Read Ten Tickets a Week
Every layer between a leader and reality has an incentive to make the news better. Not from dishonesty — from kindness, from summarization, from the fact that every metric is being optimized by someone and every summary is written by a person who knows what you hope to hear. Charlie Munger spent fifty years cataloging how incentive-caused bias bends information on its way up an organization. The bend is structural. You cannot fix it. You can only route around it.
The cheapest route-around I know: read ten full support tickets a week. Personally. Forever.
Not summaries of tickets. Not a dashboard about tickets. Not your CX lead’s excellent weekly digest, which is genuinely excellent and is also a filter. The raw conversations — what the customer actually typed, with the typos and the third rephrasing and the sentence where politeness finally cracked; what your company actually said back, including what your AI said when no human was watching.
The selection rule matters more than the count: choose badly. Happy-path tickets teach nothing. The ten that pay are escalations, churned accounts’ last conversations, tickets that bounced between owners, anything the customer reopened after it was “resolved,” and a few from the deflected pile your containment metric counts as wins. You are not sampling for representativeness — the dashboard already does that, honestly enough, for the middle of the distribution. You are sampling the tails, because the tails are where the dashboard has no categories. Every operationally important discovery I’ve made in a support queue — a billing failure that read as “user confusion,” a product gap disguised as a training issue, an AI answer that was fluent, confident, and policy-violating — was invisible in the aggregates and obvious in the third transcript.
Why the founder specifically, and why it survives delegation attempts: the value isn’t that the tickets get read — your QA team reads hundreds. The value is that the person with authority to change anything reads them. A QA analyst who finds a product flaw files a report that crosses four desks, softening at each one. A founder who finds it changes the roadmap on Thursday. Reading the tickets yourself collapses the distance between signal and decision to zero — that’s the entire mechanism, and it’s also a latency argument about your org chart in miniature. This is the same reason the practice must not end when you hire a real support org. The week you stop is the week the laundering resumes, with better stationery.
The practice: a recurring calendar block, non-negotiable and non-delegable. Ten conversations, selected by the bad-news rule. One line of notes per ticket — what surprised me — in a running log; the log matters because patterns emerge across weeks that no single ticket shows. Every surprise files somewhere real: a knowledge-base bug, a product ticket, a question for the next ops review. And once a quarter, do it as an executive team — ten tickets on a screen, no slides. It will be the most clarifying meeting of the quarter and routinely the least comfortable, which I have come to believe are the same property.
In the AI era, the practice has acquired a second job. Your AI now conducts most of your customer conversations, and its failures are confident, identical, and silent — invisible to a satisfaction average, glaring in a transcript. Reading raw conversations is no longer just founder discipline; it’s the human audit layer over a machine that talks to more of your customers than your whole team combined. Ten a week is the minimum honest dose.
The standard objection is that it doesn’t scale. Correct — and that’s the design. It is not a coverage function; it is a sensing function. Your metrics give you coverage. Ten raw conversations a week give you the thing metrics structurally cannot: contact with what your company feels like from the outside, unbent by anyone’s incentive to tell you it’s fine. The companies that drift aren’t the ones with bad dashboards. They’re the ones where nobody with power has heard a customer’s actual voice in a year.
Sources & further reading
- The Psychology of Human Misjudgment — incentive-caused bias: why every layer improves the news.
- ‘Improving Ratings’ — why the metrics can’t save you: targets bend measures.
- Related: Deflection Is Not Resolution — what the deflected pile hides; Goodhart’s Revenge — the theory this practice routes around.