Of all the bosses likely to mount a defence of gut feel, Ted Sarandos is among the less expected. Netflix’s co-chief executive built his reputation on data: hundreds of millions of viewing events, sliced and weighted to decide what gets commissioned. He has called himself a human algorithm. He has described House of Cards, the political thriller that launched Netflix’s originals slate, as algorithmically generated. Asked lately how the biggest calls actually get made, though, he put the split at 70 per cent judgment, 30 per cent data. Algorithms, Sarandos tells his staff, only describe yesterday.
Such an admission would have sunk careers not long ago. Everyone in business has spent a decade telling everyone else how ‘data-driven’ they are; ‘gut feelling’ has come to sound like a euphemism for idleness. Money has kept pace. Gartner, a research firm, reckons global IT spending will reach 6.31 trillion US dollars this year, 13.5 per cent up on 2025. Of that, 2.5 trillion US dollars will go on artificial intelligence alone (a 44 per cent leap on last year), most of it on kit that promises to turn judgment into arithmetic. Chief data officers are now standard-issue. Dashboards have taken over the corporate eye-line.
Much of this is well spent. Credit-card networks catch fraud that no human eye sees in time; logistics software routes lorries better than the most grizzled dispatcher; retailers have wrung billions from A/B testing on pastel buttons and repositioned checkboxes. Wherever the past reliably predicts the future, machines outperform humans. Chess settled that one a quarter-century ago. Most business, alas, is not chess.
Christmas spoils the model
The drug class reshaping the American grocery basket was on nobody’s forecasting radar three years ago. GLP-1 medicines (Ozempic, Wegovy and their kin) have moved fast. Households that have put someone on one have cut their food spending by around six per cent within six months; savoury snacks are down 10 per cent, with sweets and baked goods taking similar hits. JP Morgan reckons the number of Americans on the drugs will triple, from 10 million this year to over 30 million by 2030. Food giants are hurrying out products built for the shrunken appetite (Nestlé’s first new American brand in nearly three decades, Conagra’s ‘GLP-1 friendly’ labels on frozen meals), but the pivot is catch-up, not prescience. Dirk Van de Put, chief executive of Oreo-maker Mondelez, insists the impact on his volumes will be “minimal”. A consumer-goods boss who dismisses a health trend with confidence has, historically, a thin record.
The data is not wrong, exactly. It is merely backward-looking, and therefore mute on any future that does not resemble the past. Nassim Nicholas Taleb put it bluntly: the turkey fed every morning for a thousand days has excellent time-series evidence that the farmer adores him. Christmas spoils the model.
Dashboards fall foul of Goodhart’s Law, too. Once a measure becomes a target, it stops being a good measure. The call-centre agent scored on handling time rushes grandmothers off the line. The sales team paid on quarterly bookings hoards deals until January. The teacher whose pay hinges on test scores teaches to the test. Data makes it easier than ever to manage what can be counted at the expense of what cannot. Anyone who has worked under a KPI sadist knows the shape of it.
Which is where instinct creeps back in. Gary Klein, an American research psychologist, spent years studying firefighters, critical-care nurses and battlefield commanders who routinely make sound calls with patchy information and no time to consult a spreadsheet. What looks like mystical gut feel, he argued, is usually pattern recognition at speed (compressed expertise, built up over thousands of cases). The hunch is a library.
More than a spreadsheet
Daniel Kahneman, the late Nobel laureate famously sceptical of intuition, spent six years arguing the point with Klein. Their 2009 paper, tellingly titled A Failure to Disagree, concluded that expert intuition works, but only within limits. It flourishes in domains that offer quick, reliable feedback (chess, meteorology, anaesthesiology) and gets stuck in those that do not (stock-picking, long-range political forecasting). Most management decisions sit between those poles (product launches, senior hires, whether to spend 100 million US dollars on a political thriller), and there, the spreadsheet alone will not do.
The savvier operators have worked this out. Jeff Bezos harps on about it. Amazon Prime, now one of the most profitable subscriptions ever devised, went ahead despite, by Bezos’s own telling, “not a single financially savvy person” supporting it. Every spreadsheet forecast disaster. The rule he has drawn from the experience, since borrowed across Silicon Valley, is that any decision waiting for 90 per cent of the information you want is already too slow. Some 70 per cent is usually plenty.
And then there is the AI irony. As machines swallow more of the analytical grind (contract review, demand forecasting, customer-service boilerplate), the prized skills thin down to what the machines cannot do. Judgment. Taste. Knowing when the dashboard has lost the plot. Computers are excellent at answering questions. Working out which question is worth asking in the first place is rather less their thing.
The typewriter is long gone. The hunch, it turns out, isn’t going anywhere. Whatever the dashboards say.
Photo: Dreamstime.
