customer's conversation with the chatbot. They asked to be put through to a person. The bot sent them to the FAQ page. They asked again, more specifically. FAQ page. Again. The build had no escalation mechanism — not hidden, not queued. Absent. This wasn't a service failing to help; it was a service being performative about helping. And when I looked at what that team was rewarded for — call volumes, down — the missing handover stopped reading as an oversight.

Nobody writes down "make the customer give up." They write down a target: cut cost per contact, raise the self-service rate.

Accounting researchers have a name for this — surrogation. Tie someone's pay to a measure, especially a single dominant one, and it stops being an imperfect stand-in for the goal. It becomes the goal. Choi, Hecht and Tayler showed it in a 2012 lab experiment; carrying it into a contact centre is my leap, not theirs.

It also isn't a law. Whether targets corrupt whatever they measure is genuinely contested — Kelman and Friedman tested the English NHS's four-hour A&E target across all 155 hospital trusts and found large improvements and no measurable gaming. Measures don't always eat their goals. They eat them under conditions: one dominant number, and real money riding on it.

The number here is a containment metric: it counts contacts that never reached a person. It stands for "we helped them without needing an agent" — but it can't tell you whether the problem got solved.

This isn't an argument for putting the phone number back: self-service is genuinely good, and less of it is not automatically better. Tracking customers over time, Scherer and colleagues found a U-shaped relationship — an intermediate mix of self-service and personal service kept the most customers. And for many, the phone is the barrier, not the route past it: in Money and Mental Health's survey work, 54% of people who had experienced mental health problems found the telephone difficult or distressing.

So the defensible rule isn't a channel, it's a guarantee: whatever door someone comes through, a route to a person who answers and can act — the thing the logistics build didn't have.

Here's how I'd sort the instruments. The third column is the argument. Not because those numbers are honest — they get gamed too. Because Choi and colleagues' finding was about being paid on one measure versus several:

None of it is exotic. HMRC ran the first one on itself — then published the 61% the same day it announced a longer closure. Measuring the gap is cheap. Acting on it isn't.

The wall isn't personal, and often isn't a decision anyone made about you. Somewhere upstream is a number that goes up when you give up, and nothing next to it goes down. So the question worth asking of any service — including the one you work on — is what goes down next to it when someone gives up. If the answer is nothing, the loop isn't a bug. It's the number working.

Reply and tell me: when did you last go hunting for a human — and what worked?

Next Wednesday: Who designed the other side of the counter? — the staff tool that adds three steps to the job it was bought to simplify. Backstage needs designing as deliberately as frontstage.

Forwarded this? The Listening Loop pulls apart one invisible piece of service design every Wednesday.

See you next week.

Go deeper: Chatbots in Consumer Finance (Consumer Financial Protection Bureau, 2023) — a regulator patiently documenting what happens when a financial company's front door is a bot, in the complainants' own words.

Sources: