OpenAI's Aliisa Rosenthal Shares Lessons on Scaling AI Sales from Self-Serve to Enterprise

Aliisa Rosenthal joined OpenAI over four years ago when the company's revenue was in the low millions, primarily driven by a self-serve API used by developers who found the platform, swiped a credit card, and started building. Her role involved scaling the organization to hundreds of employees and billions in revenue. Her experience spans the full spectrum of client engagement, from nurturing early-stage startups to managing massive enterprise accounts.

Rosenthal advises founders to reverse traditional go-to-market strategies, emphasizing automation over immediate hiring for sales. She recommends building the sales machine first by leveraging modern tools to identify and streamline workflows before bringing on human sales personnel. This approach ensures that the process is optimized and efficient before scaling with staff.

When ChatGPT launched in 2022, it lacked enterprise features, but it quickly gained traction among large organizations. Rosenthal recounted being "bombarded" with requests for features like Single Sign-On (SSO), Non-Disclosure Agreements (NDAs), and invoicing, none of which were initially available. In response, OpenAI dedicated nine months to building a sophisticated, high-end, and expensive product designed to meet these specific enterprise demands.

OpenAI's self-serve model, released in January 2024—just four months after the enterprise version—unexpectedly cannibalized the enterprise business. The self-serve growth significantly outperformed the enterprise segment. This indicated that most customers preferred to bypass the traditional sales process, opting for a direct, self-guided approach to product access.

Rosenthal shared a key regret from her time at OpenAI: not collecting enough information from inbound leads when they were overwhelmed by up to 10,000 new leads daily with a small team. She specifically recommended adding more fields to sign-up forms, such as phone numbers. This practice ensures that once automation is in place and lead handling is streamlined, there is sufficient data for effective follow-up and AI-driven outreach.

Rosenthal highlights the effectiveness of contemporary tools in automating various aspects of the outbound sales pipeline. She suggests using platforms like Clay and Nooks to streamline research, sequencing, and dialing. By automating these initial stages, sales representatives can dedicate their time to engaging with qualified leads and focusing on high-value conversations.

Rosenthal frequently encounters founders who are stuck in what she terms 'pilot hell,' where numerous pilot programs are running but none are converting into paying customers. She warns that pilots are often high-maintenance and, even if they do convert, they typically require restarting the entire sales process from scratch. Consequently, pilots should be reserved only for the most valuable, highest-revenue opportunities to justify the investment.

Rosenthal suggests several strategies to circumvent pilot objections without losing control of the deal. Instead of offering a pilot, companies can introduce prospective clients to existing customers for validation or conduct customized data demonstrations over Zoom, rather than granting full product access. Additionally, offering opt-out clauses or limited-term contracts can shift the burden of product validation to the buyer, streamlining the sales process.

OpenAI initially priced its products based on the internal cost of service, which led to setting prices too high in the market. This strategy became unsustainable as competitors like Copilot, Gemini, and Anthropic entered the market with much lower price points. The intense market competition ultimately forced OpenAI to correct its pricing strategy.

Initially, OpenAI's fixed monthly fees for product licenses restricted adoption to specific teams within organizations. To overcome this, the company transitioned to a model that included a bare license fee combined with usage-based pricing. This change significantly lowered the barrier to entry, encouraging company-wide spread and making product adoption much easier across a broader user base.

Rosenthal advises that the optimal time to hire a sales team is when the self-serve process has reached its limit and customers explicitly begin demanding human interaction. This typically occurs as companies move up-market and engage with larger organizations that require human trust-building and personalized relationships. At this stage, a sales team becomes essential to facilitate these interactions and drive further growth.

Rosenthal terms the current trend the "revenge of the steak dinner," observing that as more sales activities become automated, companies increasingly crave in-person interactions with humans. This underscores the heightened value of human contact in an automated era. Conferences and dinners remain highly effective channels for building relationships, as humans are still essential for facilitating these personal connections.

While OpenAI could offer attractive equity packages that increased in value, making fixed compensation viable, Rosenthal notes that this isn't typical for most startups. Generally, it's challenging to recruit top-tier enterprise sellers without offering a variable compensation component. She recommends simpler plans with significant upside potential for high-performing individuals to attract the best talent.

Humans continue to excel at value selling, effectively communicating a product's worth to potential buyers. Rosenthal believes that AI, while advanced in many areas, still has significant ground to cover in fully replicating the nuanced and persuasive aspects of human-led value selling. This suggests that the human element remains crucial for converting complex sales.

Rosenthal distinguishes Product-Led Growth (PLG) from high-touch Proof of Concepts (POCs) by defining PLG as self-serve product access, often gated by features or usage tiers. In contrast, POCs are contracted, time-bound engagements that demand extensive service and support, typically requiring a sales engineer or salesperson to supervise the process. This 'handheld' approach is resource-intensive and distinct from the autonomous nature of PLG.

There is a severe talent shortage for forward-deployed engineers, with approximately 55,000 job openings for only 5,000 qualified individuals, making them both expensive and difficult to hire. Despite the high cost, deploying these engineers creates significant product stickiness. Once a customer has invested in the services element of having an expert assist with product integration, they are unlikely to switch providers, thus locking them in.

The first ten customers for an AI startup should ideally serve as design partners rather than targets for automated outbound sales. These partners need a more 'handheld' approach and should be chosen for their ability to provide high-quality feedback, understanding the limitations of an early-stage product. Rosenthal also suggests using internal, non-production projects to secure prestigious logos, effectively bypassing rigorous security and compliance hurdles common with large enterprises.

Initial design partnerships for early-stage product validation should ideally be built on existing professional relationships. These connections can come from prior employment, introductions from investors, or personal friendships within companies. Creating a friendly, low-friction environment through these relationships is crucial for successful early experiments.