Five Payroll Agents, 900 Inquiries, and a Four-Hour Statutory Clock
A payroll COE has five agents handling about 900 inquiries a month. Roughly 70% are status questions: "where is my payslip," "when is the bonus paid." All work lands in one assignment group and agents pick from a shared list. Median first response is 2.1 days, 45% of cases are reassigned at least once, three of the five agents handle 80% of the volume, and CSAT sits at 3.4 out of 5. No agent has any skill recorded. Two agents are US payroll certified and one is EMEA certified. A new statutory pay-shortage process must be answered within 4 business hours, with evidence of response time retained for audit.
1. Which portion of the 900 should stop being cases, and how do you remove them without hiding real problems?
Read the worked answer
Target the 70% status queries - roughly 630 records a month - because they consume agent time without needing judgement. Remove them through three layers:
- Self-service supply - a payslip access route in the portal and a pay-calendar article that answers the recurring date questions. Without this, deflection just moves the question to a different channel.
- Deflection at intake - knowledge suggestions and a virtual-agent topic before the case is submitted.
- An auto-response/auto-close path on a narrowly scoped "pay inquiry" type**, not on the payroll COE's whole intake.
The guardrail that matters: the questions that look like status queries but are not - missing payslip, wrong amount, incorrect tax - must land on a different HR service so they can never be swept into auto-close. Then measure the guardrail, not just the deflection: reopen rate, "employee replied after closure" count, and a weekly manual sample of 20 auto-closed records. If reopen rate climbs while volume falls, you have hidden work, not removed it.
Also set the expectation that 630 is a ceiling, not a forecast. Realistic first-quarter deflection is well below it.
2. Design routing for the remaining volume plus the new four-hour statutory process. What do you configure, and in what order?
Read the worked answer
Order matters - data before automation.
- Record skills on all five agents (US payroll, EMEA payroll) and identify who can cover statutory work. Routing built on absent skill data is just a queue rename.
- Split the intake into distinct HR services - payslip exception, pay inquiry, statutory pay shortage - each with its own assignment group and template. The split is what allows different SLAs and different capacity treatment.
- Advanced Work Assignment on the groups: capacity weights so the high-volume simple work cannot consume an agent's whole day, and skills-based matching so EMEA work does not sit with US-only agents. The 80/20 concentration is a symptom of pull-based picking; capacity weighting plus push assignment is the fix.
- Statutory service configuration: own assignment group, default priority, 4-business-hour SLA on a schedule that actually reflects business hours, escalation before breach, and audit-friendly timestamps retained. If it must never queue behind 600 payslip queries, give it a dedicated group rather than sharing a queue with capacity weighting.
- Then tune - measure, adjust weights.
Do not route by keyword or subject line. The statutory process is defined by which service was chosen, so make the employee-facing choice unambiguous and, where possible, drive statutory intake through a channel that cannot be misclassified.
3. Which metrics would tell you within 30 days whether reassignment fell for the right reason?
Read the worked answer
Track a small set that can be read together - any one alone is gameable:
- Reassignment rate, broken down by service - the headline, but only meaningful split by service. A blended number hides a service that got worse.
- First-assignment accuracy - share of cases that never left their initial group. This is the one that actually moved.
- Time to first assignment - reassignment can fall simply because cases sit unassigned longer, which is not an improvement.
- AWA offer rejections and expiries - high rejections mean skills or capacity weights are misconfigured.
- Workload distribution - top-agent share, to confirm the 80/20 concentration is actually easing.
- Reopen rate and CSAT - the harm check. Faster closure with rising reopens is a regression dressed as a win.
- Unassigned aging / WIP - capacity weighting can starve low-weight work if the weights are too aggressive.
State the hypothesis before the change: "reassignment falls because cases start in the right group, with time-to-first-assignment flat or better, and reopens flat." If any of those three conditions fails, the number moved for the wrong reason and you should not report it as success.