Databricks Couldn't Get a UBS Meeting. Microsoft Could.

Here's a fun problem. You build something good. Really good. You email a giant bank, and nobody writes back.

That's roughly where Databricks sat in its early years. Today it's valued at $190 billion and counts more than 70% of the Fortune 500 as customers. But Arsalan Tavakoli-Shiraji, one of its co-founders, told a TechCrunch Disrupt audience that the early days looked nothing like that. The company couldn't even get a meeting with UBS.

What happened next is a useful lesson for anyone selling to bigger companies, and you don't need a data platform to use it.

He said young startups assume their technology will sell itself, and it won't. Tavakoli-Shiraji, who is senior vice president of field engineering at Databricks and helped land its first large deals, put it plainly: good technology is required but not sufficient. "People and processes matter the most," he said.

TechCrunch's Julie Bort wrote up the talk on October 26, 2026. It covers how the company got into big enterprises, what customers worry about when they buy AI, and where he sees openings for new startups.

Let's pull out the pieces a first-time founder can use.

Because the buyer isn't evaluating your product in a vacuum. A large company has procurement teams, security reviewers, lawyers, and a dozen vendors already in place. Your demo is one small input.

Think of it like a restaurant opening in a building with no foot traffic. The food can be fantastic. If nobody walks past the door, the kitchen stays quiet.

Most first-time founders I talk to spend 95% of their energy on the product and about 5% on how a stranger at a big company would ever hear about it. Flip that ratio a little. Not all the way. A little.

Databricks became an Azure partner, and Microsoft's sales team started booking the meetings. Without that, Tavakoli-Shiraji said, the company couldn't get in front of UBS. After the Microsoft partnership, Microsoft reps advocated for Databricks inside accounts, and sales could run under Microsoft's existing terms.

The relationship didn't appear overnight. VentureBeat reported in 2018 that the two companies had worked together for about two years, that Azure Databricks entered beta in November 2017, and that it left beta on March 22, 2018. The integration took many trips between Databricks' San Francisco office and Microsoft's Redmond campus. Board member Ben Horowitz helped through his relationship with Microsoft CEO Satya Nadella.

Microsoft has offered Azure Databricks as a first-party service since 2018, according to Databricks' own newsroom. In June 2025 the two announced a multi-year early extension of the partnership.

Did you catch the effort in there? Two years of work, plenty of travel, and compliance investments made earlier than planned. A shortcut to distribution still costs you something.

Demand shows up fast. Too fast, sometimes.

Tavakoli-Shiraji said the partnership produced a flood of inbound interest, and that created new headaches. Customers asked for features in countries Databricks hadn't entered yet. If you've ever had a customer promise to sign if you add one thing, you know the feeling, only louder.

So before you chase a big partner, ask a boring question. Can we support what happens if this works? Do we have support staff, a security story, and a roadmap that survives ten new requests a week? If not, you're borrowing trouble.

Context, choice, cost, and control. Tavakoli-Shiraji uses these four words to group the worries large buyers bring to AI projects.

• Context: The AI has to learn the company's own terms and processes. He compares this to onboarding a new employee.
• Choice: Buyers want to switch models instead of being locked into one provider.
• Cost: Customers need to manage what the product costs to run.
• Control: Safeguards should govern what the AI does and protect sensitive data.

Even if you aren't selling AI, the pattern travels. Whenever you pitch a bigger company, write down their four worries before you write your slide deck. Then answer each one on a single page.

Old, expensive systems. He estimates that 99% of the CIOs he talks to sit at the lower levels of a one-to-five AI adoption scale, where level one means employees simply have access to ChatGPT. Many enterprises haven't found uses for AI agents beyond engineering teams.

His suggestion is to look at legacy systems, like IBM AS/400 machines and mainframes at banks. His reasoning: "Nobody wanted to touch it because nobody knew what that code did." AI agents can now read and analyze code that few humans understand, which makes modernizing it newly possible.

That's a decent pattern to steal. Look for work everybody avoids because it's tedious, risky, and expensive. New tools sometimes turn "nobody wants to touch it" into "we can finally afford to."

You don't need a partnership with Microsoft. But you can plan the same way, at a smaller size.

Start with a simple map. Who already has the trust of your target buyer? A software platform they use, a consultant they hire, an industry group they join? Those are your possible umbrellas. You can sketch the map in a spreadsheet, a Notion page, or a planning tool like Foundra that walks first-time founders through go-to-market sections, and the format matters less than doing it.

Then rank the umbrellas by two things: how much access they give you, and how much they'll cost you in time and control. A partner who books meetings is wonderful until you realize they also set the terms.

Tavakoli-Shiraji said early deals stalled because the startup wasn't ready for legal, contractual, and security questions. That's a classic trap.

You can get ahead of it without hiring a team. Keep a short folder with your security practices, data handling policy, and standard contract terms. Know who on your side answers each kind of question. When a buyer's reviewer asks for something, a two-day reply beats a two-week reply almost every time.

And yes, you should talk to an attorney about contract terms. This article isn't legal advice.

One more story from the talk. He recalled a meeting with a prospect who had just signed a big deal with a major competitor and had no interest in hearing his pitch.

Ouch. But it's a good reminder. Before a meeting, ask two things. Is there budget this year? Did they recently sign with someone else? A five-minute conversation upfront can save you a week of preparation.

If you want to hear more from operators like this, TechCrunch lists Toast's Steve Fredette and Bessemer's Kent Bennett among the speakers at its Founder Summit on November 4 in Boston, with sessions on customer acquisition and product-market fit.

• Databricks co-founder Arsalan Tavakoli-Shiraji says technology is necessary but not sufficient; people and processes matter most.
• Databricks couldn't get a UBS meeting until its Microsoft partnership opened doors.
• Partnerships take real work and can bring more demand than you can serve.
• The four C's of enterprise AI buyers are context, choice, cost, and control.
• Old, expensive systems may be an opening because AI can now analyze code few people understand.
• Prepare for legal and security questions before your first big deal.
• Qualify prospects early to avoid pitching someone who already chose a competitor.

Do I need a big partner to sell to enterprises? No. Databricks used one, but you can also build trust through pilots, references, or a respected adviser. A partner is one route, not the only one.

What is a cloud partner? It's a company that sells or hosts your product on a larger cloud platform. In Databricks' case, that was Microsoft's Azure, where Microsoft's sales team could introduce the product to its own customers.

How long can a partnership take to build? VentureBeat reported about two years of work before Azure Databricks left beta in 2018. Yours may be shorter or longer.

What does "context" mean when selling AI? It means the AI understands a company's terminology and processes, similar to how a new employee learns the business.

How should early-stage founders prepare for security questions? Keep clear written answers on data handling, access controls, and contract terms. Have a named person who responds, and talk to an attorney about legal terms.

Is selling to large companies right for every startup? No. Enterprise deals can be slow and demanding. If you're pre-product, talking to ten small customers may teach you more.

1. Databricks co-founder shares his secrets for winning big enterprise customers, TechCrunch, October 26, 2026
2. How Microsoft and Databricks crafted a unique partnership for AI data processing, VentureBeat
3. Databricks and Microsoft Extend Strategic Partnership for Azure Databricks, Databricks newsroom
4. TechCrunch Founder Summit 2026, TechCrunch