Don't Waste a Good Complaint
Turn friction into leverage.
Your chatbot escaped. The customer didn't.
In a recent Reddit post, a Booking.com customer described arriving at a hotel only to learn the reservation had been canceled. The app offered a call button. The call connected to an AI agent.
Their complaint was admirably direct:
“Does anyone know how to get hold of a human at customer service? ... The ‘call us’ option within the Booking.com app is an AI agent—impossible to actually talk to a person.”
The hotel room was gone. The customer needed judgment, authority, and a next move.
The system offered conversation.
That difference is the whole problem.
New here? Every two weeks, Don't Waste a Good Complaint takes one real operating failure, scores the hidden opportunity, and shows what AI should handle, what humans should decide, and what leaders should measure.
The Friction File
The complaint is not that customers hate every chatbot.
They hate being trapped inside one after it has stopped being useful.
Parloa, an AI customer-service vendor, commissioned a survey of 1,001 U.S. adults. More than half said they had tried to circumvent an automated system. Of those, 43.9% had yelled “human” or “person.”
A separate vendor-commissioned study of 6,000 consumers across the U.S., U.K., and Canada found rising frustration with AI service and a growing preference for human support. The vendors have commercial interests in the outcome. The pattern is still worth measuring inside your own operation.
We trained machines to understand natural language, then trained customers to shout one magic word—HUMAN—like they are trying to escape an IKEA designed by Kafka.
The absurd part is what happens inside the company.
The chatbot may count the interaction as contained. The customer did not reach an agent. Handle time stayed low. The automation dashboard remains green.
Meanwhile, the customer repeats the request, opens another channel, calls again, abandons the purchase, disputes the charge, or leaves.
The conversation was deflected.
The customer may have been, too.
F2L Score: 78 / 100
The Friction-to-Leverage Score asks whether a complaint is just noise or whether it exposes hidden operating value.
This one scores 78 because the failure is frequent, measurable, economically important, and fixable.
The score breaks down this way:
Input | Score |
|---|---|
Customer effort | 27 / 35 |
Cost of false containment | 28 / 35 |
Escalation quality | 23 / 30 |
Total | 78 / 100 |
Customer effort measures how hard someone must work to explain the issue or reach the right authority. False containment captures repeat contacts, cancellations, disputes, and churn hiding behind a successful deflection. Escalation quality asks whether the human receives the context and authority needed to resolve the problem.
A rough key: under 40, it is noise—let it go. From 40–70, it is redesign territory. Over 70, it is worth funding and governing this quarter.
This clears the bar because customer-service AI is no longer an experiment. Companies are under pressure to deploy it, reduce support costs, and show automation gains. If the measurement system rewards containment without verifying resolution, it will optimize the wrong behavior at scale.
The Real Bottleneck
Most chatbot strategies start with:
What percentage of conversations can we automate?
Start with a different question:
Which customer intents can the system resolve safely, and what must happen when it cannot?
That change sounds small. It changes the operating model.
A useful AI agent should resolve routine, bounded work quickly: order status, simple account information, appointment confirmations, basic troubleshooting, and policy questions.
It should also recognize when the job has changed.
A missing hotel room at check-in is not an FAQ. A disputed medical bill is not a search query. A cancellation involving money, risk, safety, or emotion is not improved by a more conversational dead end.
The same failure appears inside regulated operations: a benefits applicant trapped in eligibility logic, an employee blocked by an identity workflow, a contract manager unable to resolve an exception, or a patient routed through automation after the situation has become clinical rather than administrative.
The AI does not need to become human.
It needs to know when a human is required.
The Move
Build an escalation contract, not an escape hatch.
Name resolvable intents. Define the requests the AI can complete, not merely discuss.
Name mandatory escalations. Refunds, cancellations, safety, legal, medical, billing disputes, vulnerable customers, and emotionally charged exceptions should have explicit rules.
Detect the loop. Repeated intent, rephrased questions, shorter messages, negative sentiment, multiple contacts, and the words “human” or “representative” are operating signals.
Transfer context. The customer should not have to repeat the account, history, failed steps, and desired outcome.
Give the human authority. An escalation to someone who can only read the same policy aloud is theater with a headset.
Measure resolution. Track repeat contacts, customer effort, time-to-human after an escalation signal, post-contact churn, and whether the promised action actually occurred.
Deflection can be useful.
Resolution is the outcome.
Confusing the two is how a cost-saving program quietly becomes a customer-deletion program.
Know an operator whose dashboard says success while the customer is still stuck? Forward this issue to them. They can subscribe here.
One Hard Thing
At the 2026 Hardrock 100, Courtney Dauwalter and Ludo Pommeret beat their own course records across 100 miles of the San Juan Mountains.
What looks superhuman from the finish line is a series of ordinary decisions made under fatigue: adjust, solve the next problem, and notice when conditions have changed. Resilience is adapting before the failure becomes the story.
One Song Worth Hearing
“Grey Cell Green” by Ned's Atomic Dustbin
Two bass players, sharp elbows, no unnecessary polish. It moves like someone just found the human-agent shortcut.
One Moment of Awe
An aurora, viewed from above
ESA astronaut Sophie Adenot captured a green aurora moving below the International Space Station. From Earth, it hangs above us. From orbit, it wraps around the planet.
Same system. Entirely different perspective.
What's your F2L Score?
The newsletter scored the chatbot escape room at 78.
The AI Opportunity Assessment applies the same logic to your actual workflows: where friction is real, where AI can create leverage, where human judgment must remain, and what should be funded, redesigned, governed, or killed.
For $5,000, you get ranked AI opportunities, a governance and risk map, and a 90-day sequence for action.
Your turn
Where does your customer have to fight the system to reach someone who can actually decide?
Reply with the workflow. I am looking for the places where automation reports success while the customer is still stuck.