Support automation
The L1 team handled a high volume of requests that were genuinely repetitive: status checks, known failure classes, and standard data pulls.
PROBLEM
Skilled engineering time was being spent on retrieval, not on judgement. That capped how many accounts the team could absorb.
HYPOTHESIS
A meaningful share of L2 escalations were not escalations at all — they were L1 requests lacking a self-serve answer.
RESEARCH
Classified escalated tickets by whether resolution required new judgement or only known information, and measured the split.
DATA USED
Escalation reasons · resolution notes · time-to-first-response · repeat-request frequency
DECISION MAKING
Chose to automate retrieval and leave judgement with humans — the split the data supported, and the only split enterprise security would accept.
SOLUTION
Runbook standardisation, self-serve answers for the top repeat requests, and ultimately the assist-first model behind AI Marine.
BUSINESS OUTCOME
Escalation mix shifted toward genuinely novel problems, which is what an L2 team should actually be spending its time on.
REFLECTION
Automation is a prioritisation exercise disguised as an engineering one. The data told us what not to automate.
SQL INVESTIGATION
SELECT resolution_type,
count(*) AS n,
round(avg(first_response_minutes), 1) AS avg_frt
FROM escalations
GROUP BY resolution_type
ORDER BY n DESC;