The problem
A few kinds of escalation always went to the same three people. When one of them was on leave, those tickets waited. When all three were busy, the whole queue slowed down.
How I measured it
# escalations by who resolved them
awk -F, 'NR>1 && $3=="escalated" {r[$7]++} END {for (k in r) print r[k], k}' tickets.csv | sort -rn | head
# hours waiting in queue for escalations, split by whether a top-3 person was the resolver
awk -F, 'NR>1 && $3=="escalated" {top=($7=="ana"||$7=="raj"||$7=="lee"); w[top]+=$8; c[top]++}
END {print "top-3 avg wait:", w[1]/c[1], "others:", w[0]/c[0]}' tickets.csv
Most escalations of those kinds were resolved by the same few names, and they waited longest.
Why it kept happening
It's quicker to hand a hard ticket to the person who's solved it before than to teach someone else. Every single time that's the right call, and over a year it builds a bottleneck.
The process change
A brown-bag session every other week, an hour over lunch. Each one took a real escalation from the past month (anonymized) and walked through it: the symptoms, what we checked, what we got wrong first, the fix. Whoever had solved it presented, and a junior specialist co-presented the next time that kind of issue came up.
Sessions were recorded and the notes went into a shared folder, organized by symptom, not by date, so people could find them when they had the problem.
What I learnt
The experts weren't hoarding anything. There just wasn't a time set aside for passing it on. A fixed slot on the calendar did more than asking people to "share knowledge".
What I do now
- Teach from real cases. People remember a story better than a procedure.
- Organize the notes by symptom, the way someone will search for them.
- Measure who resolves what. A short list of names is a risk.
stage 3 includes leaving the knowledge behind → see the engagements