AI at enterprise scale: a conversation with Naren Gangavarapu
PhoenixDX sat down with Naren Gangavarapu GAICD — Chief AI Officer, ranked #1 Technology Leader in Australia by iTnews — to talk about the realities of moving beyond the AI pilot phase.
This week, Naren will take the conversation further as a panellist in the fireside chat at our exclusive executive experience: The Last Competitive Advantage — Leading Agentic AI Before It Leads You.
Q. You’ve worked with AI in different forms for more than 25 years. What feels genuinely different about this current AI moment?
Naren: There are two distinct factors that are rarely discussed together. The first is accessibility; for decades, AI required specialist coding skills in languages like Python to build pipelines. Now, the interface is natural language, making AI accessible to everyone. The second factor is a "failure mode inversion." In the past, AI failed loudly - it was obvious when a prediction was off. Today, generative AI is so sophisticated that it often "fails elegantly" - it presents errors so convincingly that unless you are a subject matter expert, you may not realise it has failed at all.
Q. After a year building agentic systems, you’ve argued that rules-based and prescriptive AI are winning for many enterprise use cases. What led you to that conclusion?
Naren: For decisions that require regulatory compliance, audit trails, and fiduciary accountability, rules-based AI is outperforming Generative AI. Rules-based systems provide guarantees, testability, and clear inputs and outputs. Generative AI, by contrast, samples from a distribution—the coverage is never complete, and the output can drift, making it difficult to prove why a decision was made. I advocate for using rules-based AI to build the "backbone" of your system and using agentic AI on the edges for tasks like summarisation and user-facing interfaces where downstream impact is lower.
Q. Are organisations becoming too focused on GenAI and agents rather than starting with the business problem they actually need to solve?
Naren: Yes, this is a recurring issue with new technology. Organisations often find a tech package and try to force it into their operations rather than identifying the root cause of a problem. The second mistake is over-engineering - using expensive generative models to solve problems that could be addressed by much simpler solutions. The key question should be: "Does this business area have proprietary data and knowledge worth protecting?" If you do, you should own the model rather than renting a subscription, which effectively hands your competitive edge over to the vendor.
Q. Where have you seen agentic AI genuinely earn its place in production — and what made those implementations successful?
Naren: For one of Australia's largest tourism and hospitality groups, we successfully implemented 25 agents across the booking, wholesale distribution, and cruise operations. The results were clear: a 34% drop in tier-one support escalations, a reduction in onboarding time from 23 days to 3, and a 24% rise in booking conversion. The secret is "boring" foundation work: building a solid backbone, using the source of truth for your data rather than spreadsheets, and avoiding manual manipulation. Treat agents like employees - they require foundation, structure, and accountability.
Q. You’ve said that organisations often confuse AI capability with AI reliability. Why is that distinction so important at enterprise scale?
Naren: AI capability demonstrates that a technology can perform a task in a demo environment. Reliability demonstrates that it will perform consistently, repeatably, and compliantly in production. Building reliability is the hardest part. You cannot let an AI "self-mark" its own exam - you need rigorous, external testing mechanisms to ensure that the system isn't drifting.
Q. What does a board need to see before it should confidently approve a significant investment in AI?
Naren: Boards should focus on four critical questions to move beyond the hype:
- What specific decisions will this system make or influence, and what happens if it is wrong?
- Can we produce a complete, auditable record of how it reached that output?
- Do we have the specialised skills and tools to test it adequately both before and after deployment?
- What controls exist to immediately detect and address behavioural drift?
Q: What separates an impressive AI pilot from an AI initiative that delivers genuine ROI?
Naren: It comes down to operational discipline. You must avoid the trap of measuring success by declining demand or "vanity metrics”. ROI comes from ensuring your data comes directly from the source of truth, measuring impact during peak operational windows, and ensuring your processes aren't just "automated" - they are optimised and defensible.
Q: If a CIO told you, “We have dozens of AI pilots, but we’re struggling to demonstrate value,” what would you advise them to do next?
Naren: Go back to the basics. Stop chasing market pressure. Focus on your proprietary data and build your own models where you have a unique edge. Look at your operational habits - if you can treat AI as another employee whose performance and accountability are tracked, you will have a much higher chance of a successful implementation.
Q: What is one piece of conventional wisdom about enterprise AI today that you think technology leaders should challenge?
Naren: Challenge the idea that AI is the default solution for every problem. The conventional wisdom is that if you aren't using the latest AI models, you are falling behind. In reality, the leaders are those who are using the right type of AI - whether rules-based, prescriptive, or generative - to build a robust, reliable backbone for their specific business needs.
Biography
Naren Gangavarapu, GAICD, is a Chief AI Officer and senior technology executive based in Sydney, Australia. Ranked Australia's #1 Technology Leader by iTnews in 2024 and #5 CIO by CIO50 in 2022, he has spent 25+ years delivering more than $1B in business value across government, private equity and complex service industries.
Naren specialises in AI-powered transformation, post-merger integration and large-scale digital strategy, with a track record of deploying AI in production rather than in pilots, helping organisations embed it as a competitive advantage, not just a capability. His hands-on experience spans from early expert systems and machine learning through deep learning, NLP, and computer vision to today's agentic and GenAI applications, including Copilot rollouts, ESG reporting automation, customer sentiment analysis, and legal and compliance tooling.
Career highlights include establishing the award-winning omni-channel platform for Service NSW as the single window for government transactions, achieving 98% customer satisfaction and delivering a government priority three years early; halving state significant project assessment timeframes from 298 to 145 days to unlock $18B in capital investment and 59,000 jobs; and leading a $500M+ innovation program spanning digital licences, live traffic, IoT, facial recognition and payments.
