How to Improve Weak Signal Judgment

Everyone collects weak signals now. Most of what they collect predicts nothing. A weak signal isn't a thing you spot, it's a prediction you make, and the edge goes to whoever bets on it while being wrong is still cheap.

So how do you become the one placing the bet, not the one still collecting reports? Let's get into it.

A weak signal is a faint piece of evidence that points to something a customer will want before they can name it, and before the market has priced it in. Faint, because if it were loud, everyone would already be acting on it. Deniable, because you can always explain it away as noise, and most people do. That deniability is the whole point. The moment it becomes undeniable, the advantage is gone and the price has moved.

Ten years ago, noticing was hard. You needed sources, a network, time to read widely, a feel for the edges of your industry. That was the moat. It isn't anymore. Every team has a trend report and three newsletters and an AI tool surfacing emerging behaviors on a schedule. The noticing got automated. What didn't get automated is the judgment about which signal predicts a structural change and which points to nothing real, and the nerve to act early.

I sat on Roche's diagnostics innovation board, the only outsider in the room, helping decide which ideas got funded. At one point we took on diabetes care.

I am not diabetic. So I had Roche ship me every meter and test strip they made, and I pricked my finger up to a dozen times a day to feel what their customers felt. You cannot innovate for a customer whose day you have never lived. Skip that, and everything after is a guess.

Roche was a leader in blood glucose testing with its Accu-Chek meters, and the math looked obvious. Someone with type 1 diabetes tests around eight times a day, every day, for life. A big, stable business. Type 2 was the smaller story per patient. Those patients tested once, maybe twice a day, so each one looked worth less, and we filed the category under "less interesting." We could already see type 2 climbing. We weighed it against the per-patient math and explained it away.

Then type 2 diagnoses exploded into one of the fastest-growing chronic conditions in the world. And the category stopped being about counting tests per day at all, because monitoring went continuous, the always-on sensors people wear today. We had seen the early edge of both shifts. We even predicted them. We just didn't move fast enough, and the reason is the one that kills most weak signals inside a big company. Project approval and annual budgets are built to fund what's already proven, not to chase something still faint.

Roche got there. Accu-Chek SmartGuide, its real-time continuous monitor, is on the market now. I just wish we had moved the moment we saw it, instead of waiting for the next budget cycle to make it safe.

We didn't miss the type 2 signal for lack of noticing. We noticed. We missed it on the three things that come after, and those you can train. The moves start once you've got a signal you can't quite dismiss, and the skill is what you do with it.

A canary in a coal mine matters because the air changed. It signals something structural, a shift in the environment that affects everyone in it, whether they've noticed yet or not. A costume is the opposite. A few people put it on, it's striking, it spreads for a season, then they take it off and the room is exactly as it was. On day one the two look identical. A behavior appears, it's unusual, it's spreading. The only question that matters is whether it predicts a change a customer can't reverse, or a moment that will pass.

Back in 2018 I wrote about telling a trend from a fad, and the test still holds: ask what need the behavior reveals. Type 2 was a canary, and we read it as a costume, because we counted testing frequency instead of the need underneath it. That need, millions of people learning to manage a lifestyle disease, only grew.

The discipline is refusing to let the size of the spike tell you which one you're looking at. Costumes spike too, sometimes higher. You're reading for the need, not the noise.

A signal's window is short. Too early, you can't tell it from noise and you waste resources chasing ghosts. Too late, it's obvious, everyone sees it, and the advantage is already priced in. The value lives in the narrow gap between. Waiting for more evidence feels like better judgment, but the evidence that finally convinces you has already reached your competitors. Certainty and advantage move in opposite directions, so by the time you're sure, sure is just another word for too late. The question isn't whether the signal is real yet. It's how much longer you can be the only one taking it seriously.

This is the move that separates the people who read signals from the people who collect them, and almost nobody is willing to make it. A signal you predict but never act on is still just watching. Ideas without execution are a hobby, and I'm not in the hobby business.

The whole value of an early signal is that you move before it's confirmed. Wait for proof and you've waited too long. So you act on thin evidence. And thin evidence is wrong a lot, which means you will be wrong a lot. People hear that and freeze, because they picture the cost of being wrong as the failed product, the wasted year, the budget burned on a guess.

People call that caution. It isn't. The skill is structuring the bet so that being wrong is cheap. You don't commit a product line to a deniable signal. You commit a prototype. A landing page. One conversation with ten customers. A two-week test that costs you a sprint and buys you information you can't get any other way. Being early and wrong should cost you a week. Being early and right should put you a year ahead.

You're not betting on being right. You're buying the option to be right, cheap enough that being wrong doesn't hurt, and you scale up only as the signal firms up.

That's why the noticing crowd never gets here. Noticing carries no risk, so it never builds the muscle for cheap commitment. They watch, they report, they wait for certainty, and they call it foresight. It's the safe choice, and it's worth nothing.

Pick one behavior you've been dismissing as noise. Something you've seen more than once, in your customers, your kids, your own habits, that you waved off because it looked too small or too strange to matter. Then run it through the three moves.

1. Canary or costume. What need does the behavior reveal? A need the person can't go back from, or a novelty they'll set down in a season? Write the answer in one sentence. If you can't, you don't understand the signal yet.

2. Find the window. How much longer does this stay deniable? Who else is likely seeing it? If the honest answer is "it already feels obvious," pick a different signal. You're late on this one.

3. Design the cheap bet. What's the smallest thing you could do this month to test whether you're right, where being wrong costs a week and being right puts you ahead? Name the bet. Name the cost. Name what you'd learn.

Do this with one real signal and you'll feel the difference between collecting signals and using them. Collecting is comfortable. Using one costs you a decision.

If you want a sparring partner for that, I built one. From Signal to Bet is a set of AI prompts that run a signal through these same three moves and argue with your read at each one. It's free at innovation.tools. The exercise teaches you the moves. The prompts make you defend them.

The signal was always there, for you and for everyone reading the same reports you read. The edge was never in seeing it. It was in what you were willing to do before it was safe to do anything at all. Get good at that, and you stop reacting to the future and start arriving early.