Xcelerate
Menu
Get in touch
All articles on forecasting and prediction

Forecasting and prediction

Forecasting building inspection demand for NZ councils

Estimate inspection slots using local bookings, consent stages and reinspection history. Test how well a weekly forecast would support a council roster.

A planning team reviews inspection demand forecasts beside a construction site.
AI-generated editorial image.

A building consent can lead to several inspections at different stages. Several dwellings can share a consent. Neither the number of consents nor the number of dwellings tells a council how many inspectors to roster next Thursday. Build the forecast around booked and likely inspection slots, then ask which staffing decision the estimate can still change.

Which figures represent inspection work?

Use inspection records to distinguish requested visits, completed visits, cancellations, repeat visits and time spent travelling. Specify the forecast unit. A half-day of specialist work and a short routine visit both count as one appointment, yet need very different resources. Where possible, convert visit types into workload using locally observed durations and separately show the number of jobs.

Stats NZ’s building consent methodology explains that its new dwelling measure counts dwelling units. The series describes consenting activity in its reference month. It can provide context for a longer horizon. Use local consent stages, actual inspections and lead times for the operational estimate. Test whether the published series adds any predictive value after those local signals are included.

How much work is already booked?

Separate confirmed appointments from work that might arrive. Start an example week with 26 booked visits. If the illustrative planning assumption is four cancellations and nine new visits, the expected total is 31. Display all three components. A scheduler can respond differently to a large volume of confirmed work and a wide range of uncertain, late requests.

Add a skill and travel check. Two rural inspections may take most of a morning even if their on-site work is brief. Show where workloads exceed the hours available for a particular endorsement or district. Avoid publishing precise addresses in a planning dashboard where an area grouping will support the same allocation decision.

Show the service owner a forecast range as well as the central figure. If the lower and upper estimates both fit the available hours, the scheduler may leave the roster alone. If only the upper estimate exceeds capacity, they might agree an on-call arrangement. That distinction should be visible before anyone authorises extra staffing.

Would the forecast have changed the roster?

Replay past roster dates using only bookings and consent stages that staff knew at that point. Compare the proposed model with the current scheduling rule. Report errors in inspector hours by week and area, then examine the busiest missed week with a scheduler. A good district-wide average offers little comfort if the same rural area repeatedly misses out.

Attach the original forecast to each later comparison. If a subdivision pauses or a contractor reschedules a block of visits, record when that news arrived. This gives the service owner a fair distinction between an avoidable modelling error and a change nobody could have known at the roster deadline. Ask for a live queue view and a fallback estimate for weeks when source data arrives late.

Sources and further reading