What we asked, how we ran it, what came back
Can browsing behavior show cable-to-satellite migration pressure while a quarter is still running — weeks before operators publish subscriber losses?
Scored every panel user on a 0–5 migration conviction score from four behavioral signals, then compared quarterly panel pressure against reported operator losses across two years.
Panel pressure moved with reported losses in four of five clean quarters. The signal is coincident, not leading — but it is readable weeks before the print.
Six calls written down on July 9. The operators reported on July 23–24.
Every call was frozen before the Q2 results were published, then graded under a fixed HIT / PARTIAL / MISS rubric — including the one that missed.
We wrote down what we expected before the companies reported. Four of six were right, one was half right, one was wrong — and we show all six.
Full-session browsing → one 0–5 migration conviction score
How far into the provider's own funnel the session actually went — not whether the domain was merely visited.
How many competing providers were evaluated in the same behavioral window.
Whether the session touched cancellation or billing paths at the current provider.
Whether the behavior repeated across weeks, separating intent from a single curious visit.
Validation: top-tier scored users reach the order path at roughly 4.2× the rate of the lowest tier, and that separation held under three separate reweightings of the score.
The alignment is directional evidence, not quantitative accuracy. The sample is small and the result is underpowered — stated, not buried.
Operators attribute their losses to fixed wireless, fiber overlap and mobile substitution. We measure migration pressure toward satellite; the prints do not separate satellite-driven churn, and neither do we.
The score identifies who is in a migration state now. Forward ranking was tested and does not hold. The full brief publishes an 11-item limitations register.
Starlink Migration Intelligence
The complete 12-page brief: the full grading table, the validation series across two years, the scoring methodology, the second-order geographic reads, and the published limitations register.
- ✓ Six frozen calls with reported outcomes
- ✓ Eight-quarter validation against reported losses
- ✓ Scoring methodology and reweighting tests
- ✓ Full limitations register and two self-corrections
Want the monthly issue?
Standing series with alert thresholds, frozen calls, and public grading — refreshed monthly.
Behavioral signals, not verified migration · Coincident, not leading · Monitoring expressions, not investment recommendations