Technology

Why the Best Software Does Not Always Win Big Company Deals

Martin HollowayPublished 3w ago4 min readBased on 4 sources
Reading level
Why the Best Software Does Not Always Win Big Company Deals
source:databricks.com

Arsalan Tavakoli-Shiraji, a co-founder and senior vice president of field engineering at Databricks, told an audience at TechCrunch Disrupt that startups selling to large companies often assume the best technology wins. He argued that people and processes matter more. The remarks were reported by TechCrunch on October 26, 2026.

Databricks helps businesses store and analyze large amounts of data. According to the same report, the company is valued at $190 billion, and more than 70% of the Fortune 500 are Databricks customers. The Fortune 500 is a yearly list of the 500 largest companies in the United States.

Those numbers describe a company that grew through sales to big organizations, not only through individual engineers choosing its tools.

Tavakoli-Shiraji said Databricks first tried and failed to get a meeting with UBS, the Swiss bank. The company's route into large accounts changed when it became an Azure cloud partner through a partnership with Microsoft. Azure is Microsoft's cloud platform, where businesses rent computing power over the internet. Sales agreements were covered under Microsoft's terms. After that, Microsoft's sales staff booked meetings for Databricks and backed the startup inside their own customer relationships.

A listing on a cloud marketplace works a bit like a shop in a large shopping mall. Its name in the directory does not bring in shoppers by itself. Visitors arrive when someone from the mall walks them over and vouches for the shop.

When large companies raise objections to AI projects, Tavakoli-Shiraji sorts them into four groups. He calls them the four Cs: context, choice, cost, and control. Each one matches a different worry. They cover how an AI model is grounded in a company's private data, who decides which models and vendors get used, and what the ongoing running costs look like once the work grows.

The framework shifts the conversation away from benchmark scores, the standard tests used to compare AI systems. It moves it toward the questions a chief information officer, the executive in charge of a company's technology, and that person's purchasing and security teams will ask.

Databricks also works with startups, including Hightouch, Addepar and Tecton, by selling their products under Databricks' own terms and cybersecurity setup. That lets a small vendor pass through the security review that a large bank or insurer would otherwise require it to clear separately.

Databricks later acquired Tecton, a machine learning startup. The partnership ultimately became an ownership relationship.

For context, Matei Zaharia, Databricks co-founder and chief technology officer, received the 2026 ACM Prize in Computing. TechCrunch reported the recognition on April 8, 2026. In May 2026, TechCrunch Disrupt also hosted a session with Tavakoli-Shiraji on what kills enterprise AI deals, a topic the October remarks return to.

In my view, the most transferable point is about timing. Many infrastructure startups build for the engineer who tries a tool first. Then they learn that the person who signs the contract works on a different schedule, with different levels of risk tolerance and a different chain of approvals. As Tavakoli-Shiraji described the UBS episode, a closed door at a large institution may say little about the product and a great deal about the way in.

The broader context here is that buying enterprise AI now sits where data governance, model choice and cloud contracts meet. Vendors that can reduce friction in those areas, through partnerships with large cloud providers, packaged security controls, or acquisitions that bring a capability in-house, may find the road to real use is shorter than a feature comparison suggests. Whether that holds across the market is an open question.