Why Enterprise Sales Still Fail Even When the Technology Works

The Technology Fallacy

Arsalan Tavakoli-Shiraji sat across from a prospect he had spent weeks trying to reach. The conversation lasted less than five minutes. The executive had recently signed a large contract with a competing vendor - a decision he had staked his professional reputation on. No pitch, however polished, was going to change that calculus.

The episode remains a formative lesson for the Databricks co-founder and senior vice president of field engineering. "Everybody when they're selling to enterprises at startups thinks it's all about the tech and having the best tech," he explained at a recent industry event. What he learned through repeated setbacks was simpler and harder to fix: "people and processes matter the most."

Databricks, now a data storage and AI platform valued at $190 billion with more than 70 per cent of the Fortune 500 on its client roster, began as a company that couldn't get meetings. Tavakoli-Shiraji was responsible for bringing in the first large contracts. The path from cold outreach to category dominance offers a blueprint for startups navigating the enterprise procurement maze - one that has less to do with feature sets than with alignment, access and organisational readiness.

The Partnership Unlock

When Databricks was still building its early customer base, it tried repeatedly to secure a meeting with UBS, the Swiss financial services giant. Every attempt failed. The startup lacked the credibility, the security certifications and the distribution channels that banks require before they will consider a vendor.

Then Databricks became an Azure cloud partner. The arrangement meant that sales agreements would run through Microsoft's legal and commercial framework. More importantly, Microsoft's enterprise sales representatives began booking meetings on Databricks' behalf and advocating for the product in rooms the startup could not access on its own.

"All of a sudden, that meeting wasn't hard to get," Tavakoli-Shiraji recalled. "It was a huge boon to us that we were able to partner with somebody who had that access and distribution."

The floodgates opened. Microsoft's field teams began pulling Databricks into opportunities across geographies. New challenges emerged - prospects demanded features for countries where Databricks had no local presence, and the company scrambled to meet compliance requirements it hadn't anticipated. But these were, in Tavakoli-Shiraji's words, "good problems to have."

The dynamic has since reversed. Databricks now serves as the umbrella for younger companies, embedding their products under its own terms of service and cybersecurity infrastructure. Hightouch, which builds AI-powered marketing and customer-data tools, sells through Databricks' agreements. Addepar, which provides data platforms for investment professionals, does the same. Tecton, a machine learning operations startup, was eventually acquired by Databricks after being part of this ecosystem.

At Opentechwire, we've tracked similar partnership models across the region - from Seoul-based AI infrastructure firms riding Samsung's enterprise rails to Singapore fintech startups leveraging DBS Bank's procurement channels. The pattern is consistent: distribution matters as much as differentiation.

The Four Obstacles Enterprises Raise

Tavakoli-Shiraji groups the most common objections and concerns he encounters into what he calls the four Cs: context, choice, cost and control.

Context refers to an AI system's ability to learn organisational specifics - terminology, workflows, priorities - in the same way a new employee would. Generic models that cannot adapt to a company's internal language or processes struggle to deliver value beyond novelty.

Choice means the freedom to swap models or vendors without rewriting integrations. Enterprises resist lock-in, particularly when the underlying technology is evolving as rapidly as large language models are today.

Cost is about predictability. Enterprises want to know what running an AI product will cost at scale, not just in the proof-of-concept phase. Usage-based pricing that balloons unpredictably is a common deal-breaker.

Control encompasses governance, auditability and data protection. Who can see what? What guardrails prevent the model from generating problematic output? How is sensitive information isolated? These questions surface in legal and compliance reviews, and startups that cannot answer them in detail lose momentum.

Tavakoli-Shiraji notes that many vendors focus on the technical elegance of their models while underestimating how much enterprise buyers care about these operational dimensions. The best algorithm in the world does not close a contract if procurement, legal and security teams cannot sign off.

The Legacy System Opportunity

When Tavakoli-Shiraji asks chief information officers where they stand on AI adoption - on a scale from one, where employees merely have access to ChatGPT, to five, where new AI capabilities have fundamentally reshaped business processes - most executives position themselves at three or four. His assessment is blunter: "99 per cent of the ones I talk to are in bucket one and two."

