AI funding boom pulls Big Four deeper into startup diligence

“Investor due diligence has become deeper over the years as we raised funding,” said Bhinge, whose venture has raised over $60 million since inception. “There is still a bit of analysis paralysis even now.”

That is widening the circle of people pulled in to conduct diligence, from outside specialists to the Big Four advisory firms such as PwC and Deloitte. While they were always part of deal scrutiny, it was mostly for financial diligence, tech and cybersecurity reviews. AI has changed that playbook. Investors also want to know how the data is sourced, how models were trained, how reliable the outputs are, and whether customers are seeing real value.

Increased diligence coincides with growing investor interest in AI startups, as the technology threatens to disrupt traditional models and the software-as-a-service model. Indian AI ventures raised $832 million in 2025, with another $633 million flowing in during the first quarter of 2026 alone, according to Tracxn data.

Blackstone in February led a more than $1 billion investment in Neysa, while Emergent raised $70 million a month earlier. Sarvam AI is also reportedly in talks for a $200-250 million round from investors including Nvidia Corp, Bessemer Venture Partners and HCL Technologies Ltd.

“AI has shifted venture diligence away from just asking typical SaaS-era focus go-to-market efficiency questions to more towards a deeper focus on product and engineering capability,” said Krishna Mehra, an AI partner at Elevation Capital.

For Bharat Innovation Fund (BIF), which has backed deep-tech and AI startups since 2018, the main challenge is not just that AI startups require a new technical lens, but their numbers have surged.

BIF has screened more than 6,000 AI startups, narrowed that to over 500, and held deeper discussions with roughly 300-400 of them, said Somshubhro Choudhury, co-founder and partner at the fund. He said diligence has become harder because investors are struggling to tell genuinely differentiated products from thin wrappers built on existing AI models.

Choudhury said founders are now asked to explain the full stack, how workflows are split into tasks, which models are being called and how the system has been trained. “We ask them to take us through the entire (code) stack.”

The fund brings in chief technology officers (CTOs) from portfolio companies and other experts from its network to pressure-test those claims before investing.

In practice, that means checking who owns the data, whether it was acquired legitimately, whether consent and privacy safeguards are in place, how reliable and fair the outputs are, and who is liable if the AI fails.

“This kind of work is more common in private equity than in early-stage venture capital, because PE firms usually enter once a company has customers, revenue and enough operating history to test,” according to Vishwanath.

Strong AI products coexisting with weak governance show up “very often”, especially when a PE investor is evaluating a company for the first time after earlier rounds of capital, said Vishwanath. He said roughly 70-80% of PwC’s reports in such AI-linked PE deals carry recommendations across governance, privacy, cybersecurity, compliance and team capability.

Rohit Madan, who leads Deloitte’s forensic due diligence practice, has scrutinized more than 30 startups over the past 18-24 months, giving the firm a bird's-eye view of how AI startup claims hold up under scrutiny.

After initial founder checks, he said, the review moves to business-model traction, clientele review and whether customers are genuinely using the product as claimed. For Madan, one of the biggest diligence risks lies in the data layer: investors need to know whether the data was legitimately sourced, licensed, or lawfully acquired, rather than scraped or pulled from protected databases in ways that could later trigger disputes or litigation.

Jayant Saran, partner at Deloitte who leads forensic technology, has seen financial services emerge as a relatively easier AI category because the sector combines deep domain expertise with long experience in handling large data sets.

Healthcare and similar sectors, by contrast, are harder to underwrite because models often need to be adapted across countries, populations and systems, raising questions around data quality, scalability and real-world applicability. That makes the diligence burden heavier even when the product thesis appears strong.

Another growing concern is deployment economics. Abhishek Srivastava, general partner at Kae Capital, put it bluntly: “An AI product is expensive to run.” Investors increasingly have to ask whether those economics still make sense once a product moves beyond the demo stage and whether founders are actually building lower-cost layers of their own over time, he said.

For Infinite Uptime, this increased diligence shaped its journey from an early pitch around digitising factory equipment to creating a real AI moat. And according to Raunak Bhinge, “It’s a lot of learning from both sides–VCs as well as founders.”