Somewhere in Britain, right now, someone is on hold. They have heard the same four bars of Vivaldi seventeen times. They have been told their call is important. They have been told, in a tone of very mild regret, that due to unprecedented demand wait times are longer than usual. They are, to a first approximation, furious.
And yet, if you asked them exactly what had upset them, they might struggle to say. Was it the time? Was it the music, the voice, the word ‘unprecedented’? Or was it the creeping suspicion that nobody was coming, that the queue they had joined might not, in any meaningful sense, be a queue at all?
This is the odd thing about queuing. It is almost never about the queue.
For about sixty years, every time a technology has promised to fix the problem of waiting, customers have found fresh and imaginative ways to remain annoyed. Which brings us to the moment we are in now.
The question, after the viral chatbot disasters and the quiet chatbot triumphs of the last two years, is whether AI, with its convincingly human voices and its three billion interactions a month, can finally solve a problem that has baffled industrial engineers, airport planners, theme park designers and contact center managers since roughly the Eisenhower administration.
The short answer is: probably not in the way you think.
Why the Psychology of Queues is Almost Never About the Queue
Some years ago, Houston airport had a problem. Passengers kept complaining about how long they waited at baggage claim. The airport, being sensible, did the sensible thing. It hired more baggage handlers. Wait times fell to an industry-best eight minutes. Complaints continued.
The analysis was eventually conducted by Richard Larson, the MIT professor known, charmingly, as Dr Queue. It turned out passengers were walking a minute from the gate to the carousel and then standing for seven. So, the airport moved the arrival gates further away. Passengers now walked six minutes and waited two. Total time unchanged. Complaints vanished.
The lesson is almost banal and almost always forgotten. What we experience as waiting is not a number of minutes. It is a number of emotions.
As Larson put it:
“Often the psychology of queuing is more important than the statistics of the wait itself.”
We have built an entire multi-billion-dollar queue management industry on the assumption that queues are fundamentally an engineering problem, when in truth they are fundamentally a humanities problem.
David Maister and the Eight Rules That Still Govern How We Wait
The person who did the most to formalize all of this was a former Harvard Business School professor called David Maister, who in 1985 published a paper titled The Psychology of Waiting Lines. Four decades on, it remains the most stolen-from piece of writing in customer experience.
Maister began with what he called the first law of service: satisfaction equals perception minus expectation. If the sign says twenty minutes and you wait fifteen, you are delighted. If the sign says five and you wait seven, you are livid. The wait is the same. The sign did the work.
His eight principles explain everything that follows. Occupied time feels shorter than unoccupied time. Pre-process waits feel longer than in-process waits. Uncertain waits feel longer than known ones. Unexplained waits feel longer than explained ones. And unfair waits feel longest of all.
That last one is the killer. Research by Harvard’s Ryan Buell found customers at the back of a line are 3.5 times more likely, per second, to abandon it than those with even one person behind them. When he removed the cues of being last, abandonment fell by 43.5 percent. Customers aren’t bailing because the queue is too long. They’re bailing because they can’t see anyone worse off than themselves.
Fairness, not speed, is the thing we are actually measuring when we stand in line. We will wait a very long time if we feel the system is playing straight with us. We will leave quite quickly if we feel it is not.
Waiting Is No Longer Just a Phone Problem
There is a further complication. The waiting room has moved. Or rather, it has multiplied.
For most of the twentieth century, the queue was a single channel: you stood in a line, or you stayed on hold. The psychology Maister described was essentially linear. Today, a customer’s wait might begin in an app, migrate to a chatbot, pause in an asynchronous messaging thread, resume in a callback, and conclude – if they are patient enough – with a human agent on a video call. Each transition is its own small act of abandonment. Each new interface resets the customer’s sense of where they are in the process.
The research on omnichannel waiting is younger than Maister’s principles, but its findings rhyme with them. Customers tolerate asynchronous waits – the held email, the ‘we’ll get back to you’ message – significantly better than synchronous ones, provided they believe someone has actually received their query. The uncertainty principle remains. What has changed is the scale of the uncertainty: in an omnichannel environment, customers can be unsure not just about when they will be helped, but about whether the different parts of the system know they exist at all.
The Great AI Queue Disasters of 2024
In late February 2024, Klarna’s CEO Sebastian Siemiatkowski announced, with some fanfare, that his company’s new OpenAI-powered assistant had handled 2.3 million customer conversations in its first month. This, he said, was the equivalent work of 700 full-time agents. Teleperformance, a large, outsourced contact center company, saw its shares drop 29 percent on the news.
