What if leadership can't name the decision yet?

Then the first deliverable is helping them frame it. A good engagement often starts by turning a vague 'understand the space' into two or three decisions actually worth making. That framing is half the value, and it's the half a data subscription can't give you.

Isn't some baseline monitoring still useful?

Yes, a light standing radar has its place, but only if each signal it raises is triaged against a real question. Monitoring with no decision to feed is just an alert stream nobody reads.

Technology intelligence starts with a decision, not a database

The backwards default

It usually goes like this. Someone buys a patent-analytics tool or a data subscription, runs a broad "technology landscape" of a field, and produces a sixty-slide deck of clusters, heatmaps and trend arrows. It looks impressive. It is never opened again.

The problem isn't the data, and it isn't the tool. It's that nobody asked what decision the deck was for. A landscape with no decision behind it has no natural stopping point, no way to tell a relevant finding from an interesting one, and no format anyone will act on. It's data collection with good production values.

Start from the decision

The fix is almost embarrassingly simple: begin with the decision on the table. "Solid-state or stay with lithium-ion for the next platform?" "Is this startup's core tech defensible, or hype with a demo?" "Where should R&D place its next bet?" A real decision, with a real owner and a real deadline.

From the decision you derive the Key Intelligence Questions, the two or three things you'd genuinely need to know to decide. Only then do you choose the source mix, patents, scientific literature, corporate filings, practitioner conversations, to answer those specific questions.

This isn't just my preference; it's how the canonical model works. Cambridge's Kerr, Mortara, Phaal & Probert (2006) frame technology intelligence as a "coordinate, search, filter, scan" capability that starts from a stated information need, not from data. Lichtenthaler's (2004) study of TI in European and North American multinationals found the same pattern in the wild: the firms that got value ran TI as a decision-support process tied to strategic needs, not as a standing collection function.

Why the order matters, not just tidiness

Decision-first isn't about being neat. It changes three concrete things:

Scope. A decision bounds the search. "The landscape of AI" is infinite. "Which of these three inference approaches will be viable at our cost target within 24 months" is answerable, and finishable.
Source selection. The question tells you what you actually need: patents for who owns what and freedom-to-operate, literature for where the science is heading, practitioners for what's shipping versus what's merely published. Data-first, you collect everything and drown; question-first, you collect what earns its place.
Delivery. A decision has a shape. The person deciding wants a directional read and a recommendation, not a cluster map to interpret themselves. Start from the decision and the output almost writes itself: here's the vector, here's how confident I am, here's what I'd do.

Where AI changes this, and where it doesn't

AI has made the data-first trap cheaper, and therefore more tempting. You can now generate a plausible-looking technology landscape in an afternoon. That doesn't lower the bar; it raises the value of the one thing AI can't do for you, deciding which question is worth asking.

Use AI to move faster through the middle, search, clustering, first-pass reading of a thousand abstracts. Keep the two ends human: framing the decision at the front, and owning the judgment at the back. A model will happily produce a confident landscape of a field nobody needed mapped. It won't tell you that.

The tell

Here's a diagnostic you can run in one question. Ask whoever commissioned the "technology intelligence" what decision it will change.

If the answer is a decision, with an owner and a date, you're doing intelligence. If the answer is "we want to understand the landscape" or "keep an eye on the space," you're doing data collection, and the deck will die on schedule. The good news: it's usually fixable in the same conversation, by asking the next question, "and what would you do differently depending on what we find?"

Sources

Kerr, C. I. V., Mortara, L., Phaal, R., & Probert, D. (2006). A conceptual model for technology intelligence. Int. J. of Technology Intelligence and Planning. DOI
Lichtenthaler, E. (2004). Technology intelligence processes in leading European and North American multinationals. R&D Management, 34. link
Brenner, M. S. (1996). Technology intelligence and technology scouting. Competitive Intelligence Review, 7, 20–27. link