The gap between aspiration and implementation is widest in the core systems that underpin enterprise operations. Banks still run IBM AS/400 mainframes and legacy code that no one dares touch because institutional knowledge of what those systems actually do has eroded over decades. "Nobody wanted to touch it because nobody knew what that code did," Tavakoli-Shiraji said.

AI agents are beginning to change that calculus. Models capable of parsing and decomposing legacy codebases can now document what those systems do, map dependencies and even generate modern equivalents. The opportunity for startups is significant: products that layer contemporary interfaces onto ancient infrastructure, or that migrate enterprises off those platforms entirely, address pain points that have persisted for years.

The challenge is that these projects are long, expensive and politically fraught. Replacing a mainframe is not a technical decision; it is an organisational one that touches procurement, risk, operations and business units. Startups that understand this dynamic - and that can navigate the stakeholder map - have an opening. Those that pitch only on speed or cost savings will stall in committee.

What Kills Deals Before They Start

Early in Databricks' life, Tavakoli-Shiraji secured meetings that collapsed not because of product fit but because the startup was unprepared for the contractual and security questions that followed initial interest. Legal teams asked about indemnification clauses the company had never drafted. Security officers requested certifications Databricks did not yet hold. Procurement departments wanted volume discounts and payment terms the startup had not modelled.

These are not edge cases. They are the standard operating procedure of enterprise procurement. Startups that treat them as obstacles to be worked around, rather than requirements to be met, lose deals they thought were progressing.

The lesson Tavakoli-Shiraji emphasises is that enterprise sales is as much about organisational readiness as it is about product-market fit. A startup might have the best machine learning inference engine in the category, but if it cannot produce an acceptable data processing agreement or a SOC 2 Type II report, the deal will not close.

This is where partnerships and umbrella arrangements become decisive. By selling under a larger vendor's terms, startups inherit the legal and security infrastructure that took incumbents years and millions of dollars to build. The trade-off - giving up margin or control - is often worth the acceleration.

The Deployment Reality

Tavakoli-Shiraji is sanguine about the timeline for meaningful AI adoption in enterprises. "It's going to take many, many years before people can actually properly deploy AI in a very meaningful way," he said. Beyond enabling engineering teams to write code faster, most organisations have not yet figured out how to deploy AI agents in ways that create measurable business value.

The gap is not primarily technical. It is organisational, cultural and operational. Enterprises move slowly not because they are unaware of AI's potential but because integrating new capabilities into established workflows, governance frameworks and risk models is hard. Startups that acknowledge this reality and build products that fit into existing processes - rather than demanding that enterprises reinvent themselves - will have an easier path.

The other implication is that patience is required. Enterprise sales cycles are long by nature. When the product category is still forming and buyers are uncertain about use cases, those cycles stretch further. Startups that expect to move at consumer software velocity will burn through capital and morale.

What the Databricks Playbook Reveals

Tavakoli-Shiraji's advice distils into a few operational principles. First, accept that distribution and credibility are as important as product quality. If a startup cannot get meetings, it cannot sell. Partnerships with incumbents are often the fastest way to gain access.

Second, prepare for procurement before the first enterprise conversation. Legal templates, security certifications, reference customers and pricing models should be ready before a deal enters the pipeline. Scrambling to answer basic questions signals immaturity and kills momentum.

Third, understand the stakeholder map. The person who loves the product is rarely the person who signs the contract. Identifying the economic buyer, the legal gatekeeper, the security reviewer and the budget holder - and addressing their concerns in parallel - is what moves deals forward.

Fourth, recognise that AI adoption in enterprises is still in its earliest stages. The opportunity is large, but the timeline is long. Startups that can survive the deployment lag and support customers through organisational change will capture value that faster-moving competitors miss.

At Opentechwire, we've watched this pattern play out across sectors and geographies. The startups that win enterprise customers are rarely those with the most elegant technology. They are the ones that understand how large organisations make decisions, what risks they will tolerate, and who holds the keys to the budget. Tavakoli-Shiraji's experience at Databricks - from struggling to book meetings to serving more than 70 per cent of the Fortune 500 - is a case study in that discipline.