cx-arrival-pattern-analysislisted

# Arrival pattern analysis Staffing models need two inputs: how many contacts, and **how they arrive**. Forecasting gets the first; this skill gets the second. **The common failure is assuming Poisson arrivals because the textbook does.** Poisson means variance equals the mean — arrivals are independent and evenly random. Support queues routinely violate this: marketing sends, ticket system batch imports, outage piling, and retry behaviour create **bursts** and **correlation** that make Erlang C precise and wrong. Analyse the pattern before trusting any closed-form staffing number. ## What you are testing For each channel and interval length you schedule to (usually 15 or 30 minutes): 1. **Distribution shape** — mean, variance, coefficient of variation (CV = σ/μ). 2. **Independence across intervals** — does a hot interval predict the next one? 3. **Censoring** — are "arrivals" only contacts that waited, not those that abandoned or bounced? 4. **Batch structure** — discrete dumps vs steady drip. Record findings per queue and per interval. Patterns differ by channel; averaging hides the violation that breaks your model. ## Poisson and Erlang: when they apply **Poisson arrivals** are a reasonable working assumption when: - Contacts arrive from many independent customers without a shared trigger. - CV is near 1 (variance roughly equals mean) at your scheduling granularity. - No systematic batch import or campaign aligns to the clock. **Erlang C** (built on Poisson) i