Part III · Strategic Leadership in the AI Era

Horizon Scanning

Horizon Scanning is a standing system for detecting weak signals before they become material. The book runs it in five stages, and its objective is future foresight built as an operating process rather than an annual offsite.

What it diagnoses

An organization that sees the disruption in hindsight and can show you the email where someone mentioned it.

Horizon Scanning is a standing system for detecting weak signals before they become material. The book runs it in five stages, and its objective is future foresight built as an operating process rather than an annual offsite.

What it diagnoses is where foresight actually breaks, which is almost never collection. Most organizations are already receiving the signal. Somebody in sales heard it, somebody in procurement saw the lead time move, somebody forwarded the article. The failure is downstream: no domain definition, so everything is potentially relevant and therefore nothing is. No significance test, so the signal arrives as interesting rather than actionable. No owner, so nothing tracks it between the week it appeared and the quarter it mattered.

The second thing it diagnoses is where your bias is worst. The cognitive difficulty is highest at the periphery, precisely because there is less data there and more room for distortion. Your interpretation is least reliable exactly where the framework is most valuable.

How to run it

The book runs five stages on distinct cadences.

1. 1

   Stage 1, define your scanning domain (setup, then annual review). Name the specific territories you are scanning: technologies, regulatory movements, input costs, adjacent industries, customer behavior shifts. An undefined domain produces a feed. A defined domain produces a search. Review it annually, because the domain that mattered last year is a lagging definition.

2. 2

   Stage 2, collect weak signals (ongoing). Build the intake from multiple sources and make contribution possible from people who are not in strategy. The signal usually arrives through someone operational, and if there is no route for them to put it in, it stays in their inbox.

3. 3

   Stage 3, analyze signals for strategic significance (weekly and monthly). This is the stage that separates the system from the newsletter. Each signal gets tested against a standing question: if this continues, what breaks, and for whom. Most signals fail that test and should be dropped without ceremony.

4. 4

   Stage 4, identify and track emerging issues (monthly). Signals that survive stage 3 become tracked issues with an owner and a review date. Tracking is what converts a weak signal into an early warning, because a signal seen once is an anecdote and a signal seen three times on a trend line is a decision.

5. 5

   Stage 5, integrate into strategic decision-making (quarterly and as needed). Scanning is wasted effort if it does not change a decision. Once a quarter, a half-day leadership session takes the three to five most significant tracked issues, explores how each could play out, and gives each one a response: monitor, prepare, experiment, or pivot. Then it decides what to fund.

How operators misuse it

They collect and never triage. This is the dominant failure and it looks like diligence. A weekly digest of interesting developments, circulated widely, acted on never. Volume without a significance test produces reading, not foresight.

They scan for confirmation. An undefined domain lets you find evidence for the strategy you already have. Defining the domain in advance, and defining what would count as a threat to the current plan, is what makes the scan capable of returning bad news.

They treat it as a strategy function. Scanning owned entirely by a central team loses the operational periphery, which is where most weak signals actually enter the business. The analysis can be central. The intake cannot.

They stop at detection. Seeing it early buys nothing on its own. Every tracked issue has to reach stage 5 and leave with a decision or a monitored trigger, or the organization will have documented its own surprise in advance.

What the pattern shows

The research is consistent that the constraint is interpretation, not detection, and that technology has moved the bottleneck rather than removed it.

MIT Sloan Management Review’s work on effective horizon scanning locates the difficulty at the periphery: the cognitive challenge there is far greater than in focal areas, because there is less data to work with and correspondingly more room for bias and distortion. That is the structural reason scanning fails quietly. The organization is not missing the signal, it is misreading it, and misreading is invisible until the outcome arrives.

McKinsey’s work on weak signals makes the case that the signals that matter are available well before they are material, which reframes the whole exercise. If availability is not the constraint, then investment in more collection is investment in the wrong place.

BCG’s strategic foresight research adds the part most relevant to anyone deploying AI at this. Their position is that advanced technology has unprecedented capacity to detect weak signals, and that foresight only becomes effective when that capacity is combined with human creativity, judgment and imagination. In other words the machine has substantially solved stage 2, which moves the entire competitive difference into stage 3.

Read together, the pattern is specific. Collection is commoditized, interpretation is not, the periphery is where interpretation is weakest, and the organizations that get value here are the ones that built a significance test rather than a better feed.

Where it lives in the book

Framework 16, Chapter 5, page 290. The Future Foresight System. It closes Chapter 5, which is the correct sequence: VRIO establishes the advantage you hold, the Constellation orients attention to sustain it, and Horizon Scanning asks what is coming for it.

Questions operators ask

What is a weak signal?

An early, ambiguous indicator that something material may be forming. Weak refers to the clarity of the signal, not the size of the consequence.

How is this different from competitive intelligence?

Competitive intelligence watches known competitors. Horizon scanning watches the periphery, including the developments that will produce competitors you do not have yet.

How much time does it take?

The book runs a weekly analysis rhythm, monthly tracking and a quarterly half-day integration session. Roughly an hour a week plus the monthly and quarterly reviews. Anything less than weekly and the trend line never forms.

Can AI run this?

It can run most of stage 2 and help structure stage 3, and the foresight research is direct that detection without human judgment does not produce foresight. Automating collection and skipping interpretation gives you a faster version of the failure.

We already circulate an industry digest. Is that horizon scanning?

Only if signals in it get tested for significance, assigned an owner, and tracked. Without those three, it is a reading list.

Sources

1. Scott D. Anthony, "Four Keys to Effective Horizon Scanning," MIT Sloan Management Review, 2025
2. Boston Consulting Group, "What Companies that Excel at Strategic Foresight Do Differently," 2026