Indonesian Firms Master AI for Real-World Impact, Not Hype

The question haunting every C-suite executive today isn’t whether AI will transform their business. Instead, it’s whether they’re building castles in the cloud or foundations for sustainable growth.

At a recent panel discussion titled “The Pace of AI – Threat or Opportunity?” at the Chief Digital & Data Officer Indonesia Summit, two seasoned technology leaders from Indonesia’s corporate landscape delivered a sobering reality check on the true potential and pitfalls of artificial intelligence.

“My goal is not to be a cost center, but we should be a profit center,” declared Restu Kresnadi, chief data officer at PT Kalbe Farma, articulating what may be the most critical paradigm shift facing data teams worldwide. This is less about adopting the latest LLM or implementing another chatbot and more about engineering tangible business outcomes in an ecosystem drowning in AI hyperbole.

Beyond the algorithm: Solving real-world problems

While media headlines trumpet each new model release and venture capitalists invest billions in AI startups, Kresnadi and Norman Sasono, chief technology officer at DANA Indonesia, advocate for a decidedly unglamorous approach: start with the problem, not the technology.

Kresnadi’s team at Kalbe Farma exemplifies this philosophy through their nuclear medicine optimization system. Nuclear pharmaceuticals present a complex logistical nightmare — radioactive substances that decay rapidly, require precise dosing, and cost thousands of dollars per unit. Traditional human decision-making simply cannot process the permutations and combinations needed to optimize production timing, dosage amounts, and delivery routes.

“There are so many permutations and combinations that humans will not be able to comprehend and make decisions right away,” Kresnadi explained. “All that is done automatically without people thinking about it.”

This prescriptive analytics addresses a genuine operational bottleneck where human cognitive limitations result in a measurable business impact.

The 90% problem: Architecture over algorithms

Sasono’s experience at DANA, with its 200 million users, reveals another inconvenient truth about AI implementation: most AI projects fail not because of model performance, but due to fundamental architectural and software engineering deficiencies.

“After doing all POCs and prototyping, we realized that all these LLM GenAI efforts end up 90% being architecture and software engineering issues, not AI problems anymore,” Sasono observed. “Building all this and making it work at scale is actually architecture and software engineering problems.”

This approach challenges the prevailing narrative that AI success hinges on model selection or hyperparameter tuning. DANA’s approach looks to build a logical abstraction layer. It then serves as a “model as a service” architecture, decoupling business logic from specific AI implementations.

“You need to be able to encapsulate what varies,” Sasono explained, applying classic software engineering principles to AI systems. “Model varies, model will change, and how you shield that change and variations from impacting everything else becomes crucial.”

This architectural thinking extends to operational considerations as well. DANA, which has implemented comprehensive MLOps and AIOps pipelines, recognizes that deploying models is fundamentally different from traditional software deployment.

Data: The ultimate differentiator

Both executives converge on a provocative thesis: AI itself will become commoditized, relegating model capabilities to the status of utilities, such as electricity or cloud computing. The real competitive advantage lies in proprietary data and domain expertise.

“AI will become pervasive, no longer a key differentiator,” Sasono predicted. “What will make the difference is your data — data behind the login that only you have access to in the world.”

DANA is exploring advanced techniques, such as knowledge distillation and fine-tuning, with proprietary datasets to create models that competitors cannot replicate. While foundation models provide a baseline of capabilities, the company understands that a sustained competitive advantage requires leveraging unique organizational data assets.

Kresnadi reinforced this perspective, noting that finding technical talent — programmers and modelers — presents fewer challenges than identifying professionals who can bridge business domain knowledge with AI capabilities. “Those are the people that I need the most,” he emphasized. “People who can understand the business and translate that into an AI project.”

Governance without paralysis

The discussion also addressed the growing tension between innovation velocity and risk management. Sasono described DANA’s AI Governance Committee — a cross-functional team comprising engineers, legal professionals, and risk management specialists who review AI initiatives before they are deployed in production.

“[The committee] should not be seen as a blocker, but as implementing a proper process so that you don’t end up building something, deploying it to 200 million people, and then discovering something catastrophically wrong,” Sasono explained.

This governance committee also exemplifies a mature AI strategy: establishing guardrails without creating bureaucratic bottlenecks that stifle innovation.

The minimum viable AI approach

Perhaps most pragmatically, both leaders advocate for incremental adoption of AI rather than comprehensive digital transformation. Sasono explicitly rejected the “fix all data problems first” approach that paralyzes many companies.

“Start with the use case, identify what problem you’re trying to solve with AI, then determine what specific datasets you need,” he recommended. “Just fix that one data pipeline first, so you can immediately articulate the outcome of that AI implementation.”

This approach follows agile software development methodologies, which focus on rapid iteration and continuous value delivery over traditional waterfall-style comprehensive planning.

So, is AI a threat or an opportunity?

The panel’s implicit answer to their main question is that AI represents an opportunity for companies that approach it with an engineering discipline, a business focus, and realistic expectations. The main threat is not in the technology itself, but in the strategic missteps that waste resources and delay genuine value creation.

As AI models continue their relentless improvement trajectory and costs plummet toward zero, the companies that will thrive are those building sustainable competitive advantages through superior data assets, domain expertise, and architectural sophistication. The future belongs not to those who can deploy the latest model, but to those who can engineer meaningful business outcomes from the intersection of AI and human insight.

Image credit: iStockphoto/DisobeyArt