Why do companies make terrible decisions? With Dan Davies

Episode Summary

Executive Summary: Dan Davis argues that large organizations fail because decision-making scales more slowly than complexity, creating “accountability sinks” where nobody owns outcomes. He criticizes economists for modeling firms as black boxes, defends middle managers as vital information processors, and says AI will matter only if used to redesign organizations rather than automate old bureaucracy.

Main Topics: The Unaccountability Machine and organizational failure: Davis explains how complex systems generate bad outcomes without any single person intending them, making accountability diffuse and decisions hard to trace. Accountability sinks and biased system interactions: He uses examples like exam appeals to show how individually fair processes can combine into structurally biased outcomes that no one directly caused. Why economics misses management: He argues economists over-simplify firms and governments as black boxes, ignoring internal information flows, incentives, and organizational design. Central banking and the financial crisis: Davis says central bankers focused too narrowly on inflation and missed debt buildup because they filtered information through the wrong framework. Middle managers as essential connectors: He reframes middle managers as crucial translators of context and boundary-crossers who keep large organizations coherent. Cybernetics as the path not taken: Davis highlights Stafford Beer’s cybernetics as an alternative management science focused on feedback, bandwidth, and control in complex systems. AI, black boxes, and organizational redesign: He treats AI as another information technology and warns that its value depends on redesigning institutions, not merely accelerating existing bureaucracy.

Key Arguments: Large systems become less controllable as they grow because complexity rises faster than management capacity. Many harmful outcomes are emergent rather than malicious: no single decision-maker intended them, yet the system still produced them. An accountability sink is a structure that absorbs criticism so no individual or team can be held clearly responsible. Two fair processes can combine into an unfair system; the transcript’s GCSE/appeals example shows how bias can emerge without explicit intent. Economics has helped create blind spots by treating organizations as simplified black boxes and neglecting management science. Central banks missed the build-up to the financial crisis because they privileged inflation targets over broader signals like debt accumulation. Middle managers are not dead weight; they provide translation, feedback, and context that keep information useful across organizational boundaries. Outsourcing and delayering reduce internal information-handling capacity, often making organizations more fragile and prone to breakdown. Populist leaders attract support by offering simple explanations in a complex world, even if they cannot באמת control the system. AI does not raise entirely new ethical issues; it repeats long-standing questions about opaque, non-human decision systems. The major opportunity with AI is organizational redesign—using it to improve information processing and decision structures, not just to automate routine tasks.

Data Points: Complexity scale: 5–6 normally desirable; current world about 8, rounding up to 9 - Davis’s calibration of how close society is to “losing its mind” Financial crisis timing: Early 2000s - Referenced as the period when central bankers focused too narrowly on inflation Middle management reduction period: 30–40 years - Approximate period over which outsourcing, downsizing, and delayering reduced middle management Human life comparison: Less than a human life - Time span in which decision-making shifted from individuals to algorithms, committees, and rules KLM incident: 400 pet squirrels - Example of a bizarre organizational failure mentioned to illustrate accountability sinks Corporate management history: Post-Second World War acceleration - Davis says industrialized decision-making expanded rapidly after WWII Cybernetics origin: 1950s - Period when Stafford Beer’s management cybernetics emerged John Paul Getty example: Daily reports from rigs - Illustrates an older model of centralized, individual managerial control AI/IT historical analogies: Telegram, double-entry bookkeeping, filing cabinet - Used to argue AI is another transformative information technology

Pivotal Quotes: "there's just this sink into which any kind of negative feedback and criticism of the system can be directed" — Dan Davis: Defining an accountability sink through the exam-marking and appeals example "the central problem of management is matching the information processing capacity of the decision-making center to the volatility and variety of its environment" — Dan Davis: Explaining the core cybernetic view of management "it's like you have the opportunity to hire Galileo, Shakespeare, and Einstein, and you put them in a room and told them to memorize the phone book" — Dan Davis: Criticizing the way firms use AI and mainframe-style computing for low-value automation rather than redesign

Implications: Listeners are left with a framework for diagnosing organizational failure: widen accountability, preserve middle-management bandwidth, and redesign systems around information flow. AI may help, but only if it changes structure rather than speeding up old dysfunction.