In June, New York City hosted the largest tech week in history with over 1,500 events and 50,000 attendees participating throughout the week. One of the highlights was a forum on data-driven dealmaking organized by 1BusinessWorld , which convened technology investment bankers, tech investors, M&A lawyers and diligence providers working in AI and data-driven dealmaking. These were my three takeaways on how AI is being used across the deal lifecycle.

AI Assessments: A New Standard for Due Diligence

In Bain’s 2026 M&A Report that surveyed 300 dealmakers, one in five strategic acquirers said they had walked away from a deal because of the anticipated impact of AI on the target's business. Nearly half of all technology deals now have an AI product component. That is why PwC now counts AI risk among the core items of any pre-acquisition review. For us at Union Square Advisors and for many of the largest technology investors, AI assessments have become the new standard for due diligence.

We work with investors across venture, growth equity, private equity and private credit, and one question now precedes every investment decision: how does AI change the value and defensibility of this asset? That single question opens into an ever-expanding set of scenarios, each raising further diligence questions of its own. Having run AI assessments on more than 400 companies over the past two years, we have found that those questions can be distilled into a repeatable analytical process.

At Union Square Advisors, our AI Assessment program draws on more than 2,000 data sources for each company, from patent filings to customer reviews to analyst reports , as well as on our firm’s live market intelligence regarding the technology M&A market as it trades today. Together, that combination answers the three questions every technology investor is asking: which companies to select, when is the best time to transact, and what are the best paths to monetizing the asset.

Early-Stage Investing and Data Driven Investing

A mature technology company can be evaluated against a public record the market has already produced. At pre-seed and seed, however, that record does not exist yet, and today's earliest-stage companies are growing faster than any cohort before them.

For the past nine months the entire Y Combinator batch has compounded 10% a week , a rate without precedent in early-stage venture . Roughly a quarter wrote 95% of their code with AI, and some crossed $10 million in revenue with teams of fewer than ten. Nearly half of a recent batch were AI-agent companies , and more form every quarter. These are the companies we want to track early, and they are precisely the ones with no record to read.

Across the earliest stage of investing, some venture firms are using AI to crowdsource data from networks of operators and using AI to power data analysis. Orange Collective , a VC fund backed by 150+ Y Combinator grads and exclusively investing in AI companies from Y Combinator, is one clear example. Orange Collective runs AI agents on Firecrawl to scrape each company’s profile the moment it appears, sends its network a daily AI-written digest of the entire batch, and draws on more than 150 YC alumni as limited partners who grade each cohort from the inside . It assembles a scored dataset of founder, traction and team signals to rank every company a Y Combinator batch produces, the kind of proprietary data that exists nowhere else. Then the dataset and its benefits compound: every batch graded, every signal checked against what the company became, sharpens the next call, and the advantage increases through time.

Diligence Providers and AI Systems

The advisory firms that handle legal, financial and tax diligence are also rebuilding their work on AI and data-driven systems.

M&A lawyers are rewriting the terms that govern the purchase of an AI acquisition target. Buyers now demand representations that a company's models have been trained on lawfully licensed data, built with disclosed open-source components, and deliver explainable, bias-tested, reproducible outputs . Because an AI company's value is associated with its models and datasets rather than just its code , sellers carry increased, dedicated liability caps and a broader range of representations and indemnities , with longer survival periods.

Even the deal insurers are pulling back. On most transactions, reps-and-warranties insurance is used to support the seller's various promises about its business, ultimately paying out to the buyer if one or more of these promises turns out to be false. But as law firms like Skadden have noted , insurers are now studying AI deals closely enough that they may exclude the two hardest promises to price – i.e., that the training data was lawfully obtained and that the models keep performing – leaving the buyer to catch those risks in diligence or carry them alone after closing.

As for financial, tax, and operational diligence, Deloitte is now deploying Claude to its 470,000 employees , Anthropic’s largest enterprise rollout, in versions built for the accountants who run that diligence and the developers who build its tools. KPMG put Claude in front of its 276,000 staff and became Anthropic's preferred partner for private equity .

Humans in the Loop, Humans in Charge

The broad consensus across the forum was that these data-driven systems work best when they are built to keep humans responsible for the work and accountable for the outcome.

After all, dealmaking is a services business, whether the work is sourcing, diligence or execution. Ultimately, accountability and responsibility for the work rest on the shoulders of the people doing it, not the systems assisting them.