Conversation
Highlights

Why Private Equity Firms Said Yes to an Unproven Startup (With a <1% Conversion Rate)

Most AI tools flooding into finance in 2023 made the same mistake: build a ChatGPT wrapper, add financial data, pitch it to investors.

Rohan Parikh watched this play out and understood why it wouldn’t work.

In a recent episode of Category Visionaries, Rohan Parikh, Co-Founder and CEO of Keye, explained how his team built AI-powered quantitative analysis for private equity by rejecting the summarization playbook and engineering for deterministic accuracy instead.

Why Search and Summarization Fails in PE

Rohan and his co-founder brought 15 years of combined finance experience—trading, investment banking, Wharton MBA, Vista Equity Partners, Goldman Sachs. They knew PE workflows from the inside.

The first 90 days focused on a single question: where in the workflow could AI create the most value?

“Private equity as an industry has many workflows. So you could be in sourcing, you could be in due diligence, you could be in portfolio monitoring, you could be post deal value creation,” Rohan explained.

They saw tools attempting to bring ChatGPT-style interfaces into PE. Document search. Memorandum summarization. Q&A on deal files.

The problem? “If you look at how private equity professionals work, they are not making decisions on search and summarization. Summaries don’t lead to decision making.”

The actual work happens after initial document review. Investors request detailed financials. Then comes data scrubbing, creating multiple analytical views, combining datasets, extracting insights about customer cohort performance and business trajectory.

“This entire process of scrubbing, creating multiple views, combining those to build your analysis and then extracting insights was something that was extremely critical,” Rohan said. “If you did not go through the same process, first, you will not have the same rigor, second, your probability accuracy is going to be affected.”

Keye found their wedge: quantitative analysis workflow where investors spend the most time making the highest-stakes decisions.

Engineering for Domain-Specific Accuracy Thresholds

Most startups would have applied existing AI models to this problem. Rohan’s team recognized that wouldn’t work.

Financial analysis operates under different constraints than content generation or customer support. Mathematical errors in PE due diligence can mean millions in misvalued deals.

“We say unless it’s 100%, it is 100% wrong,” Rohan stated.

AI models at the time achieved roughly 99% accuracy on math operations. Impressive for most applications. Unusable for PE investors evaluating billion-dollar acquisitions.

“It’s still, you’ll get 99% accuracy, which I think in this industry we say is unless it’s 100%, it is 100% wrong,” Rohan explained.

So they built hybrid architecture: “We needed to build something which was, first of all, 100% accurate, completely deterministic, but we also had flavors of AI.”

Deterministic calculations. AI contextualization layered on top.

This architectural decision was a GTM decision. Investors “have gone through the same process over 50 years at this point in time.” They needed confidence that comes from proven methodology, not probabilistic outputs.

The Design Partner Economics Nobody Talks About

Before Y Combinator, Rohan’s team did something counterintuitive: they sold an unbuilt product.

Not through deception. Through clarity about what they planned to build in six months.

They reached out to PE firms—through connections, through Wharton alumni networks. They showed their vision for the tool.

The conversion rate: less than 1%.

“100 goals and we would hit one client would be like, okay, I’m ready to take a bet on you guys,” Rohan said.

But those rare wins paid $10,000 to $20,000. Three to four paying design partners before significant product development.

The firms’ calculus: “Seem like you guys will do something out of this, but for us it’s $10, $15, $20,000. Who cares?”

For Keye, those customers provided validation, feedback, and proof that someone would pay for their approach—not just express interest.

The lesson isn’t that 1% conversion is acceptable everywhere. It’s that in relationship-driven industries selling to sophisticated buyers, extreme selectivity filters for believers who understand the problem deeply enough to bet on an unproven team.

Outbound Stratification: Beyond Job Titles

Rohan’s cold email approach works because it maps pain to workflow position and fund structure—not biographical details.

“I mean, I do definitely use it, but that’s not the biggest point,” he said about alma mater mentions. “That gets the open rates come from there, but the meetings don’t exactly come from there.”

