Workload and forecasts
Connect historical arrivals with time-slot forecasts, validation and staffing coverage assumptions.
Slide the divider to compare the same workspace.
DarkThe right people. At the right time.
Open a queue to bring its waiting pressure, historical arrivals, forecasts and coverage into the same view. Then test a staffing choice before the next busy period.
1. Choose a queue
Click its name in the current-pressure table. The selected period and analysis settings stay with you.
2. Read its demand
See the queue’s waiting trend, historical load and expected arrivals in the chosen intervals.
3. Try the coverage
Set service targets and staffing. Compare the time slots and saved scenarios before choosing how to organise the team.
What changes if demand rises?
Choose an example queue, then change arrivals, conversation time or scheduled staff. The calculation updates immediately using the same Erlang C model as Obelisk.
Interactive example · illustrative dataExample settings: answer 80% within 20 seconds; maximum occupancy 85%; after-call work 30 seconds; unavailability 20%. In Obelisk you can change all these assumptions.
This staffing meets both the service target and the occupancy limit in the example.
There is room to improve coverage. Try the calculated staffing and compare the estimated wait.
Available staffing cannot sustain stable service for this demand. Try the calculated staffing.
How to read the simulation
Erlang C relates arrivals, handling time and agents available to estimate waiting and staffing. It models a homogeneous team in stable intervals, with random arrivals and service times and customers remaining in the queue. Use the result as a planning estimate alongside your real queue context.
Net availability accounts for unavailability; occupancy concerns conversation plus after-call work. Obelisk supports saved scenarios, variable time bands and comparison against the same traffic, so you can evaluate different choices on a consistent basis.
Know what to look for.
Choose the scope of the decision
Select the historical period, section and queue, then choose the forecast history and 15-, 30- or 60-minute intervals. The history you inspect and the completed days used to train the forecast each have their own controls.
Open a queue. Focus the whole analysis.
Compare waiting calls and available, busy, paused and DND agents, with the time of the latest sample. Click a queue name to open its waiting trend, historical load, expected arrivals and dedicated coverage simulation with the selected context retained.
Read the demand that actually arrived
Daily queue-entry events show total arrivals, observed days and the average per day. Use the pattern to recognise peaks and quieter days; the date of the latest event helps you assess how current the evidence is.
Look seven days ahead, down to each interval
The local model uses the median of at least four comparable weekdays. Read expected daily arrivals, the minimum and maximum already observed and the comparable-day count, then select a day to inspect its interval profile and prepare the coverage scenario.
Check how the forecast performed on history
Compare the model’s mean absolute error with a simple historical baseline and read how many days were evaluated. This gives you evidence for deciding how much weight to give the forecast when planning.
Find recurring waiting pressure
The day-and-hour heatmap shows average waiting calls per queue sample. Read it alongside sample coverage and peaks to locate recurring pressure and decide which time bands deserve a closer look.
Compare service by queue priority
Compare recorded average service levels, abandonment and observed volumes for queues with a positive weight and those with zero weight. Use the selected period to assess whether priority groups receive the service you expect.
Turn arrivals into an Erlang C scenario
Start from forecast arrivals or enter a manual volume. Set conversation and after-call time, the percentage to answer within a chosen number of seconds, maximum occupancy and unavailability. When combining queues, confirm that the agents can serve them as one shared pool.
Put coverage in the right time bands
Try constant staffing or assign different headcounts to individual time bands. Compare the effect of an earlier start, a stronger peak period or reduced evening coverage; account for breaks and training through the unavailability assumption.
Keep the scenarios worth discussing
Save personal scenarios with parameters, staffing and the reference traffic. Reopen the original snapshot to review a decision on the same evidence, or create a new scenario from the latest forecast.
See the coverage gap before the busy period
The chart and interval table compare scheduled staff, net agents required and agents to schedule. Read the staffing gap, estimated answers within threshold, occupancy and mean wait to locate the specific intervals that need attention.
Compare two choices against the same demand
Compare the current simulation with a saved scenario using the same reference traffic. Read agent-hours, intervals below target and insufficient-capacity intervals together with each scenario’s service and handling assumptions to choose the most suitable coverage.
