Enterprises are investing in AI, but moving from experimentation to production remains a challenge. High inference costs, legacy systems and difficulty measuring productivity gains are holding back wider deployment, even as companies redirect existing technology budgets toward AI infrastructure.
Pari Natarajan, CEO and co-founder of Zinnov, a management consulting and strategy advisory firm, says enterprises need to rethink how they measure AI’s value and prepare their organisations to use it at scale. In this interview with Cybernetix, he discusses the obstacles to enterprise adoption, the shift in technology spending, the implications for jobs and India’s opportunity to build more AI-driven businesses.
The interview has been edited for clarity and length.
Cybernetix: Enterprises are investing heavily in AI, but there is still considerable uncertainty about the returns. From your interactions with companies, how is AI adoption progressing, and what are the biggest challenges they are facing?
Pari Natarajan: Companies are spending quite a bit of money on AI platforms, but they are struggling to articulate the value generated from that spending. Boards are asking CIOs what the impact of their investment has been, and one of the measures they often look at is headcount reduction.
Another challenge is that internal systems are not ready for AI adoption. Many enterprises have legacy systems built over the past 40 or 50 years. Making those systems work with newer AI capabilities requires investment in the data layer and other parts of the technology foundation. Many companies have not adequately planned for that.
There is also an organisational challenge. Leaders may accept that change is necessary but still struggle with how to make it happen. AI adoption requires more than introducing a new application; it can require changes to processes, systems and the way people work.
Cybernetix: What proportion of AI pilots are reaching production, and what is preventing more of them from scaling?
Natarajan: I would say at least 10% to 15% of pilots are getting into production. Running a pilot for three or four months is much easier than deploying something at scale.
Even when a pilot works, a company may not be able to roll it out everywhere immediately. It might deploy an AI application for a particular set of customers or in selected regions before expanding it further. The cost can become too high if the company tries to scale the deployment all at once.
Token costs are one factor. Organisations may wait for those costs to come down before expanding their use of AI. But cost is not the only issue. The organisation also needs the systems, data and processes required to support a wider deployment.
The situation is improving. More use cases are reaching production, and the time needed to move from a pilot to production is coming down. But scaling a successful pilot remains a different challenge from proving that the technology works.
Cybernetix: Are enterprises increasing their overall IT budgets to accommodate AI, or reallocating existing spending?
Natarajan: They are primarily reallocating budgets. Some spending is moving from services to infrastructure, while some is shifting from labour-related expenditure into IT and AI infrastructure. They are not necessarily adding an equivalent amount of new money to their overall budgets.
My estimate is that about 20% of the money is being redirected toward AI infrastructure in the markets we discussed, particularly in developed economies. That does not mean every enterprise is doing this at the same rate, and the figure should not be applied uniformly to India.
Indian companies may take a more measured approach because labour is still relatively inexpensive. They may wait until the value of a particular AI application is more clearly demonstrated. Some companies, including Bajaj Finance and newer digital-native businesses, are moving faster.
Cybernetix: Even when AI saves employees time, why is it difficult for companies to demonstrate a return on investment?
Natarajan: Suppose someone uses AI and saves two hours a day. The question is whether that person uses those two hours to do more work for the company or spends the time on something else. The company may not know what has happened to that productivity gain.
If an employee can prepare a presentation in 15 minutes instead of spending an hour on it, the time saved does not automatically become additional business value. The organisation needs to determine how that capacity will be used.
One way to address this is to push the system to do more. If a team can complete work faster, the company needs to create additional work for that team or find ways to use its capacity elsewhere. Otherwise, the productivity improvement may not translate into a measurable business outcome.
That is why companies need to look beyond whether employees are using AI or whether a task is completed faster. They need to understand how those gains affect the work the organisation can deliver.
Cybernetix: What should CIOs prioritise to get better returns from their AI investments?
Natarajan: First, invest in a strong data layer. AI applications need access to the right data, and enterprises must build the foundations required to make that data useful.
Second, consider a more horizontal organisational structure rather than relying entirely on hierarchical arrangements. A more horizontal structure can help people work across functions and move initiatives forward more easily.
Third, create a learning culture. Organisations need to bring in new capabilities, help employees develop their skills and make learning part of how they operate.
These changes matter because the challenge is not just to introduce AI technology. It is to prepare the organisation to use that technology effectively.
Cybernetix: How does India’s AI adoption compare with that of more developed markets, and where does it need to improve?
Natarajan: Developed markets are redirecting more money toward AI infrastructure. Indian companies are also adopting AI, but some may take longer to invest because labour is still relatively inexpensive and the value of certain use cases needs to be demonstrated.
There are companies in India that are doing well with AI, particularly some digital-native businesses. But if India wants to do more than adopt technology developed elsewhere, it needs to invest more in core research and development.
India spends about 0.6% of its GDP on R&D, compared with around 5.4% in South Korea, according to the figures I cited in our conversation. That is a significant gap.
India has attracted global capability centres that undertake high-quality work for multinational companies. The next step is for Indian companies to invest more in their own research, invention and the development of new capabilities. Building applications on top of existing models requires investment and innovation, too.
Cybernetix: There is considerable concern that AI will reduce employment opportunities for young engineers. Do you expect AI to create more jobs than it eliminates?
Natarajan: It is unclear right now, but my view is that AI will create more jobs over time. What is less clear is exactly what those jobs will look like.
We are seeing changes in junior-level hiring. In our discussion, I referred to an estimate from Draup’s analysis indicating that hiring at the junior level was about 20% lower globally than it had been previously. That is a change worth watching, although it does not, by itself, establish that AI is the sole cause.
AI will also make it easier for people to start companies. Consider someone who wants to launch a business. AI-enabled services could help provide capabilities such as finance, customer relationship management, sales and customer support, allowing a founder to concentrate on the core business rather than building every support function from scratch.
We have estimated that 100 million entrepreneurs could emerge globally over the next 10 years as starting and operating businesses becomes easier. That is an estimate of potential entrepreneurship, not a forecast of 100 million jobs. The employment impact will depend on how these businesses develop and how many people they hire.