Instead, Keye segments by role-specific workflow pain:

Associates: “constantly crunching through numbers”

VPs: “trying to look at high-level details”

Principals: managing “different relationships”

But job titles don’t tell the whole story. Fund structures create different buying dynamics.

“Some funds have full committees, other funds are just operated through associates trying to find the right tools,” Rohan explained.

A committee-driven fund requires different messaging than one where associates have tool selection autonomy. The decision-making process, approval requirements, and pain points vary dramatically.

“Understanding the demographic and how funds are operating—what is top of mind for the personas inside—go back to your startup fundamentals: what is the pain point you’re trying to solve for each persona?”

The emails that work demonstrate understanding of specific daily friction, not generic pain points.

Exploiting Second-Order Timing Effects

Keye launched when two conditions converged—one obvious, one subtle.

The obvious: AI technology reached a point where contextualization at scale became feasible.

The subtle: PE firms became receptive to AI tooling because of what they witnessed in their portfolio companies.

“Five years or six years back did not have the same effect in terms of GTM. It was very tough to get any stakeholder’s attention,” Rohan said.

What changed wasn’t just technology maturity. It was that operating partners at PE firms watched their portfolio companies transform with AI firsthand.

“Operating partners are very wary of what AI has brought into the picture just because their portfolio companies themselves have gone through a massive revolution in terms of technology.”

This created unusual openness to internal tooling. “Both technology has been brought in, but at the same time the willingness to look at technology has also drastically improved.”

Keye didn’t just ride AI hype. They timed their entry to when PE decision-makers were experiencing transformation in adjacent contexts—making them receptive to evaluating how similar technology could improve their own workflows.

The GTM implication: sometimes the best time to sell isn’t when your technology is ready, but when your buyers are experiencing related shifts that create receptiveness to change.

Multichannel GTM in Relationship-Driven Markets

Keye’s current go-to-market spans brand recognition, ecosystem association, cold outbound, and word-of-mouth—each serving distinct purposes.

Brand recognition signals data security capability. PE firms handle sensitive financial information. “They want to know that you have the right resources, the right team, the right means in order to protect and have all the measures that can be taken on the data privacy side,” Rohan explained.

Associating with credible ecosystems—alumni networks, Y Combinator—provides trust baseline. “Just getting your name out there and being focused about it. So we just do private equity as of now.”

Cold outbound continues working because of timing and persona-specific messaging. “Outbound has been a really good channel.”

Word-of-mouth compounds in small industries. “This is an industry where it’s a small industry, people know each other,” Rohan noted. Prospective clients ask each other: “Have you guys seen Keye? Pretty interesting concept.”

The multichannel approach reflects an understanding that in conservative, relationship-driven industries, no single channel builds sufficient trust. Brand signals capability. Outbound demonstrates understanding. Word-of-mouth provides social proof. Together, they create conditions for deals to close.

The Vision: Compressing Time to “No”

Keye’s long-term vision anchors to the original pain point: PE firms want to evaluate more deals without sacrificing selectivity.

The constraint isn’t finding opportunities. It’s having resources to properly analyze them.

Rohan frames Keye’s value as “the tech layer which helps you to say no to a deal earlier in the process so you can do more deals as a result.”

Not automating deal selection. Not replacing investor judgment. Compressing the timeline to “no” so firms can focus resources on legitimate opportunities.

“The pain point is they’re not being able to do more deals and they want to look at more deals, hopefully with the same selectivity, do more deals,” Rohan said. “And so pain point remains the same. The vision is exactly helping that pain point.”

This framing shapes product priorities. Features that help investors reject bad deals faster take precedence over features that marginally improve good deal evaluation.

For B2B founders building in conservative industries, Keye’s approach offers a blueprint: understand workflows at granular detail, engineer for domain-specific reliability thresholds, time market entry to second-order receptiveness signals, segment outbound by workflow position and organizational structure, and frame your value proposition around the constraint that actually limits your buyers—not the problem you’re technically capable of solving.