Put the support math to work.

Convert a monthly asynchronous workload into a practical FTE floor, accounting for shrinkage and sustainable occupancy.

Formula: contacts × handle time ÷ productive hours. This is a workload model, not a queueing model; use Erlang C below for synchronous demand. Compare assumptions with the benchmark library.

Estimate the agents required to answer a synchronous voice or chat queue within a service-level target, without assuming abandonment.

Erlang C assumes random arrivals, no abandonment, identical work, and a single pooled queue. Real schedules need shrinkage, concurrency rules, skills, and interval-level arrival data. See the occupancy definition.

Build an honest fully loaded unit cost instead of dividing payroll alone by ticket volume.

Formula: labor + tooling + allocated overhead ÷ resolved contacts. Pair it with quality and repeat-contact measures; a cheaper unresolved contact is not a saving. Start with the metric definition.

Estimate the random sample needed to measure a proportion at a chosen confidence and margin of error.

Uses the conservative 50% response assumption and finite-population correction. Stratify high-risk and rare contacts separately. Download a starting rubric from the template library.

See the uncertainty hidden behind a satisfaction percentage using a 95% Wilson confidence interval.

The interval describes random sampling error, not whether respondents represent all customers. Always report response rate and segment mix next to CSAT. See the CSAT benchmarks.

Separate eligibility from confirmed automation, include inference and platform costs, and see the monthly payback before calling deflection ROI.

Confirmed resolution means the customer did not need a human or repeat contact—not a bot session that ended. Run low/base/high scenarios and add quality-failure costs before approving spend. Use the AI pilot readiness checklist first.