Startups have never been under more pressure to grow quickly. To do that, many of them must land large enterprises, the deep-pockets of the IT world, as customers. But as any company that sells to enterprises will tell you, it’s not easy to win such contracts.

One critical mistake most younger startups make is believing that their tech will practically sell itself, says Arsalan Tavakoli-Shiraji, co-founder and senior vice president of field engineering at Databricks, which helps businesses store and analyze huge amounts of data.

Yes. The tech must work well. But that’s not enough, he warns. Databricks today is an AI and big data storage giant, valued at $190 billion, with more than 70% of the Fortune 500 as customers. But it was once a tiny startup, and Tavakoli-Shiraji was responsible for helping bring in its first big deals.

“Everybody when they’re selling to enterprises at startups … we think it’s all about the tech and having the best tech,” Tavakoli-Shiraji told attendees at TechCrunch Disrupt last week. But, he learned the hard way that “people and processes matter the most.”He remembers securing a meeting with a prospect in the early days, only to discover that it was the same person that signed a mega deal with his major competitor. “He has no interest in listening to me because he bet his career on that competitor,” Tavakoli-Shiraji said.

He also remembers early deals fizzling because the startup wasn’t prepared for legal, contractual, or security questions. He offered his stop tips for startups to win big customers.

Sell under a bigger tech company’s umbrella When Databricks was young, it tried and failed to score a meeting with UBS, the large Swiss bank. Then it did “a partnership with Microsoft,” becoming an Azure cloud partner. That meant sales agreements would be covered under Microsoft’s terms.

But it also meant that Microsoft’s sales folks actually booked the meetings and advocated for the startup. “All of a sudden, that meeting that wasn’t hard to get,” the founder recalled. “It was a huge boon to us that we were able to partner with somebody who had that access and distribution.”With Microsoft’s sales team helping, “we started getting pulled everywhere.”That created new problems, like prospects wanting features for countries Databricks hadn’t expanded to, but they were good problems to have.

Today, Databricks is the big company that works with startups, selling their products under Databricks’ own terms and cybersecurity setup. Examples include Hightouch, which provides AI-powered marketing and customer-data tools (Databricks has also invested in it); Addepar, which offers data tools for investment professionals; and Tecton, a machine learning startup that Databricks later acquired. When selling AI, address the “four C’s”Tavakoli-Shiraji groups enterprises’ main concerns and objections into four categories he calls the four Cs: context, choice, cost, and control. Context means an AI that can be trained “like bringing on a new employee,” he says, learning the company’s terminology, processes, priorities, and so on.

Choice is the freedom to switch AI models rather than be tied to a single provider. Cost is about controlling how much the AI product will cost to run. Control involves the safeguards that govern the AI’s behavior and protecting sensitive information. Attack the expensive legacy systems enterprises think they’re stuck with Tavakoli-Shiraji jokingly asks CIOs where they stand with AI on a scale of one to five. A one means employees merely have access to ChatGPT, and a five means the company has adopted new AI capabilities that have changed its business processes.

“And everybody loves picking, ‘I’m three leaning towards four.’ But 99% of the ones I talk to are in bucket one and two.”Enterprises still haven’t figured out how to deploy AI agents in ways that create value beyond allowing engineering teams write code. “It’s going to take many, many years before people can actually properly deploy AI in a very meaningful way in the enterprise.”

One big opportunity for startups is to bring that value is to modernize expensive legacy systems. “If you go to any bank today, they still have [an IBM] AS/400 and mainframes,” he described. “Nobody wanted to touch it because nobody knew what that code did.”But today, AI agents “can go in and understand and actually decompose what that code does.”That, he says, opens the door to products that do everything from adding modern interfaces for those ancient systems to replacing them altogether.

Tavakoli-Shiraji also shared his predictions for the future of enterprise AI adoption, along with an anecdote of what happened when the company was young and he pretended to be a lawyer to land a big enterprise contract. See the full interview below.