AlgoMags
Why marketing still can’t predict exactly
Essays · November 14, 2026 · 8 min read
Marketing has more data than ever, more powerful models and real-time feedback. It still cannot say with certainty what any one person will do next. The gap is not a temporary engineering problem. It reflects something about how people work.
Prediction works in aggregate
Marketing models are good at populations. They can estimate how many of ten thousand visitors will convert, which segment responds best to a message, and how demand will shift with seasonality. These estimates are reliable enough to run businesses on.
For individuals, the same models produce probabilities, not answers. A person with a 30 percent predicted likelihood to buy is still either buying or not, and the model cannot tell which.
Why individuals resist prediction
Several forces keep individual behavior uncertain:
- Context the data cannot see: a conversation at dinner, a bad day, a budget change
- Changing preferences, which shift with life events faster than data updates
- Reflexivity: people change behavior when they sense they are being targeted
- Noise in the signals themselves, from shared devices to accidental clicks
Research shows digital traces can predict personality traits with surprising accuracy. Traits shape tendencies. They do not determine choices in a given moment.
Quantified psychology and its limits
The hope behind much of modern marketing technology is that psychology will eventually be quantified closely enough to predict decisions directly. Progress has been real: personality inference, emotion detection, behavioral modeling. But each advance improves the odds rather than eliminating uncertainty.
Even if measurement became perfect, a deeper limit would remain. People respond to the systems that try to predict them. A market where every message is precisely targeted produces buyers who discount targeted messages. Prediction changes the thing being predicted.
What follows for practice
The practical answer is to design systems that work under uncertainty: test instead of assume, measure ranges instead of points, keep feedback loops short, and treat every forecast as a hypothesis. Marketing that admits what it cannot know tends to outperform marketing that pretends it can.