Skip to content
Confidirect

One engagement, three notes.

Five years inside one large New Zealand government organisation, building the forecasting and reporting that a large operational function relied on. The organisation and the function are not named, and what is written here is limited to what can be stated plainly.

3,000+

FTE covered by the planning models

Capacity and FTE planning for a 3,000+ FTE operation

New Zealand government · large operational function

Situation
A large operational function inside a government organisation, with a workforce of more than three thousand FTE, needed capacity and operational planning it could rely on rather than rebuild by hand each cycle.
Built
Demand, capacity and FTE forecasting models, with the assumptions written down and the reporting that put them in front of the people making roster and budget decisions.
What changed
Capacity and operational planning for the function ran on these models, covering a workforce of more than three thousand FTE.

Service line: Forecasting and capacity planning

Automated

Forecast refresh, previously manual

Forecasting pipelines that replaced manual work

New Zealand government · same engagement

Situation
Forecasts were produced through manual processes: exports, re-keying and hand-built refreshes that depended on the people doing them and left little time for the forecast itself.
Built
Automated forecasting pipelines that refresh on a schedule, with the checks that flag when reality departs from the forecast.
What changed
The manual processes were replaced: refreshes ran unattended, and the time that used to go into the mechanics went into the forecast instead.

Service line: Data and analytics engineering

50+

Dashboards shipped into production

Reporting infrastructure that a function relied on

New Zealand government · same engagement

Situation
An operational function needed reporting it could run on week to week, not proofs of concept that stalled after a demo.
Built
More than fifty dashboards shipped into production over five years, on the forecasting and reporting infrastructure built alongside them.
What changed
The reporting became the infrastructure that the function relied on. The same person scoped, built, reviewed and maintained it throughout the five years.

Service line: Advisory and embedded leadership

Independent work

The same discipline, applied to different problems outside the engagement.

  1. Demand-forecasting platform

    A hybrid Prophet/LightGBM model forecasting 6.2 million NYC 311 complaints across 14 categories: 27.6% mean error blended across a range from 7.1% on high-volume complaints to 59.2% on strongly seasonal ones, with a live dashboard turning the forecast into staffing recommendations.

  2. Pit-strategy optimiser

    A CP-SAT solver for Formula 1 pit strategy that commits before a safety-car outcome is known. Across 652 driver-races, it captured 58% of the 15.61 seconds hindsight had available, cutting the gap to a 6.61-second regret.

  3. Causal-inference study

    A Callaway–Sant'Anna staggered-adoption design testing whether Starlink actually speeds up a country's internet, using public speed-test data across three launch countries and three controls. The first estimate came back negative; checking it against a likely confounded comparison moved it to a small positive result.

Code and write-ups are at jesse-obrien.com.

Next step

The detail behind these notes is for a conversation.

What can be written on a public page is limited. Ask about any of it in the scoping call and you will get specifics, including the parts that did not go to plan.

Start with a scoping call [email protected]