Published on the 08/08/2024 | Written by Heather Wright

Ready for AI, but reaping data benefits now…

“Data is not IT,” says Datin Ts. Habsah Nordin, head of artificial intelligence centre of excellence for Petronas.

Nordin, who now leads Petronas’ artificial intelligence centre of excellence, was instrumental in a three-year data liberalisation project at the global energy company.

“In order to bring the value from data, liberalisation is not an option but a need.”

It’s a project which has connected to 260 data sources and has seen the company ‘empower’ 64 techno-digital solutions in three-and-a-half years, representing about US$2 billion of value creation, Nordin told attendees at the Gartner Data and Analytics Conference in Sydney.

The project included bringing 227 separate legal entities globally into a Group Data Liberalisation Agreement, sharing their data across the wider organisation.

And that’s where Nordin’s comment about data not being IT comes into play.

“Data is within the business domain, so every single business is guarding their data.”

Data liberalisation is ‘a very risky change’ in the eyes of most of the subsidiaries.

“But in order to bring the value from data, liberalisation is not an option but a need.”

Petronas operates in more than 100 countries, including Australia and New Zealand, with more than 250 legal entities and 47,000 employees worldwide.

“We have business activities from exploration into retail and so our data is very complex and with [huge] amounts of data coming in.”

The company could see the opportunities in data analytics and AI, but needed to fix how it managed data.

In 2020 it set a critical mission to make Petronas a data-driven organisation.

Data sharing, which wasn’t available seamlessly across Petronas’ 250+ subsidiaries to the parent company,  was the first challenge to be addressed.

The data liberalisation program made data sharing from all legal entities to the parent the default standard, setting a principal for there to be a single source of of truth that would enable the business to liberalise data horizontally and vertically across the organisation.

“At the same time, we understood some data was very sensitive,” Nordin says. Seven ‘special categories’ of data were introduced, including personal data, and competitive and IP related data, which required higher due diligence processes to be undertaken by both the business and legal prior to enablement of the data into the enterprise data hub.

It was the concept of data liberalisation that was the biggest change, Nordin says. It would mean internal, open data was available to all in the business.

“It’s a very different concept we do because we believe that if you want robust analytics, you shouldn’t have a blind spot. So liberalising the data matters.

“We also liberalise data that is confidential and secret, however, that is done with the right entitlement and the right data security.”

Operating models were also changed. Prior to 2020 the organisation had seen things through the lens of document and record management. Now the focus was data management.

Petronas established data governance ‘from a blank piece of paper’, using the DAMA (Data Management Association) framework as its basis.

Centralised governance, through the Centre of Excellence, was established to lead big programs, developing and implementing them across the business, but with the option for execution to be done by the business in some cases, and in others where the project was ‘very unique’ to the business, for the business to decide on design and execution themselves.

“With that we have institutionalised and created data roles,” Nordin says. Executives, including the president, CEOs of every subsidiary, senior VPs and EVPs take responsibility and accountability as data owners. A data committee brings all the executive data stewards across the organisation to plan the next tools and approaches.

“This set up of a data committee doesn’t just apply at parent company, it is also being repeatedly simulated across all the business, and the chair at each business is the data owner themselves.”

It’s an initiative that has seen Petronas upskill 750 personnel.

Building an Enterprise Data Hub

A dual cloud-based Enterprise Data Hub (EDH), based firstly on Microsoft Azure, has been implemented, moving the company from disparate platforms to an integrated ecosystem and that all important single source of truth. An immersive front-end portal, Data+, enables employees to access data and analytics products and an Enterprise Knowledge Intelligence Hub to enable employees to discover contextualised data and knowledge is being implemented.

Data quality was key, Nordin notes.

“Data quality is not a project. To us it is business as usual. It is embedded in the process.”

Two checkpoints were implemented for data quality, with profiling at the data source and again at the EDH.

“We want to make sure transparency of data quality matches to the employee in order for them to trust the output they will get from the analytics.”

Petronas set a target quality score improvement of 90 percent by December 2023 and achieved 91 percent, based on 84 data sources, in January 2024.

“Along the way we have defined data standards across multiple domains within Petronas because it allows for data interoperability as well as the metadata and data security and access control.”

Nordin says the EDH, designed as a single domain, functions as data-as-a-service, advanced analytics-as-a-service and platform-as-a-service.

But Petronas is also keen to harness all the unstructured data within the organisation – and AI.

“The Enterprise Knowledge Intelligence Hub is the next milestone. How do we convert data and AI so we can power up the right insights to the organisation?

“We have seen from the techno-digital projects that we are only just tapping the surface. There is only 10 percent of the data that we have used for our analytics. But the bigger insights are from the unstructured.

Cognitive services, including object detection, image and text extraction and named entity recognition were introduced.

Nordin says 59 analytics models were deployed, and 2.1 million pages were processed with an extraction rate of around 0.8 seconds per page, resulting in a 75 percent man-days reduction. Manual extraction of all Petronas’ documents would take 21 years, with EDH the company believes it will be able to do it in three months.

An additional eight-month project also saw generative AI integrated with semantic search and knowledge graph, to reduce the risk of hallucinations.

Released in July 2023, J.AI is now part of Petronas’ enterprise GPT and Nordin says it’s brought ‘enormous’ benefits in productivity and efficiency.

“What we are seeing today is if you want to move into adoption of AI at scale, fixing your data will be the first thing you need to concentrate on, because if you are not able to do that and it’s not sustainable and not a data culture within the organisation, scaling of AI will be a big challenge for many organisations.”