A signal is an observable phenomenon of goal attainment
Aliases: GSM signal · observable phenomenon · signal layer
What it is
In Goals–Signals–Metrics, a signal is a phenomenon the world can see when the goal is met, not yet a counting rule. The goal says “the new editor becomes a daily writing tool.” A signal might be “the author finishes a complete draft in it on consecutive days and opens it again unprompted.” “Seven-day opens per person” is a later metric chosen for that signal. The signal answers what success looks like; the metric answers how to turn that look into a comparable number. Fold the two and teams jump from goal to a convenient figure, skipping a phenomenon description that could be contradicted.
Why it happens
Goals are normative; phenomena are empirical. The middle layer forces a sentence of the form: if the goal were truly met, what would someone who has never seen the dashboard observe on site. That description can be behavioral (reopening, exporting a file, sending a link to a colleague) or an observable attitude (unprompted recommendation, positive wording in a session). Signals are more stable than metrics: swap an event definition or a survey item and the signal can still be “a complete draft was finished”; what changes is the measure. Without a signal layer, metric disputes impersonate goal disputes—one person wants opens, another wants finished drafts—when they may already agree on the phenomenon and disagree only on the count. Writing the signal is what lets a candidate metric be judged as capturing that phenomenon, or as capturing something nearby that is easier to collect.
Studying it
For each goal, collect two or three non-redundant phenomenon descriptions and ask colleagues outside the design to judge from the description alone whether attainment occurred; record disagreement. Map live metrics back onto phenomena; a metric that will not map is not a measure of that goal. Field or diary work can check whether the phenomenon actually appears among successful users and is systematically absent among unsuccessful ones. One phenomenon can have several metrics, and one metric can be mis-bound to several phenomena; the mapping must be an explicit many-to-many, not a default one-to-one.
Where it stops holding
Some goals have phenomena that cannot be watched directly: trust, fairness, long-horizon skill. Signals then have to be indirect, and that indirectness belongs in the limitation; the indirect phenomenon must not be renamed as the goal. A purely internal feeling that never enters any observable expression is not yet a signal. More signals are not better: stacking phenomena that cannot contradict one another is the same as having none. A one-off demo or a sales script that “looks like success” is not a signal of goal attainment.
Applying it
- For each goal, write the signal as a sentence of what can be seen on site; ban percentages and means at this step.
- List at least two candidate metrics for one signal, and say which part of the phenomenon each captures.
- In review, read the signal before the number: if the number rose but the phenomenon does not match, mark the metric as failed, not the goal as met.
- When swapping events or survey items, confirm the signal still holds before changing the counting rule.