In uncertainty marketing, mistakes become a strategy

Nobody is perfect, you know. But what is new is that not even machines achieve infallibility. That is why a strategic alliance between humans and humanoids becomes essential. This is something Anthropic believes in so much that it is dedicating its television debut in grand style to the subject. The American colossus founded four years ago and architect of the recent $13 billion mega round that brings the valuation to $183 billion has decided to launch its first advertising campaign for Claude. Contrary to the mainstream and even daring in its positioning, it takes the rhetoric of the artificial intelligence that will replace us head-on, going so far as to present the perfect model to be a companion. Not a substitute, but a co-pilot of our browsing experience. So much for thematic audiences, we go on generalist TV.

The campaign is called Keep Thinking and marks the official entry into the consumer market with a multi-million dollar investment. So the Ia should not replace the human mind, but amplify it. In the ninety-second commercial the message is made explicit: an age full of challenges can become fertile ground for those who never stop thinking. So much humanity and even imperfection in these words. A reversal of the dominant narrative. We are light years away from that daring and criticised Apple commercial launched in May 2024. Very bombastic title - Crush! - to promote the iPad Pro at the time. In the Cupertino giant's commercial, various musical instruments, books, cameras and sculptures were crushed by a press, leaving only a very thin iPad in its place. After that came the deluge, the implication read between the lines by many analysts and users, even though the intentions were otherwise: to show the hi-tech product in a small space. But in the time of predictable unpredictability, good intentions no longer count.

Back to the present day, in the phase that the Economist has called the age of uncertainty. Unstable markets, widespread wars and energy crises have made linear planning increasingly fragile. A strategy battering under the weight of contingency. But beware: for brands, the worst move is to stop communicating. Thus the phenomenon of going dark must be countered with companies that delude themselves into thinking they are protecting themselves from error, but actually lose value: data show up to13% less brand equity after six months of silence. Meanwhile, the Cmo Survey 2025 - an annual survey sponsored by Duke University, Deloitte and the American Marketing Association - reports that 44% of chief marketing officers have reduced budgets in the past year, mainly due to inflationary pressures and economic caution. Yet we have to learn to live with uncertainty and error, even of judgement.

Machines are also wrong because all indicators are altered. Marek Eliaš, a lecturer at the Department of Computing Sciences at Bocconi University, has shown that even predictive algorithms are not infallible: so even artificial intelligence gets it wrong. But rest assured. There is no need to chase predictive perfection because even when making mistakes, if the error is handled well, performance remains close to optimal. "It is important to recognise the limitations of any technology we use. Machine learning and artificial intelligence can be extremely powerful in making predictions under ordinary conditions, based on large amounts of past data. But when sudden changes occur due to climate or geopolitical factors, those same predictions can become unreliable. After all, the Ia can certainly guide us, but it does not really know the future. Because nobody knows it,' says Eliaš, who works on systems that do not seek perfection, but learn from their mistakes.

Concretely: instead of relying on a single prediction, the algorithm combines several strategies, recognises which ones work best and reduces 'regret'. Mathematical models can thus learn from mistakes. "They do not try to eliminate errors completely, but to monitor risks and adapt to them. The transition to renewable energies makes the problem clear. To maintain efficiency, algorithms are needed that can handle minute decisions - such as when to recharge the electric car or start the washing machine - under everyday conditions. But they must also be able to adapt to disruptions, whether in consumer behaviour or weather conditions. This requires algorithms that not only predict, but also assess risks, choose actions that circumvent them, and interpolate between models trained for different scenarios. My theoretical results show that in many cases this is possible,' says Eliaš.