Skills CX operations
Detecting contact spikes
Support volume is one of the more strongly seasonal series in a business, and almost every naive spike detector fails on the same thing: it compares a Monday to a Sunday and declares an incident.
npx rulebase-skills install cx-contact-spike-detectionWhen to use it
Reach for this when someone says any of these — they are the phrases the skill itself triggers on:
- “are complaints about X rising”
- “detect when an issue spikes”
- “was that a real increase”
- “alert us to unusual contact volume”
How it works
The method, in the order the skill runs it. The full procedure — tables, worked examples and the edge cases — is in the skill itself.
Model the seasonality, or the detector is noise
At minimum, support traffic has: Business-driven cycles, Trend, from growth in the customer base.
Rate and count, together
A topic's share of contacts and its absolute count answer different questions and fail in opposite directions.
Set the threshold on the historical distribution
Do not pick a round multiple. "Double the baseline" is arbitrary and will be far too sensitive for low-volume topics and far too insensitive for high-volume ones.
Multiplicity: the trap nobody accounts for
Monitoring 50 topics at a 1-in-100 threshold produces roughly one false alarm every other period, every period, forever. Then a fortnight later the channel is muted.
Confirm before escalating
A statistical spike is a hypothesis. Before anyone treats it as an incident: Rule out the boring causes, in this order.
Onset detection matters more than magnitude
For incident use, the useful output is when it started, not how big it got. Report the first period that deviated, the trajectory, and whether it is still rising — alongside the earliest few conversation ids, which are what an incident responder actually wants.
Related skills
Free and open source, and vendor-neutral — it reads the conversations from whichever helpdesk you already run. Browse all 149 skills · connect your helpdesk over MCP · source on GitHub
