The Seduction of the Decimal Point

The third in this week's series looking at how to avoid marketing analytics misfires and be successful in 2026. On Monday issues with AI-driven attribution were discussed, on Tuesday it was ensuring you used appropriate metrics to assess audience optimization, and today we are taking a skeptical look at outputs that are overly precise.

There is a specific kind of analytics misfire that often happens in high-stakes boardrooms. It’s an over-reliance on precision - if an AI model can provide very precise estimates, it must be right! As I was cleaning up some old files over the break I came across an old report where a consultant (not me!) promised a CMO that new customers would spend an average of $142.34 each over the next six months. Not $140. Not "somewhere between $130 and $150."

The company treated that number as gospel. They built CAC targets around it, scaled their spend, and hired staff based on it. When the company encountered tough market conditions and the actual average spend for that period came in at $125, the strategy collapsed - and the consultant was out.

Why does the "AI" give you such a precise number?

When we talk about "AI" in this context, we are talking about a predictive model. Whether it’s a simple regression or a complex neural network, the algorithm arrives at its output by finding the mathematical center of gravity for what is often a massive, messy dataset. And the sophistication of the analyst or the tool in incorporating externalities like market conditions has a significant impact, because the process is inherently backward-looking. Since it is calculating a statistical mean across potentially thousands of variables—some high, some low—the resulting 'best fit' almost never lands on a round number. That $142.34 isn't an intentionally targeted prediction; it’s just the cold, calculated middle of a noisy cloud of data.

Because math is exact, the output is exact. The AI isn't trying to be misleading in an intentional way; it is simply performing a calculation. It's a machine. If you divide 1,000 by 7, you get 142.857... The decimal points aren't a sign of insight; they are just a byproduct of the math.

The trap is thinking that mathematical precision equals real-world certainty.

The Misfire: Precision as a Proxy for Credibility

The danger of a precise number is that it can shut down critical thinking. Our minds are wired to believe that "142.34" requires more rigorous work to calculate than "140" while in reality, the model may have been trained on historical data that didn't account for shifting market conditions or competitive price pressures and that is only a general target number, not the basis for setting specific expectations.

In marketing analytics, a model output is a point estimate—a single dot in the middle of a massive cloud of probability. By presenting only the dot and not the cloud, practitioners can lead their clients astray.

The Lesson: Bet on the Range, Not the Decimals

Precision is not the same as accuracy. You can be precisely wrong (hitting the exact wrong spot every time) or vaguely right (hitting the general target area). One of the first things I have always taught analysts on my team: don't put decimal points in management presentations because you are often called on to defend the precision; better to use a reasonable range and get people nodding their heads about the logic of setting the range, because they can't question the algorithm.

I once had a client insist I provide an estimate of the response rate for a direct mail campaign. I refused. I told them I couldn't predict the exact result, or even give them an accurate range, because this was a new channel for them, and what I had seen for other clients didn't reflect the same conditions. They were frustrated—until the returns came in. Because we had planned for a range of outcomes, we were prepared for the reality. The model was in service of a different metric; the campaign was hugely profitable, even though it didn't hit a specific response rate target.

The Takeaway: If a tool or a consultant gives you a decimal point in a prediction, ask for the variance. If they can’t give you a range, they aren't helping you develop a strategy—they’re giving you a math problem. And inviting you to set unrealistic expectations.

What do you think about the level of precision from your analytic outputs? Do you agree or disagree with this perspective? Want to chat? Find a convenient time here.