Video · September 15, 2025
How Endeavour Energy went from a black box Python model to 98% forecast accuracy in 10 minutes
About this video
Endeavour Energy, an electricity infrastructure owner and operator serving 2.7 million people across New South Wales, Australia, had built its own Python machine learning model for revenue forecasting. The self-developed code was effectively a black box: results were difficult to explain, drivers were unclear, and maintaining the model required specialized skills that created significant key person risk across the Finance team.
SensibleAI Forecast replaced that opaque model with a transparent, Finance-owned forecasting capability. Accuracy improved from 92% to 98% and forecasting time compressed from two days to 10 minutes. The Finance team now maintains the model without data science expertise, explains forecast drivers to regulators and shareholders with confidence, and engages in far more intelligent conversations about energy consumption patterns and customer behavior.
Speakers
Head of Commercial Finance | Endeavour Energy
Key takeaways
- A self-built Python forecasting model created a black box Finance couldn't explain,maintain, or trust fully. Key person risk, opaque drivers, and an inability to clearly communicate forecast rationale to regulators made a better solution essential.
- SensibleAI Forecast delivered the transparency a regulated utility needs tosubmit forecasts with confidence. Understanding the relative impact of different drivers on revenue forecasts is not optional for Endeavour Energy. It is a regulatory requirement.
- Forecasting accuracy improved from 92% to 98% after implementingSensibleAI Forecast. For a business thatsubmits annual regulatory forecasts, that six-point accuracy gain has direct implications for compliance, credibility, and stakeholder confidence.
- Forecasting time compressed from two days to 10 minutes, freeing Finance for higher-value analysis. The team now spends time understanding patterns and drivers of customer behavior rather than running and maintaining a complex manual forecasting process.
- Finance teams nowmaintain the model without data science skills or the ability to write code. Removing the specialist dependency that created key person risk gives Endeavour Energy a sustainable, Finance-owned forecasting capability that can grow with the business.
Read Full Transcript
My name is Rebecca Yu, Head of Commercial Finance at Endeavour Energy. We're an electricity infrastructure owner and operator in New South Wales, Australia. Our network covers from Western Sydney to the Blue Mountains all the way to the South Coast.
We have 2.7 million people living, working in our network. Before using OneStream, we had particular challenges around planning our network as well as forecasting our revenue. That's why we embarked on self-developing, self-teaching Python code to start because we had a self-developed code for the Python machine learning model around revenue forecasting.
It was pretty much a black box. We couldn't understand it very well in terms of the thought process and the drivers that led to the forecasting results. We also had a lot of manual effort for a key person risk in terms of maintaining and tweaking the Python code.
Wetting that black box, that's when we realised we needed a more efficient solution to help us navigate the complexity of our revenue forecast. OneStream's sensible AI forecasting really stood out, giving us the transparency that we needed in terms of forecasting revenue, understanding the relativity of different drivers, the impact to our revenue forecasts.
We as a regulated utility business, we need to submit our forecasts to the regulator every year. We're benefiting from the output from the sensible AI forecasts in our regulatory submission process. The benefits we've seen since we went live with sensible AI forecasts, our forecasting accuracy have improved dramatically.
Our forecasting accuracy was sitting around 92% prior to our implementation and over the last six months we're now able to achieve 98% of forecasting accuracy. In addition to that, we have shortened our forecasting time frame down from two days to 10 minutes.
Having the finance team being able to maintain the model without being a data scientist, knowing how to code, knowing how to code and we now have visibility of the relativity of different drivers impacting our forecast that we can explain.
We're able to have much more intelligent conversation with our shareholders, with our management in terms of what's impacting our energy consumption at a much more granular level, understanding the patterns and drivers for our customer behaviour.
So much innovation has been invested by OneStream. As a customer, we're really excited to see the pipeline in terms of innovation and we really want to be along the journey with OneStream.
Related resources
Parkdale Mills
How Parkdale Mills eliminated seven Excel consolidations and cut close by two days in year one
BDO
How BDO replaced hundreds of weekly emails with one unified Finance platform
Cor Partners
How Cor Partners moved from version-controlled Excel files to one real-time source of truth
BlueScope
How BlueScope eliminated 75 spreadsheets and accelerated Finance with OneStream