Read the signal.
Choose the next step.
Prepare for the next peak with evidence
Start with the seven-day forecast and inspect the comparable weekdays, historical range and evaluation against the baseline. When the history is sufficient, select a day and interval length to see how demand changes across the day. Where evidence is insufficient, keep that limit visible when making a staffing decision.
Put the right coverage in the right time band
A daily headcount can look adequate while the busiest half-hour remains uncovered. Enter talk and after-call time, the service target, occupancy and shrinkage; then compare the net agents required with the staff to schedule. Try variable time bands and read the gap, estimated service level and mean wait before adjusting the real plan.
Compare two choices on the same demand
Save a fixed-coverage scenario, then test a different distribution by time band. A/B comparison applies both staffing approaches to the same reference traffic from scenario A, so staffing hours, under-target intervals and capacity gaps remain comparable. The Erlang C model assumes a homogeneous pool and stable intervals: use it to support a team leader's judgement, not as automatic rostering. No Gemini account is needed.
Use evidence to prepare the next decision: select a queue, review its demand, estimate the coverage and compare saved scenarios. Historical data, local forecasts and Erlang C help you decide where and when to strengthen the team.
Put it to work
Review actual queue arrivals, recent waiting-pressure samples and their freshness, short-term waiting trends, load heatmaps and SLA patterns for the selected queues.
Forecast the next seven days locally from weekday medians, with at least four comparable historical weekdays and profiles in 15-, 30- or 60-minute intervals.
Inspect comparable-day counts and the historical minimum/maximum range, then compare past forecast error with a simple baseline. The historical range is not a confidence interval.
Use forecast arrivals or manual volumes, vary demand, and set a target percentage answered within a chosen number of seconds plus a maximum occupancy.
Set conversation time, after-call work and shrinkage separately so the Erlang C model distinguishes call-handling demand from time unavailable to take calls.
Compare fixed staffing with variable time-band coverage to see where a daily headcount still leaves peaks uncovered.
Read required, planned and net staffing, the coverage gap, estimated service level, occupancy and mean waiting time for each interval.
Save personal scenarios and compare A/B coverage on the same reference traffic, including staffing hours and intervals below target. Calculations run locally without Gemini and do not change staffing schedules or PBX settings.
What you can do
Start with the observed load
Review actual queue arrivals, recent waiting-pressure samples and their freshness, short-term waiting trends, load heatmaps and SLA patterns for the selected queues.
Look seven days ahead
Forecast the next seven days locally from weekday medians, with at least four comparable historical weekdays and profiles in 15-, 30- or 60-minute intervals.
Check the forecast against history
Inspect comparable-day counts and the historical minimum/maximum range, then compare past forecast error with a simple baseline. The historical range is not a confidence interval.
Set the service you want to deliver
Use forecast arrivals or manual volumes, vary demand, and set a target percentage answered within a chosen number of seconds plus a maximum occupancy.
Account for the time people need
Set conversation time, after-call work and shrinkage separately so the Erlang C model distinguishes call-handling demand from time unavailable to take calls.
Shape coverage around the peak
Compare fixed staffing with variable time-band coverage to see where a daily headcount still leaves peaks uncovered.
Read the staffing gap and its consequences
Read required, planned and net staffing, the coverage gap, estimated service level, occupancy and mean waiting time for each interval.
Compare alternatives before deciding
Save personal scenarios and compare A/B coverage on the same reference traffic, including staffing hours and intervals below target. Calculations run locally without Gemini and do not change staffing schedules or PBX settings.
Read the detail with confidence
No Gemini usage
No Gemini usage; current view uses local calculations.
Minimum four comparable weekdays
Minimum four comparable weekdays; insufficient data remains explicit.
Forecast historical min/max is not a confidence interval.
Forecast historical min/max is not a confidence interval.
Erlang C assumes one homogeneous pool and stable intervals
Erlang C assumes one homogeneous pool and stable intervals; no abandonment, skills, prior backlog or shift planning.
Saved scenarios do not change PBX or staff schedules.
Saved scenarios do not change PBX or staff schedules.
History filter and forecast-training history are separate.
History filter and forecast-training history are separate.