Originally written in French. Translated by AI — the meaning has been preserved, not the prose.
Angle
The object of competitive intelligence is the set of criteria a target customer decides on, and the movement of that set over time. Direct competitors' releases, general-purpose tools that are enough to prevent adoption, and standards imported from other markets are three sources on that same object, and none of them covers it: the first shows what one player believed it understood about the customer without saying its reasoning, the second acts outside the radar built by product category, the third raises the threshold of acceptability without any player in the market having done anything. A setup that follows only the first declares the market stable at the precise moment the other two are shifting it, and the loss shows up as deals that are lost to nobody.
Synthesis
A survey of competitors' features is not merely poor intelligence, it is intelligence whose object has been badly chosen. What a competitor ships is the part of its work that does not carry its reasoning: neither the customer it was aiming at, nor the outcome sought, nor the alternative it was fighting. Collecting that material and drawing a list of catch-up items from it amounts to reproducing gestures whose intended result nobody in the copying team knows.
The object that makes the collection interpretable lies elsewhere: what the customer decides on. Their needs, their alternatives, their frustrations, what they judge to be good enough. This set of criteria is what competitors sometimes reveal — and it is also what they sometimes fail to see.
Once the object has shifted, the population to observe shifts with it, and that is the least intuitive result. A cross-cutting tool like Notion wins no feature comparison against a specialized product, and it is enough for the comparison never to take place; its victory takes the form of an absent project, hence of an absent signal. Symmetrically, consumer applications raised the standard of professional tools without ever entering their market: what was accepted as normal at the office stopped being so, and AI is today replaying that movement at the level of the acceptability threshold itself — searching a long time for a piece of information, producing by hand what a piece of software could prepare. Three sources, a single object.
This object has the property of being in motion, which imposes on intelligence a temporal dimension that the comparison table does not have. A decision criterion first shifts in the form of weak signals — recurring frictions, workarounds, advanced usage in one profile — that nobody has yet designated as a subject. Reading them consists in formulating a hypothesis and testing it, not in predicting; and the organization that defers this work is not choosing the absence of risk, it is choosing the risk of arriving once the criteria have formed without it.
What remains is what makes the whole thing practicable, and it is not in the collection: positioning. With no target customer and no named alternative, a signal gathered is judged on the only criterion left — the fact that it exists — and a sound observation produces a false conclusion. That is also what separates the four levels a single meeting conflates: comparing what is visible, interpreting a movement, stating a tenable place, committing. Intelligence occupies the second; it produces neither the first nor the third, and an organization that stacks them leaves with observations where it thought it had decided.
Tensions / contradictions
Two requirements clash without being settled here. Positioning is what makes signals legible, and it is also what filters them: a team may set aside, in the name of its target, the very movement that announced that target shifting. Nothing in the setup says at what moment the frame of reference should be reopened rather than defended.
A second tension, about scope. The criterion that brings general-purpose alternatives into the field — whatever is enough to prevent adoption — has no natural bound: inaction, the budget, another internal priority answer to it too. Widened, it makes intelligence exhaustive and therefore impracticable; narrowed, it reinstates the category radar the thesis sets aside.
Questions
- Can a shift in decision criteria be observed other than after the fact, once a customer has already put it into a refusal?
- Can a team that has not yet won a customer in its target segment read that segment's criteria, or does it only read those of the customers it already has?
- Who, in an organization, keeps the trace of a weak signal long enough for a coherence to appear across several isolated occurrences?