Key Takeaways

  • Less than 1% of enterprise data has been absorbed by large language models, explaining the massive gap between corporate experimentation and working production systems.
  • Vista Equity Partners instituted a portfolio mandate requiring every company to build a working agentic angle before going to market.
  • In a recent Vista exit where revenue grew 3.5x and EBITDA grew 5x, the buyer's primary interest was acquiring agentic capacity to retrofit across its own operations.
  • Enterprise data security requires bringing AI models to proprietary data behind corporate firewalls, rather than feeding proprietary data into external foundation models.

The One Percent Data Trap

Robert Smith sees a clear structural reason why corporate AI rollouts stall in pilot phases. Foundation models have absorbed almost none of the private data that runs corporate systems.

“Less than one percent of the enterprise data can be absorbed by these large language models,” Smith told Hugh MacArthur. “And as a result of that, you actually have a very high level of use, but very low level of, call it, fully scaled enterprise or agentic solutions, right across the board.”

When software vendors try pushing public foundation models onto enterprise stacks, corporate leadership stalls the process. CIOs fear proprietary trade secrets, customer records, and operational logs leaking into general model weights. Enterprise buyers refuse to risk their core data assets training external systems.

Bringing Models to the Data

The fix is an architectural reversal. Instead of shipping terabytes of sensitive records to third-party endpoints, companies must pull foundation models into secure, private environments.

“Look, the CEOs that I spend time with all know the theoretical power of what this new technology can bring,” Smith noted. “Their challenge is how do they do it in a way that not only is reliable, but doesn't create a problem in taking their data and leaching their data out into these foundational models or these LLMs.”

The technical baseline is straightforward: “You've gotta take these models to the data, not the data to the model.” By running specialized weights directly against isolated databases, portfolio assets retain their proprietary edge while enabling automated execution loops.

Why Strategic Acquirers Buy Working Agents

Financial execution and revenue growth still matter, but strategic buyers now evaluate software targets through an operational automation lens. Vista experienced this firsthand on a recent portfolio realization.

“We just exited one of our companies,” Smith said. “One of the reasons, beyond the fact that we grew the revenue more than three and a half times and EBITDA more than five times since we bought it, those are all great things, what they really liked about it was they saw the agentic capacity of what we had built in that business and saw how it, in essence, can affect the rest of its business.”

Strategic buyers no longer evaluate SaaS assets purely on recurring revenue expansion or cross-selling opportunities. They want working autonomous workflows that can be lifted and dropped across their existing corporate infrastructure.

Why It Matters

Strategic acquirers are treating mature agentic workflows as an operational acquisition target rather than standard product expansion. Sponsors who isolate proprietary data architectures and deploy functional agents are creating an exit valuation arbitrage that traditional margin expansion alone cannot match.