Nearly everybody has a story: the chatbot that won't escalate, the phone menu with no path to a human, the hour on hold, the case that gets closed without resolution.
It's tempting to attribute this to incompetence, but the pattern is too consistent across too many industries. What's actually happening is a set of rational decisions producing a collectively irrational outcome.
Support as a cost centre
The structural starting point. In most organisations, customer service sits in the budget as a cost, not as a source of revenue.
Costs get minimised. That's what finance functions do, and it's not unreasonable in general. The problem is that the benefits of good support — retention, word of mouth, reduced churn — accrue elsewhere in the business and are difficult to attribute to specific support interactions.
So the costs are visible and measurable, and the benefits are diffuse and contested. Any negotiation between a support director and a finance director under those conditions has a predictable outcome.
The metrics that break things
What gets measured determines behaviour, and the standard support metrics have well-known pathologies.
Average handling time. Reward agents for shorter calls and they end calls faster. Some of that is efficiency and some of it is closing cases that aren't resolved, because the customer calling back tomorrow is a new case rather than a failure of this one.
First contact resolution. Better as a metric and gameable in its own way — a case marked resolved is resolved, and whether the customer agrees is a separate question.
Cost per contact. Drives deflection. Every interaction that can be pushed to a self-service channel reduces the number, which is why finding a phone number has become an exercise in persistence.
Satisfaction surveys. Response rates are low and skewed towards extremes, and agents know which behaviours produce good scores, which are not always the behaviours that solve problems.
None of these metrics is stupid. The issue is that optimising them individually produces a system that's efficient at everything except helping people.
Deflection as a strategy
The term used internally is deflection, and it's worth knowing because it explains a great deal.
The goal is to resolve queries in the cheapest available channel: help articles cheapest, chatbot next, live chat next, phone most expensive. Every layer exists to reduce the number of people reaching the next one.
Applied well, this is genuinely good — many queries are simple and self-service is faster for everybody. Applied as a cost-minimisation exercise, it becomes a maze designed to exhaust people before they reach a human.
The tell is whether escalation paths are clearly signposted. A well-designed system offers self-service and makes it easy to skip. A deflection-optimised one hides the exit.
The chatbot problem
Automated support has improved substantially and the deployments frequently haven't kept pace.
The failure mode isn't the technology; it's configuration. A system with no confident answer should hand over immediately. Many are configured to keep trying, because every handover costs money, which produces the experience of arguing with something that cannot help you and will not let you go.
A well-implemented automated system handles the simple majority and escalates cleanly. A badly implemented one is a wall, and the difference is a policy decision rather than a capability one.
Why complaints get results
A practical observation that reveals the incentive structure: escalation to a public channel or a regulator frequently produces resolution within hours after weeks of nothing.
That's because those routes have different cost structures attached. A public complaint carries reputational cost. A regulatory complaint carries a case fee in some sectors, plus reporting obligations.
Suddenly the interaction is worth resolving. Which tells you the resource always existed; it just wasn't allocated to you.
What good looks like
Organisations that do this well tend to share a few features.
Support has a route into product decisions, so recurring problems get fixed at source rather than handled repeatedly. This is the single highest-return intervention and it requires organisational structure rather than budget.
Agents have authority to resolve without escalation, within defined limits. Most bad experiences involve somebody sympathetic who isn't allowed to help.
Metrics are about resolution rather than throughput.
And there's a visible, easy path to a human. Companies confident in their product don't need to hide.
Practical advice
For dealing with the system as it is: be specific and unemotional, because the person you're speaking to has limited authority and no responsibility for your problem. State what you want to happen rather than describing your frustration.
Ask directly whether they have the authority to do what you're asking. If not, ask to be transferred to someone who does.
Keep a written record with dates, names and reference numbers.
And know the escalation routes that exist in your jurisdiction — ombudsman schemes, regulators, chargeback rights. Not to threaten with, but because knowing they exist changes how you approach the conversation, and because after a reasonable attempt they're frequently the only thing that works.
The offshoring question
Worth addressing directly because it comes up constantly and is frequently misdiagnosed. Poor support experiences get blamed on offshored operations, and the causal story is usually wrong.
Agents in offshore centres are typically working from the same scripts, with the same authority limits, against the same metrics, as agents anywhere else. When they cannot help, it is nearly always because the system does not permit them to, not because of where they are sitting.
Where location genuinely matters is context — knowledge of local regulation, local product variants and local expectations. That is a real problem and it is a training and resourcing decision rather than a geographic inevitability. Well-run offshore operations handle it; badly-run domestic ones frequently do not.