IT services companies were one of the quickest to integrate generative AI and launch GenAI offerings, besides the hyperscalers and SaaS companies.
However, the burning question on everyone’s mind is: How much revenue are these IT companies actually generating from generative AI alone?
Companies like Accenture have announced billions of dollars in revenue from generative AI so far. Indian IT giant TCS, too, has revealed a generative AI deal pipeline worth $900 million.
However, according to Tanvir Khan, executive vice president of cloud, infrastructure, digital workplace services, and platforms at NTT DATA, much of the generative AI revenue being reported is generative AI integration into existing business lines and not new business avenues.
“If you look at the revenues people are posting on generative AI, and the number of use cases, it’s disproportionately higher than the incremental revenue in the industry. [Going by] all the revenue that people say is being generated by AI, it should have created much more growth. The fact is it hasn’t,” Khan told AIM in an exclusive interview.
Generative AI Revenue is AI Mostly Washing
Khan said there are three categories within the realm of generative AI that fall under AI washing. First, there are pure AI use cases where generative AI is genuinely employed. These instances constitute legitimate generative AI revenue.
Second, existing operations, such as call centres integrating chatbots, can now be labelled as generative AI revenue despite generative AI contributing only a portion to the overall customer service revenue.
Third, in business-as-usual scenarios like managing cloud infrastructure, where generative AI enhances productivity through automation, the entire revenue stream can sometimes be portrayed as generative AI revenue, even though the underlying business model remains unchanged.
“The extent to which revenue is attributed to generative AI can vary greatly depending on how broadly the term is defined, leading to what we refer to as AI washing,” Khan said.
Deriving Value from GenAI
Enterprises have derived great value from generative AI. For instance, in software development projects in general, AI is said to make programmers more efficient by 30-35%.
“This is the productivity value. When you come to the second category of net new incremental value, which is not productivity, but driving new revenue, those use cases are just beginning to emerge,” Khan said.
Moreover, Khan believes there is a third category where mature AI-based digital products will emerge, however, those are still a few years away.
“You can actually draw parallels to the internet. Things like online banking and online brokerages took longer to emerge, even though the internet was there for a while. Hence, these types of use cases at scale might be a couple of years away,” Khan said.
Venkata Malapaka, senior director, data and analytics leader at NTT DATA, believes a lot of the tall claims has to do with the FOMO (fear of missing out) factor.
“We need to view everything in context. Very few real, impactful generative AI use cases have actually made it into production,” Malapaka told AIM.
He believes that on the consumer front, we have seen significant progress, as evidenced by applications like ChatGPT and the integration of AI into everyday tools such as WhatsApp and Google Maps.
“However, in the enterprise domain, there are fewer instances of generative AI being applied in actual production use cases,” Malapaka pointed out.
GenAI Projects at NTT Data
NTT DATA, one of the largest IT services companies in the world, made $30.04 billion in revenue in 2023. For the company, generative AI is the largest growing revenue stream.
“The reason for that is the denominator is so small that the percentages are big. I’d like to remind people that AI is a marathon, not a sprint. We are overestimating what it will do in the next three, six, or nine months, but we are greatly underestimating what it will do in the next three, six or nine years,” Khan said.
Currently, around half of NTT DATA’s projects are powered by generative AI in a major way. Whereas around 90% of the projects have been touched by generative AI in some form.
Khan also revealed that generative AI is creating new business opportunities for NTT DATA, but most of them are in the Proof of Concept (POC) stage.
“There are hundreds of use cases with project sizes ranging from a quarter million to half a million dollars. The goal is to scale these initiatives into larger projects worth $10 million to $20 million each.
“While this transformation will take time, GenAI is introducing entirely new types of use cases that were not there in the past,” he pointed out.
NTT DATA’s Foundational Models
NTT DATA is among the few IT companies that is building its own foundational models. Tsuzumi is its lightweight language model fluent in both Japanese and English and has parameter sizes of 600 million and 7 billion.
The ultra-light model is designed in such a way that it could run on a CPU.
“Building our own foundational model offers a sustainability advantage. The compute power required for training and inference is significantly smaller compared to GPT-3.5, while still achieving acceptable or comparable results,” Malapaka said.
While the company has only released the Japanese version of Tsuzumi, the English version is expected to come this summer. Furthermore, Khan revealed that it will take a few more quarters before we see real traction with the model in enterprise use cases.
Nonetheless, NTT DATA is not in the business of developing foundational models. “We are building foundational models not to compete with others in the market, but it is more of an insurance policy,” Khan concluded.
