Signals
Reading session gaps as a leading indicator
Identity Flowcore journal · used in week four of Cohort Signal Lab
“Days since last open” is a tempting single number. It is also a blunt instrument. A daily commuter who skips one weekday is not the same person as a weekend-only shopper who skipped a month. Predictive work on mobile starts by admitting that silence has a personal tempo.
Build a baseline per identity, not per average user
We ask students to compute, for each stable identity, the distribution of gaps over the last twelve observed sessions. A new gap is interesting when it sits far into that person’s tail, not when it crosses a global threshold like “14 days.” Global thresholds are how grocery apps spam festival travellers.
Separate calendar shocks
National holidays, app outages, and store-listing downtime create shared silence. If half your cohort went quiet on the same Tuesday, look at status pages before you look at psychology. Overlaying a simple outage and holiday calendar prevents a lot of false “at risk” flags.
Pair gaps with last-session quality
A long gap after a failed payment is a different story from a long gap after a completed reorder. Session length, error events, and whether the last screen was a blocker change the meaning of silence. None of this requires a deep network. It requires the event dictionary to tell the truth, which is why taxonomy week comes first.
What CRM should actually receive
Not a mysterious score from 0 to 1 without a sentence. A small set of states: on-tempo, stretched, shocked (calendar/outage), and unknown (identity unstable). Messaging rules attach to states. That is less glamorous than a leaderboard and much harder to misuse.
Week four of the Cohort Signal Lab is this pairing of gaps and notes. Bring a sceptical lifecycle lead if you can; they will keep the states operational.