should we switch customer support to AI-first?

Do not switch support to AI-first yet. Run a four-week assisted-drafting pilot for billing and account questions, where evidence is strongest and escalation risk is bounded.1

Evidence reviewed

EvidenceWhat it saysLimitation
2,400 recent tickets46% are repetitive billing or account questions1One quarter of seasonality
Agent shadow studyDrafting consumes 38% of handling time2Six agents, two weeks
Vendor benchmark22-31% handling-time reduction3Vendor-selected customers
Escalation auditSecurity and cancellation errors have high downside4Low frequency, high severity

Recommendation

Use AI to draft replies for two queues: billing explanations and account administration. A human reviews every draft before sending. Retrieval must be limited to approved help-center and account-policy sources; the model should not infer refunds, credits, or security status.

Keep these queues out of the pilot:

Decision criteria

MetricBaselinePilot gateWhy
Median handling time11.4 minat least 20% lowerMeasures operating value
QA score92%no more than 1 point lowerPrevents speed from masking quality loss
Repeat contact in 7 days14%no increaseDetects superficially correct replies
Unsafe suggestion ratenot trackedbelow 0.5%Hard safety boundary

Four-week pilot

WeekWorkExit condition
1Build retrieval set and 100-ticket offline evaluationNo critical policy errors
2Shadow mode with five agentsReviewers agree on quality rubric
3Human-reviewed live drafts at 25% of eligible volumeUnsafe suggestions below 0.5%
4Expand to 50% and measure repeat contactAll decision gates reported

Open questions

The largest uncertainty is not draft quality in common cases; it is whether agents become less attentive when most drafts are acceptable. Track edit distance, skipped review time, and reviewer disagreement to detect automation complacency.2