Errors, high cost among reasons GenAI not moving beyond concept stage

Senior industry executives and analysts say a high error rate, steep cost of deployment and the relative newness of the technology are the main reasons for GenAI’s weak adoption by companies.

“So first and foremost, accuracy of the results is critically important. And second, the adoption of that solution is critically important. And the adoption only takes place if the accuracy levels are high enough that they’re providing you with a distinct competitive advantage,” said Rohit Kapoor, chief executive officer of ExlService Holdings Inc., in its post-earnings interaction with analysts in May 2024.

The New York-based ExlService is a Nasdaq-listed BPO company which ended the three months through June 2024 with $448 million in revenue, up 2.7% from the preceding three months. Its revenue for the year ended December 2023 totalled $1.6 billion, 15.5% higher than in 2022.

Kapoor, who is among the first CEOs to highlight the high error rate in GenAI-led solutions, said the goal was to improve their accuracy.

“So when we start with a new solution, the accuracy levels are very low. They are like 65% to 70%, but we have to drive that up to 90% to 95%, and that is driven by a strong understanding of the domain and understanding of how and what the data is telling us, and therefore, the fine-tuning of our algorithms and our ability to embed this into the workflow so that it is actually useful for every single person that is adopting these solutions,” said Kapoor.

GenAI made waves after the launch of ChatGPT in November 2022 catapulted the new technology into boardroom discussions of the largest companies in the world. It seemed its widespread adoption was a foregone conclusion.

Nearly two years after, GenAI hasn’t quite lived up to its billing.

In an interview with Mint on 12 July, K. Krithivasan, CEO of India’s largest IT services company, Tata Consultancy Services Ltd (TCS), said that delivering differentiations to clients is key, instead of riding a hype cycle. However, the road to adoption is still hazy.

GenAI in outsourcing firms

Experts have put a question mark on the adoption of GenAI by outsourcing companies as most of the tests conducted by these companies have not gone into the production stage.

“According to industry feedback, more than 85% of proof-of-concepts around generative AI have failed to move into production. We think this is contributing to a disparity between generative AI investment in the hardware and training front versus actual revenue generation by IT service providers and software application providers,” read a note by BMO Capital Markets analyst Keith Bachman dated 31 July.

IT firms’ AI shift

Most IT services companies in their latest earnings commentary have said that they have trained their employees on AI and GenAI, without delving deeper into those training modules.

Gartner has said that almost a third of GenAI projects would not see the light of the day beyond the proof-of-concept stage.

“At least 30% of generative AI (GenAI) projects will be abandoned after proof-of-concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value,” the IT research and consulting firm said on 29 July.

While accuracy is a concern, stakeholders are wary of the novelty of the new technology. Analysts said that enterprises were uncertain on actual adoption of GenAI because of its relative newness.

“Only 5% of GenAI has moved into production in major enterprises, but this will increase significantly as real business cases emerge. Remember, we are barely 18 months into this and enterprises are still figuring out where it adds most impact and value,” said Phil Fersht, CEO of US-based HFS Research, an outsourcing research firm.

Also, experts said the cost of deployment was high for GenAI adoption to be immediate.

“For commercialisation, GenAI proof-of-concepts have to be integrated into the IT infrastructure of the outsourcing companies. That requires multiple building blocks to be in place like accurate data layering. Then, there is also a cost of implementation, which will require outsourcing companies to earmark a significant cost towards this integration. Adoption will take a couple of years,” said an industry expert working at a BPO company on condition of anonymity.

Mint had reported on 4 July, citing analysts, that GenAI posed a threat to BPO units as the new technology, which can create content in audio, visual, or written form by means of a prompt, can replace routine human work. BPOs offer customer support to their clients, much like a call-centre operates.

In fact, global IT services companies, including Accenture Plc, and Cognizant Technology Solutions Corp., have highlighted GenAI as a risk in their annual reports. 

“We increasingly use AI-based technologies, including GenAI, in our client offerings and our own internal operations. As with many innovations, AI presents risks and challenges that could adversely impact our business. The development, adoption, and use of AI technologies are still in their early stages and ineffective or inadequate AI development or deployment practices by us, our clients, or third parties with whom we do business could result in unintended consequences,” said Cognizant in its annual report for the year ended December 2023.

Cognizant follows a January – December financial calendar whereas Indian IT services companies follow a April – March financial calendar.

Caution towards GenAI

The caution towards GenAI was no different at home.

“GenAI, large language models and Metaverse could lead to legal liabilities through plagiarism, deep fakes and privacy and copyright infringement issues. The efficacy of AI models depends on the quality of the data they are trained on. Accuracy, bias/fairness risks could cause reputational damage and legal liabilities. GenAI technologies could disrupt software development and testing activities, changing customer expectations in the short term,” read TCS’s annual report for the year ended 31 March.

TCS’s Krithivasan said that clients are taking a calibrated approach while charting their GenAI adoption roadmap.

“While the interest is strong, organizations are taking a calibrated approach to measure the risk potential and organizational impact while chalking out their roadmap for GenAI adoption,” said Krithivasan during the company post-earnings interaction with analysts on 11 July.

Yet, notwithstanding the current slow pace of GenAI adoption, experts said the technology is bound to replace basic human functions in areas of customer experience.

“The main work being disrupted is in areas such as customer experience, marketing and IT development. Existing workforce will be reduced by 20-50%, and those remaining trained to use the tools,” said Fersht.

The BPO industry expert cited earlier said humans will always be a part of the loop to deal with complex queries even as chatbots replace basic customer experience functions.

“What companies have started doing is that they have started to integrate GenAI into a human-centric process. It is now human plus AI which is improving the efficiency of the person involved as well as the process. That is the stage we are in where it will never be a binary move. Humans will always be in the loop,” the expert added.