Sirion Completes Majority Investment from Haveli, Aiming to Accelerate AI Push in CLM Market

Sirion, an AI-native contract lifecycle management platform, has completed a majority investment from Austin-based private equity firm Haveli Investments, the companies announced today.

With Haveli’s partnership, Sirion said, it will aim to accelerate product innovation, expand its global go-to-market presence, and enable organizations to move from static contract repositories to intelligent, workflow-driven contracting.

“Sirion is building AI that works directly inside enterprise contracting workflows, helping teams draft, negotiate and manage agreements faster and with greater confidence,” said founder and CEO Ajay Agrawal.

The deal, first disclosed in January, gives Haveli a controlling stake in the company and clears out Sirion’s earlier investors, including Sequoia and Tiger Global. Agrawal, in an exclusive interview with LawNext, describes the deal as not a traditional fundraise, but a strategic recapitalization.

“What it’s not is a buyout or a kind of a huge strategic fundraise,” Agrawal told me in an interview Friday. “The company was already profitable before this transaction. We had more than half the money from our previous round still sitting on our balance sheet.”

Instead, Agrawal explained, the deal was driven by a need for board-level simplification. With investors from different eras – some on the cap table for more than 11 years – Sirion was navigating competing interests and motivations at the board level.

Haveli’s entry resolves that by buying out the earlier investors entirely, leaving the company with what Agrawal calls “a cohesive, unified voice at the board level at a time when corporate strategy and business strategy need to keep pace with the developments on the frontier model side.”

Who Is Haveli?

Haveli may not be a well-known investment firm in legal tech, but Agrawal made clear he considers it a major get. The firm was founded in 2021 by Brian Sheth, who previously cofounded Vista Equity Partners.

Haveli’s debut fund, Agrawal said, is the largest first-time software-only private equity fund ever raised – a $4.5 billion vehicle – and the firm is highly selective about its portfolio companies, bringing with it a value-creation playbook developed across more than 400 software companies.

Haveli has not previously invested directly in legal tech, although it has invested in legal-adjacent companies such as document management company M-Files and ERP platform Certinia. That is fitting, because, in our interview, Agrawal himself pushed back on the legal tech label for his own company.

“I don’t really see CLM as a legal tech category,” he said. “Ninety percent of the funding comes from a combination of procurement, sales operations and IT, with only 10% coming directly from legal.”

While he acknowledged that “down-market” CLM players, as he described them, such as Ironclad, Agiloft and ContractPodAi, engage heavily with legal buyers, he said that “the moment you move up market into the enterprise segment, legal is an interested party, but not the one with the cash.”

A Different Kind of PE Deal

Agrawal was emphatic that this is not a typical private equity transaction.

“This is not your typical private equity investment, where the playbook is to come in and cut cost,” he said. “The first thing they said to us is, ‘How can we expand our product leadership in this category? What are the investments that we need to make?’”

For Sirion’s customers, the deal should be a positive signal, Agrawal said. While daily operations will not change, Haveli’s backing will accelerate Sirion’s generative AI roadmap, including investments in GPU infrastructure and expanded R&D, Agrawal said.

On the growth front, Agrawal said the company has been expanding at nearly 50% annually and is profitable.

The transaction also creates a liquidity event for employees with stock options, something Agrawal said is becoming a “Sirion tradition,” in that this is the second such employee liquidity opportunity since the company was founded in 2012.

 The Platform’s AI Architecture

The last time I had an in-depth conversation with Agrawal was in 2022, when Sirion raised an $85 million Series D round, and he was a guest on my LawNext podcast in 2020, not long after the company had raised a $44 million Series C round.

In our conversation Friday, we talked about how Sirion’s AI approach has evolved since then – and what Agrawal sees as his company’s enduring competitive edge.

He traced the evolution in three stages. When we spoke three years ago, Sirion was using small- and medium-language models to extract structured data from contracts and flag deviations from playbooks – what the company called “contextual deviation.”

With the arrival of GPT-3.5, the platform moved from identifying problems to explaining them in plain English – what Agrawal called “issue detection and issue summarization.” The third phase, enabled by GPT-4 and subsequent models, was auto-redlining – the ability to mark up contracts much like a junior attorney.

Today, Sirion is emphasizing what Agrawal calls “conversational search” – allowing users to query their entire contract portfolio in natural language.

But he was careful to emphasize that the key underlying engine is not the large language model itself. Rather, it is a layer of what he calls “template imprinting micro APIs” that translate a user’s plain-English query into the specific language of a company’s SQL database.

Ask for all contracts expiring in southern Europe without an ESG clause, and the system can figure out that “southern Europe” is not a category in the database, but Italy is, so it converts the query accordingly, showing its work in the chain of thought so the user can understand.

Agrawal said that this architecture reflects a core conviction he has held for years, which is that LLMs alone cannot be trusted for high-stakes contracting decisions.

“LLMs today are probabilistic in their reasoning process, not deterministic,” he said. “You can’t pay payroll saying Bob will get his salary 80% of the time.”

The ‘Workhorse of Enterprise Contracting’

For that reason, he argues that sub-million-parameter, highly deterministic small language models will remain the workhorse of enterprise contracting for the foreseeable future. He further believes that competitors who skipped the decade-long work of building those models will not be able to close the gap simply by plugging in the latest frontier model.

Agrawal pointed to Sirion’s 2024 acquisition of Eigen, a leader in intelligent document processing, as a key expansion of this capability.

The Eigen technology allows Sirion to read engineering reports, service-level backup data, and other non-contract documents, then connect them back to contract obligations. That enables Sirion to flag, for instance, when a supplier’s documented underperformance means a client is being overbilled.

On the acquisition front, Agrawal signaled interest but patience. He sees the CLM market heading toward a reckoning, with some players ending up on “the wrong side of AI history” and becoming available at steep discounts.

At least for the near term, his preference is for what he calls “tech tuck-ins” – targeted acquisitions of interesting AI technology, similar to its acquisitions of Eigen and Zendoc, which it acquired in 2022 – rather than for large revenue-driven M&As.

On the Foundation Model Threat

With the topic much in the news of late as to whether foundational AI companies such as Anthropic or OpenAI represent competitive threats as they push further into enterprise applications, Agrawal does not think they do, at least not in any lasting way.

His view is that foundation model providers will increasingly function as infrastructure, with economic value flowing upward to the application layer.

“They’re going to become like a lasagna layer,” he said. “The capture of their economic rent will progressively decline and become asymptotic over time, and the bulk of the economic rent will go to the applications that sit on top.”

He sees Sirion’s data advantage – the company has parsed tens of millions of contracts, approaching 100 million – as a moat that no general-purpose model can replicate.

“They have access to a very thin slice of enterprise data, only 1%,” he said of the foundation model players.