The pattern repeated across the industry. A Canadian tribunal ruled Air Canada liable after its chatbot wrongly promised a bereavement discount – the airline had argued the chatbot was a “separate legal entity” responsible for its own actions.
In each case, the technology answered the call. It just failed at the human part. The maths won, and the psychology broke.
In each case, a company reached for AI not to augment the wait but to replace the human behind it. In each case, the technology worked brilliantly at the narrow engineering task and failed spectacularly at the broader human one. The calls were answered. The orders were taken. The chatbot did, in a technical sense, respond.
And yet the customer was left feeling less heard, less certain, less fairly treated.
Where AI is Calming Customers Down
The success stories of AI customer service share a wholly different architecture. Bank of America’s Erica has handled more than three billion interactions across 50 million customers. She has no personality to speak of, no generative flourishes – just a tightly scoped, deterministic system that won’t answer questions it isn’t sure about. Similarly, Octopus Energy’s AI, used to draft emails for human agents to review, lifted customer satisfaction from 65 to 85 percent. The humans weren’t replaced; they were given tools to help them do their jobs more efficiently.
Look carefully at these examples. None of them is pretending to be human. None of them is claiming to replace 700 agents. They are tightly scoped. They are deterministic or retrieval-based rather than broadly generative. They tell you what they are. They resolve quickly when they can and escalate cheerfully when they cannot. They assume, reasonably, that customers would like to be waited on rather than waited around.
They are, to borrow Maister’s framing, doing the psychology before they do the maths.
What Modern Analytics Sees That We Don’t
There is a version of this problem that moves from the philosophical to the operational. If the real cost of a queue is psychological – the moment a customer decides they are being ignored, or treated unfairly, or that the system cannot help them — then the question becomes whether that moment can be detected in time to prevent it.
This is where modern customer analytics and intelligence platforms are beginning to do something genuinely new. Journey analytics can map the precise points in a service interaction where customers abandon or escalate, translating Maister’s abstract principles into addressable friction. Behavioural modelling can identify, before the customer does, the patterns that predict dissatisfaction. Sentiment intelligence – applied in real time to voice, text and chat – can detect frustration at the word level, or even the pause level, and route accordingly.
Predictive abandonment analytics goes further still. Rather than waiting for a customer to leave, it identifies the conditions under which they are likely to – the wait duration, the channel, the query type, the hour – and intervenes before the decision is made. A callback offer at minute six rather than minute nine. A different routing choice for a customer who has already transferred twice. An escalation flag raised before the customer raises their voice.
None of this replaces the psychological insight Maister and Larson identified. It operationalises it. The principles are the same: reduce uncertainty, demonstrate fairness, give the wait a shape and a reason. The difference is that modern AI orchestration systems can now apply those principles at a scale and granularity that no human workforce ever could – not because they are cleverer than the customer, but because they are paying closer attention.
Can AI Solve our Queuing Anxiety?
The latest data finds 85 percent of consumers still prefer a human for customer service. Forrester predicts a third of new AI service rollouts in 2026 will fail outright, and Gartner predicts agentic AI will autonomously resolve 80 percent of common issues by 2029.
The question is whether board-led AI initiatives can strike a balance with customer preferences. Perhaps both things can be true.
The old question – can AI answer faster than a human? – has been answered. Yes, for a tenth of the cost. However, the real question is whether it can do the other things a human does without realising: acknowledge, explain, reassure, apologise, hold the line of fairness, notice you’ve been waiting, notice you’re upset.
Those are Maister’s eight principles in new clothes. Every AI customer service triumph of the last two years has honoured at least one of them. Every disaster has ignored them.
The AI that finally soothes our queuing anxiety won’t be the cleverest or the cheapest. It will be the one whose designers understand the psychological principles behind our stresses.
FAQs
What makes a successful AI customer service deployment?
The strongest examples like Bank of America’s Erica and Octopus Energy’s drafting tool are tightly scoped, honest about their limits, and designed to escalate gracefully rather than pretend to be something they’re not.
Is faster service always better service?
No. Research consistently shows that customers will tolerate longer waits if they feel the process is fair, explained, and that someone knows they’re there.
How does omnichannel support change the waiting experience?
Every channel transition resets a customer’s sense of progress, compounding uncertainty across the journey rather than just at a single point.
Can AI predict when a customer is about to give up?
Yes – predictive abandonment analytics can identify the conditions that precede dropout and trigger interventions, like a callback offer, before the customer makes the decision to leave.
Will AI ever fully replace human customer service agents?
Gartner predicts AI will autonomously resolve 80 percent of common issues by 2029, but with 85 percent of consumers still preferring humans for service, the most effective model is likely to remain a collaborative one.
