Rohan Parikh Is Teaching AI to Show Its Work

A spreadsheet can be a confession. Its cells reveal what the analyst assumed, what the source data withheld, which formula carried the argument and where confidence quietly became guesswork. Rohan Parikh knows this language because he spent years speaking it. Before he was the co-founder and CEO of Keye, he moved through credit analysis, consulting, research, emerging-markets trading and investment banking. The job titles changed. The obligation did not: make the number explain itself.

That obligation now sits at the center of his software company. Keye builds diligence tools for private-equity investors, the people asked to inspect a business quickly enough to win a competitive process and carefully enough to avoid a costly mistake. Its product ingests raw deal files, structures the data and produces analysis that investors can audit and edit. Artificial intelligence helps understand the question and the context. Code handles the calculation.

It is a division of labor with an almost old-fashioned charm. Let the machine be conversational, but make the arithmetic wear a name tag. In January 2026, Keye gave that approach a Norse name: Odin, a copilot that accepts questions in plain English and executes the analysis through transparent logic and formulas. Parikh calls the underlying idea “Deterministic Creativity.” The phrase contains his whole argument. Discovery can roam. The answer still has to come home.

“Unless it's 100%, it is 100% wrong.”Rohan Parikh on financial-analysis accuracy

01 / Before the formulasFour languages, a football, a stage

Parikh grew up in Mumbai in a business-oriented family that spoke four languages. Long before he was presenting software to investors, he was collecting other kinds of rooms. He represented his school, district and state in soccer competitions. He performed on Mumbai theater stages as a young artist. At the University of Mumbai, where he studied electronics and telecommunications engineering, he rose from cultural-event volunteer to cultural secretary.

Those details resist the tidy founder myth of the child already rehearsing a pitch deck. They show range instead. A team sport taught coordination. Theater demanded timing and attention. Engineering made systems legible. Meanwhile, managing his father's stock portfolio drew him toward finance. He earned CFA and FRM credentials and pursued a master's degree in mathematical finance at Boston University. The trajectory was not a rejection of engineering. It was engineering applied to money, with consequences attached.

His early work moved through CRISIL, where he analyzed credit, and Accenture, where he built Excel-based financial reports. A research internship at State Street followed. Then came Natixis CIB Americas. Across global markets and investment banking, Parikh worked on emerging-market products and climbed to Director. In that role, he helped drive a sevenfold increase in deal flow and revenue. More importantly for the product he would later build, he saw the repetitive machinery beneath high-stakes decisions.

Investment work can look cinematic from the outside: a decisive committee, a signed term sheet, a celebratory dinner. Inside, conviction is assembled from far less glamorous material. Raw files arrive in inconsistent shapes. Analysts clean data, reconcile tables, build customer cohorts, test retention and chase anomalies through sprawling workbooks. The polished presentation waits at the end. The late night lives near the beginning.

02 / The first keyA startup takes the scenic route

At Wharton and the Lauder Institute, where Parikh completed a dual MBA and MA between 2021 and 2023, entrepreneurship became less of an abstraction. His original Keye was a different company. It aimed to relieve subscription fatigue by letting people buy credits and spend them across paywalled services. The irritation was personal: Parikh had found himself using fresh email addresses for trials and paying for products he opened only a few times.

That version won a 2022 Jacobson Venture Award, joined Penn's VIP-X accelerator and reached the university's Startup Challenge semifinals. The founders had partnerships and an argument about access. Yet the Keye that exists today is focused on private equity, with a founding team that includes CTO Lalit Lal and COO Conor Brown. The name survived while the problem changed.

A pivot can make the past look disposable. Here, it reads more like apprenticeship. The first Keye taught Parikh how to seek feedback, use an entrepreneurial community and test willingness to pay. He said a central lesson from Penn's Venture Lab was not to fear reaching out. The second Keye would require all of that patience, now aimed at customers who guard both their data and their professional reputations.

03 / Selling skepticismOne yes in a hundred

Before Y Combinator, Parikh and his team tried to sell. He has described a hit rate below one percent: about one design partner for every hundred outreach attempts. Those first believers paid roughly $10,000 to $20,000. For a product entering a conservative, relationship-driven industry, rejection was not a surprise. Trust was the actual inventory.

The team learned to make outreach about workflow rather than biography. An associate cleaning data has a different anxiety from a vice president synthesizing a case or a principal managing the deal team. Funds also distribute authority differently. Some choose tools through committees. Others give junior investors room to experiment. Keye's message became narrower as its understanding grew.

This is the practical piece other founders can borrow. A tiny conversion rate can conceal a useful market when the problem is expensive, the contract is meaningful and early customers actively shape the product. The customer who says yes is not merely a logo. In Keye's case, some users later became investors. By July 2025, when the company emerged from stealth, it said the platform had been deployed across more than 20 funds representing over $1.4 trillion in assets under management.

The company announced a $5 million seed round led by a group that included Sorenson Capital, General Catalyst, Y Combinator, Entrepreneurs Roundtable Accelerator and other firms and angels. The financing mattered, but the customer-investor overlap made the more interesting statement. People who had seen the product inside a live workflow were willing to back the company behind it.

04 / The determinism wallWhy 99 percent fails the meeting

Generative AI thrives on plausible variation. Financial analysis does not. Ask a language model to summarize a memo twice and two elegant answers may both be useful. Ask it to calculate customer retention twice and two answers are a problem, even when one is correct. Parikh refers to this boundary as the determinism wall.

Keye's architecture responds by separating intent from execution. AI interprets what an investor is asking. Auditable code and formulas perform the analysis. The resulting work can be traced to source data and exported into Excel, where investors can inspect assumptions and continue the model. The familiar spreadsheet remains because its openness is part of the trust mechanism. A visible formula is an argument another person can challenge.

This approach also clarifies what Parikh means when he says summaries do not lead to decision-making. A deal team rarely needs another neat paragraph about a company it is considering buying. It needs to know whether growth depends on a handful of customers, whether retention improved broadly or through a few large accounts, whether margins hide an anomaly and how the conclusion changes when an assumption moves. The output must invite interrogation.

“AI shouldn't replace judgment. But it should compound it.”Rohan Parikh at the launch of Odin

05 / What remains humanThe autonomous team, with an editor

Parikh's larger ambition is what Keye calls the Autonomous Deal Team: coordinated agents examining financial, commercial and operational questions in parallel, with an investor orchestrating the work. The phrase can sound like an empty conference room. His description places the human at the console instead.

That distinction suits the path that brought him here. Trading taught him that speed matters. Banking taught him that defensibility matters. Engineering offered structure. Theater, perhaps, supplied the instinct that a performance should hold together for the audience. Keye tries to combine those demands without pretending they are identical.

The useful destination for those saved hours is attention: testing the strange line item, asking the impolite question, noticing that management's story and the raw data have stopped agreeing. Automation earns its place when it gives skepticism more room to work.

There is a modest lesson beneath the grand language of artificial intelligence. Before software can transform an industry, it has to respect the reason an old workflow endured. Excel survives in finance because its cells can be opened, its logic examined and its mistakes found. Parikh is not asking investors to abandon that habit. He is trying to move the drudgery out of its way.

The key, in other words, is not an answer that arrives looking certain. It is an answer built to withstand doubt.