In 2001, the then Manchester United manager Sir Alex Ferguson sold Jaap Stam. Stam was one of Europe’s best defenders, a pillar of the 1999 treble. Yet at 29 he was shipped off to Lazio. Ferguson later called it a mistake.

Accounts differ on why. Simon Kuper’s is the most intriguing for anyone who works with data: Opta statistics showed Stam’s tackle count falling. The conclusion seemed obvious. Fewer tackles, therefore decline.

The statistic was accurate. The interpretation was the problem.

A defender may tackle less because he is slower. He may also tackle less because he is reading the game earlier, closing the space before the attacker even receives the ball. A tackle is highly visible. Good positioning is often the absence of an event.

The number captured what happened. It didn’t explain why. And that’s where the trouble starts.

The comfort of what can be counted

Numbers feel solid. They slot neatly into dashboards, targets and meetings. This is quantification bias: valuing what can be counted, sometimes simply because it can be counted.

The problem rarely begins with bad data. It begins with useful data being asked to do more than it can.

Philosopher C. Thi Nguyen has a name for this: value capture. A measure starts as a proxy for something we value, then slowly takes over. Goodhart’s law is the organisational version: when a measure becomes a target, it stops being a good measure.

That’s the deepest risk. We don’t just simplify reality to manage it. We start to believe the simplification is reality.

Marketing has its own tackle-count problem

Digital gave us a dazzling amount of observable behaviour: who clicked, who converted. Next to the slow, invisible work of building a brand, it felt concrete. So we followed it.

But proximity to purchase is not responsibility for purchase. Paid search may simply collect demand that was built somewhere else.

Adidas learned this publicly. Its global media director acknowledged that a focus on ROI and last-click attribution had led to a 77% performance, 23% brand split. The dashboards made it look efficient. Then paid search was switched off in one market and traffic and revenue didn’t dip. When econometrics arrived, it indicated brand activity was driving 65% of sales.

The metrics weren’t false. People did click. The mistake was treating one visible part of the journey as the whole journey. Kantar’s marketing ROI evidence shows what that costs: baseline sales can fall by more than half within four years of consistently cutting brand investment, even while quarterly numbers look manageable.

Three signals, three different jobs

At Kantar, we’ve been exploring a simple way to make sense of the data deluge in our new paper, Three Signals. However sophisticated the source, most brand signals do one of three jobs: they tell us what people do, what they think or what they express.

Behaviours are the audit trail. Concrete, but backward-looking.

Attitudes are the compass. The forces behind future choice, often moving before sales do.

Expressions are the seismograph. They pick up every tremor but can’t say which will become an earthquake.

Stam’s tackle count was a behavioural signal. The missing piece was the why. Was he declining or anticipating better? The metric couldn’t say.

Rising sales look reassuring until attitudes show predisposition weakening. Then they start to look like momentum from the past, rather than strength for the future. The reverse trap is just as real. Nike’s Colin Kaepernick campaign triggered boycott calls and burning trainers online, yet Nike’s online sales rose by more than 30%. Coca-Cola’s partly AI-made Christmas ad was picked apart on social media yet ranked among the year’s top performers when tested with everyday people through LINK+. The loudest signal wasn’t the most representative.

One mistakes visibility for value. The other mistakes volume for prevalence. Both are failures of context.

When signals disagree

The instinct is to hunt for the one number that settles the argument. But conflicting signals usually describe different parts of the same reality. Sales up while attitudes fall? Growth may be running on momentum from the past. Attitudes up while sales stall? Demand exists, but something practical is in its way. Divergence is diagnostic: it tells you where to look next.

Not every decision needs a three-signal committee. Match the depth of evidence to the stakes. More data isn’t automatically more certainty. Sometimes it’s just more data.

The part no dashboard can do

AI can spot patterns and surface contradictions faster than any of us. But deciding what they mean here is still a judgement call. And that takes experience and nerve.

So before acting on any metric, ask:

1.       What is this actually a proxy for?
2.       What else could explain this number?
3.       Which other signal would confirm, challenge or reframe my reading?

Metrics are only as valuable as the reading behind them: knowing what each number can tell us, noticing what it leaves out and applying judgement before the right number leads us to the wrong conclusion.

Download Three Signals to build your own signal strategy or speak with someone at Kantar about making decisions with greater confidence today.

Mary Kyriakidi is global thought leader at Kantar