Modernization is most effective when architects are at the helm. In an interaction with CIONow, Chander Khanduja, Group CIDO, Baxy Group explores how companies can integrate legacy systems and edge devices, minimize duplicated data, thoughtfully apply AI and select appropriate cloud models without taking on technical debt.
How do you manage the transition between legacy systems and real-time edge devices without introducing technical debt?
I think this is one of the most challenging aspects. With all the discussion around GenAI, data lakes, and related technologies, one principle has remained unchanged as technology has evolved: the importance of a strong architecture for growth. Even as new layers such as automation, IoT, GenAI, agentic AI, and edge devices have emerged, this core principle still holds true for every organization.
In my previous organizations as well, many of the technologies we talk about today did not exist then. Still, I can recollect the architecture that we had built and talked about during those times. The data, whether from the field, from machines, from a laptop or from a mobile device, it needed to consolidate in a single data warehouse. That basic approach remains relevant today. The real challenge is to ensure that there is no redundant data.
The fact is if you allow redundant data to proliferate into your database you run into real issues. The main thing we were told back when we started still reigns true, which is that your output will be garbage in Garbage out. So any initiative we undertake should ensure that there is one version of the truth.
What best practice helps remove redundant data?
What really matters is what information do you actually need? In many surveys, and this is largely correct, we find that companies hold data they rarely if ever use, as much as 80% in many cases.
The task here is not to abolish the function data plays; you will still want to capture information. But the question becomes: how do you reduce 80 fields to 20 when appropriate. Far too often, we are not careful enough in looking at this in the planning phase of any initiative. It is essential to spend enough time on this step and not just discuss it with the process owners. Data is ultimately consumed by many people across the organization, so you should form a data relevance team.
This should be a cross-functional team with representatives from different functions, and its role should be to discuss what data is required by each stakeholder. That will make the data governance strategy much more relevant.
Are you seeing organizations build such teams?
Yes, many large companies are already doing this. I would honestly say that our current setup is still at a relatively early stage of maturity, because we started our digitization journey only three to four years ago, so we have not reached that stage yet.
But many other companies have already started doing this. Many CIOs in larger knowledge groups already know these approaches, and seasoned CIOs have been doing this for some time.
How do you create bottom-line value with AI? Which use cases do you choose to pursue and which do you choose to abandon?
Step one: find the stakeholder. You need to bring the right people along with the process. If you only identify the process but the key stakeholder is not responsive, any technology implementation is likely to fail. So both people and process matter.
We began with a focused use case in our legal department. We introduced GenAI as a solution there. Our legal head was very open to implementing technology initiatives. That helped a lot.
It has now been running for about one and a half to two and a half years, and we are currently working on two or three use cases there.
My approach over the years has been simple: every technology project must serve one of four purposes.
It must improve controls and compliance.
It must create customer delight.
It must improve operational efficiency.
It must save cost.
I have followed this principle for nearly 25 years, and it still applies today. In legal, for example, compliance benefits are significant, so ROI is easier to justify.
Meanwhile, I have seen another painful trend: even when GenAI wasn’t needed for a problem, people tried to inject a GenAI solution into that problem. That is a bad problem
In reality, GenAI can become very expensive at scale. Token-based models may look attractive at first, but once deployed across the enterprise, costs can rise sharply.
Such technology costs would naturally come down in due course of time. It was the same with IoT where a 2000 rupees technology is now being sold at 100 rupees. This is because technologies evolve, consumption gets more optimized and for a consumer, the cost-benefit comes more in favour. I believe the same will happen with GenAI as well.
Are you building an enterprise-wide AI solution right now?
Not at this stage. We started our SAP implementation only three years ago but in a short period, we have moved quickly.
Right now, we have already implemented enterprise-wide IoT in our plant. For GenAI, however, we are not yet looking at a full enterprise-wide rollout across the plant. That said, we are definitely trying a few things.
We have some pilots running in manufacturing, some work happening in HR, and some initiatives in sales and marketing. Since we also have other businesses, including real estate, there are several customer-facing areas where AI can improve response times and efficiency. We have already begun exploring those opportunities.
How should CIOs think about cloud repatriation as cloud costs rise and sovereign cloud concerns grow?
First your strategy was to go to cloud first since we didn’t have the internal expertise and wanted to go to market very quickly. Cloud vendors raised prices over the last 10-15 years, and there were concerns over sovereignty, and we are also in a geopolitical environment where countries don’t necessarily want to host sensitive data and workload outside their borders. Thus, those organizations that have been to the cloud for 10 to 15 years and have already begun moving their workload to the cloud are the ones feeling the sting of the cost. And they are the first ones to start shifting their strategy towards a hybrid cloud approach.
And that is perhaps the direction to where we will all be moving; hybrid-first, not cloud-first.
The key question now is not just where the data is hosted, but how much control the organization has over it. If the data is already within the same geography, it may not make sense to pay a premium for a sovereign cloud if the cost is many times higher. In some cases, it may be more sensible to use a third-party data center or host the infrastructure on-premises in a secure data center rather than continue with an expensive cloud model.
