Ampol is an independent Australian company and the nation’s leading transport energy provider. Ampol also has a growing presence in New Zealand as the owner of Z Energy.
Ampol’s Trading and Shipping business operates out of Singapore and the USA, and has positions across the Asia-Pacific region.
Ampol’s future energy and decarbonisation strategy sets out our plans to transition our business to future fuels and energy solutions. As part of this, we have launched our electric vehicle charging and home electricity solutions to continue to deliver to our customers changing energy needs.
Best Acceleration Use Case
Best Approach for Building Trust in AI
Ampol is Australia’s leading transport fuel supplier and has operations and interests across Australia, Singapore, New Zealand, Philippines, and the USA. Ampol Trading & Shipping (T&S) is the commodity trading arm of Ampol responsible for trading all imports and exports of refined products, crude oil, feedstocks and other intermediates. Ampol T&S plays a crucial role in providing crude oil and other inputs to our Australian refinery, and the safe and reliable supply of fuel to customers.
Ampol T&S’s trading capability is underpinned by access to actionable commercial analytics. A key function of the T&S Commercial Analysis team is the delivery of fundamental analysis in relation to global crude & product markets to provide insights to its trading team.
Consolidating and assessing global product flows encompasses many data sources and multiple models are needed to appropriately capture individual geographic and product nuances.
Historically, this activity was highly manual, and many analysts worked in a siloed structure: sourcing and cleaning data, building models, and sharing outputs that were not well-integrated. Cumbersome Excel files were becoming difficult to maintain and did not provide the scalability and speed needed in a growing business.
With the team and the broader Ampol business recognizing the increasing importance of advanced analytics, there was a need to explore more flexible and agile analytical tools that could provide scalable, timely and advanced analytics to drive trading insights.
The project sought to demonstrate how forecasting processes could be achieved with machine learning, with a portfolio of models required to be scaled across 14 products, >50 countries and multiple supply/demand components. These models needed to be well understood by the Commercial Analysts who work with forecasts day-to-day.
Dataiku was used as the platform to deliver new predictive global balance models and enabled collaboration within the Commercial Analytics team.
Our Data & Analytics team adopted an agile approach to demonstrate how this could be achieved, targeting two-week sprint cycles for each product i.e. delivery of new models for feedback every two weeks. Once a portfolio of proven models was tested, we progressed to scaling models across the full set of required markets.
Each sprint began by conducting a thorough analysis of the historical data, identifying patterns, and understanding relationships between the various balance components. We engaged the Commercial Analytics team through this process to form hypotheses, and understand data and recent market trends.
We tested multiple predictive model candidates such as ARIMA, VAR and gradient boost tree. These were tested these against a naïve model benchmark, which replicated simple rule-based approaches. Selection of the best models from these forecasting techniques resulted in robust models for predicting different products, with demonstrated performance in the last 4 years' turbulent energy markets.
A key objective of the work was to demonstrate new ways of working for the Commercial Analytics team, who previously have worked primarily in Excel. To address this, we onboarded the Analytics team to Dataiku as part of the project. Collaboration on the modelling in Dataiku’s code environment helped to build understanding of Dataiku’s capability, including better understanding of the benefits of new models, and also that the platform can replicate current approaches if needed.
In only a handful of sprints, we have been able to clearly demonstrate the superior performance of new models and the ability of the platform to scale this process, resulting in strong buy-in for the next phase of deployment.
Implementing Dataiku as a modelling solution has demonstrated the ability to transform the process of developing market balances.
Previously, the forecasting process demanded substantial manual effort from analysts. A key goal of the program is to eliminate manual processes for preparing data. The balances models delivered allow us to retire 50+ Excel-based models and have facilitated a shift in our ways of working moving from a siloed individual desk approach to a collaborative workspace.
This new way of working results in a single data asset which enables a range of new views across products; the ability to roll back and govern releases; agreed sources of truth which can be shared; and greater accountability and trust in forecasts.
With centralized and trusted data foundations, the team can focus on value-generating analysis and speed of insight generation to our trading desks and executive management.
Business Area Enhanced: Supply Chain/Supplier Management/Service Delivery
Use Case Stage: Proof of Concept
The time series forecasting models developed through Dataiku have successfully provided more accurate and timely predictions for the balance components. These improved forecasts directly translate into more informed trading decisions, which contribute to the global trading P&L.
As analysts continue to become more comfortable with Dataiku and developing their own models, we are continuing to identify other analytical areas of opportunity that would similarly benefit from advanced analytics. This pipeline of areas for further improvement is now feeding back into team capability goals and future analytics strategy.
Value Brought by Dataiku:
Dataiku has been an effective platform for enabling new analytics capability in the Commercial Analytics team. The speed with which new users are able to start using the platform played an important role in facilitating the project, building trust and helping boost collaboration. The ease with which models can be scaled meant that complex modelling outcomes were delivered with confidence.
Using Jupyter Notebook to review models and demonstrate deliverables proved to be a constructive way of working and allowed more data savvy stakeholders with deep business knowledge to get into the detail and build trust throughout the process.
With our recent Dataiku upgrade from V9 to V11, we specifically looked to leverage up-to-date Python libraries. This included SKtime, a comprehensive python framework for time series forecasting which meets needs for cross-validation, out-of-sampling, and visualization. The efficiency we achieved in delivering this work would have been impossible without access to Dataiku’s V11 features.
Dataiku's resources, such as the Dataiku Academy and Community, facilitated upskilling and networking opportunities, empowering the project team to leverage the platform's capabilities effectively.
Overall, Dataiku was a strong solution for embedding advanced analytics with the commercial analyst group as they migrated from Excel towards ownership of more advanced modelling needs.