From the bidding desk to the meter: six appointments

Mandate

Turn a billing-led data estate into one the company could trust for trading, customer care and statutory reporting, and set enforceable rules for how customer data may be used.

Outcomes

• Built an early-support model walled off from collections; disconnections for non-payment fell from 6,400 in 2021–22 to 2,900 in 2025–26.
• Rebuilt the residential solar forecast from customers' meters, cutting midday error from 11.4 to 4.1 per cent and hedging losses by about A$19 million a year.
• Introduced independent reconciliation of every VPP dispatch event; match rate at or above 99.2 per cent since August 2024.
• Delivered an emissions ledger reconciled monthly to billing within 0.2 per cent for the first mandatory climate statement.
• Answered Consumer Data Right requests from the same governed source as the bill and set up a data ethics panel with an external financial counsellor.

Decisions in context

The managing director created the chief data officer role in March 2021 and asked me to take it after four years leading customer and market analytics. The mandate was blunt: our data was good enough to bill customers and not much else. I brought engineering, analytics, data science, forecasting and governance into one function of 260 people, with seven direct reports, and agreed with the executive team which decisions my teams would inform and which they would not. The early-support model came first, because winter disconnections were the problem the board asked about most often. The solar forecast followed a costly spring in 2023, the battery reconciliation a sample check in March 2024, and the emissions ledger the arrival of mandatory climate reporting. I attend the Audit and Risk Committee for data, privacy and climate items, and I sit on the board of our virtual power plant subsidiary in an executive capacity.

Reason for the move

Current role. Not seeking an executive move; considering one non-executive directorship with my board's consent.

Mandate

Give the retail business a single, reliable view of customers, prices and churn, and make use of interval data beyond billing.

Outcomes

• Consolidated interval data from about one million smart meters into a single meter-data platform.
• Built the pricing analytics used when the default market offer and the Victorian default offer took effect in July 2019.
• Reduced annual residential churn from 22 to 18 per cent with retention offers based on bill-shock risk rather than switching intent.
• Built the first settlement and performance analytics for the virtual power plant, which grew from 4,000 to 19,000 batteries.

Decisions in context

Wirrandel recruited me from banking in August 2017 to lead customer and market analytics, reporting to the chief customer officer. The team of 25 was capable but scattered across marketing, pricing and billing, each with its own copy of the customer. We built one meter-data platform for about a million smart meters and one definition of an active customer. When the default market offer and the Victorian default offer took effect in July 2019, my team produced the analysis that showed which customers would be better off on the reference price and supported the letters we sent them. I also changed our retention approach. Offers had gone to customers most likely to leave; we moved them towards customers most likely to receive a bill they could not absorb. Churn fell from 22 to 18 per cent. By 2021 the team numbered 70 and the business wanted a data function, not an analytics unit.

Reason for the move

Appointed chief data officer when the managing director created the role.

Mandate

Modernise the scorecards and collections strategies for credit cards and personal loans after a rise in arrears.

Outcomes

• Rebuilt application and behaviour scorecards for the card and personal-loan books.
• Redesigned collections treatment so that hardship indicators routed customers towards assistance; completed hardship arrangements rose from 38 to 57 per cent.
• Reduced the rate at which 30-day arrears rolled to write-off by about a fifth.
• Introduced quarterly reviews of collections outcomes for customers identified as vulnerable.

Decisions in context

I joined Coorabin Bank in November 2012 to lead credit and collections analytics, reporting to the chief credit officer. Arrears on cards and personal loans had risen and the scorecards were a decade old. With a team of 34, we rebuilt them and then turned to collections, which is where I learnt the most uncomfortable lesson of my career. In 2013 I read a sample of call notes behind a model I had built and saw that it directed our calls towards customers who could pay if pressed, while customers who had stopped coping waited until their debts were written off. We redesigned the treatment so that signs of hardship moved a customer towards an arrangement and a trained hardship officer. Completed arrangements rose from 38 to 57 per cent, and roll rates to write-off fell. I left because energy data could see a household's difficulty months before a bank's could.

Reason for the move

Wanted to return to energy, where interval data could identify difficulty far earlier than a bank's account data could.

Mandate

Build predictive churn and pricing analytics for a fast-growing mobile and broadband business.

Outcomes

• Reduced monthly post-paid churn from 1.9 to 1.5 per cent.
• Replaced blanket retention discounts with offers sized to each customer's likely value and plan fit.
• Introduced proactive plan reviews for long-standing customers paying more than a current plan would cost them.

Decisions in context

I moved to Sydney in May 2007 to join Orrinvale's customer analytics team, reporting to the general manager of customer value. The business was adding mobile and broadband customers quickly and losing them almost as quickly. With a team that grew to 18, I built churn models that drove a retention budget of about A$60 million a year, and later took responsibility for pricing analytics. Monthly post-paid churn fell from 1.9 to 1.5 per cent. The lesson I carried away was less comfortable than the result. Our best retention offers went to customers who threatened to leave, while loyal customers on old plans paid more than new ones. In 2011 I persuaded the business to start reviewing long-standing customers' plans proactively, which cost revenue and was the work I was proudest of there. I left to learn credit and collections, the other place where customer data decides how people are treated.

Reason for the move

Wanted to learn the harder end of customer data, in credit and collections, before returning to a regulated essential service.

Mandate

Improve the state demand forecasts on which the generator's daily bids and contract positions depended.

Outcomes

• Reduced day-ahead forecast error for Queensland demand from 4.6 to 2.8 per cent.
• Built a storm-season peak model that flagged high-demand afternoons two days ahead.
• Wrote the forecast documentation that the bidding desk used to challenge each day's assumptions.

Decisions in context

I moved from Melbourne to Brisbane in March 2002 to join Kowrie's forecasting team, reporting to the manager of trading analytics. The generator bid its gas and hydro capacity into the wholesale market every day, and its bids depended on how much Queensland would use. I rebuilt the day-ahead model around temperature, humidity and the timing of summer storms, and day-ahead error fell from 4.6 to 2.8 per cent. I also wrote documentation that let the traders see and question each assumption, because a forecast nobody can challenge is one nobody fully trusts. By 2006 I led a team of five. In the evenings I completed a master's degree in applied statistics, finishing in 2005. I left in 2007 because I had begun to find the households behind the demand curve more interesting than the curve itself.

Reason for the move

Wanted to work on the customer side of data, where forecasts meet the behaviour of individual households.

Mandate

Build statistical forecasting products from weather data for paying clients.

Outcomes

• Built the consultancy's first temperature-to-demand model for an electricity retailer client.
• Produced frost-risk forecasts used by grape growers in two wine regions.
• Introduced monthly verification reports showing clients how each forecast had performed.

Decisions in context

I joined Fairlie Weather Analytics in February 1999, two months after graduating, and reported to one of its founding meteorologists. The consultancy sold forecasts to grape growers, an alpine resort and, increasingly, electricity retailers, who had discovered that the new wholesale market priced every wrong guess about demand. My first independent project was a temperature-to-demand model for one of those retailers, which taught me how differently a commercial client reads a forecast from a meteorologist. I also introduced monthly verification reports, so that each client could see where our forecasts had been wrong. It was a small firm, and I learnt to explain statistics to people who were paying for the answer and had little patience for caveats. I left in 2002 because I wanted to be inside a business whose decisions depended on the forecast.

Reason for the move

Wanted to work inside an energy business where forecasts were used to make decisions, not only sold.