Founders should automate their go-to-market process first, then hire people to handle the bottlenecks that remain.
2
OpenAI launched an expensive enterprise product before self-serve, and later found that self-serve grew faster and pulled customers away from the enterprise funnel.
3
Pilots, security reviews, and high per-seat prices add work and friction, so companies should replace them with lighter evaluations, automated trust processes, and usage-based pricing.
Summary
Aliisa Rosenthal uses OpenAI's early enterprise experience to explain how AI companies should build their go-to-market motion. OpenAI had strong demand after ChatGPT launched, but it spent nine months without enterprise features, then built an expensive, high-end product for large companies. When self-serve arrived four months later, it grew faster and competed with the enterprise business. Rosenthal's advice is to start with automation and self-serve, observe where buyers need help, and add people at those points. She argues that founders should collect more signup information, respond to inbound leads, avoid sending buyers away with tasks, and treat pilots as limited exceptions. Security questionnaires and documentation can often move into a trust portal. OpenAI also learned that $60 per user per month restricted adoption, while a lower base fee with usage pricing made broader company adoption easier. Salespeople become useful when buyers need trust, value selling, and high-touch conversations.
Founders should build the go-to-market machine before hiring the team
Rosenthal advises founders to begin with the parts of sales and marketing they can automate. Modern tools can handle much of the work traditionally assigned to sales operations, BDRs, SDRs, sales engineers, and large sales teams. The process should reveal where leads stall, where buyers need help, and what remains broken. People can then be added at those bottlenecks. She contrasts this with the usual sequence, where a company hires a large sales organization first and only later tries to add automation. Her OpenAI experience changed her view: every later product launch started self-serve first, then added humans where customer feedback showed they were needed.
OpenAI lost enterprise demand during the nine months before its enterprise product existed
When ChatGPT launched at the end of 2022, companies were already asking for SSO, NDAs, and invoices. Rosenthal had none of those enterprise features and spent months asking the technical team to build them. Approval came nine months later. During that wait, the loudest companies were large enterprises, so OpenAI built a fast, expensive product with many features and went far upmarket. When the product finally launched, almost every company that had previously reached out told Rosenthal, "You never got back to me." Many had bought Microsoft Copilot instead. The company had demand, but failed to keep those buyers engaged while it built the product.
Self-serve should have come before the enterprise tier
OpenAI released self-serve in January 2024, about four months after its enterprise version. Self-serve grew much faster and cannibalized the enterprise business because many customers did not want a salesperson or the high initial price. Sales representatives then competed with the cheaper option and sometimes lost customers to it. Rosenthal says the better sequence would have been to launch self-serve, learn what customers needed beyond that product, and then build the enterprise offering around those gaps. The lesson shaped later launches at OpenAI, where the company started self-serve first and added people after seeing where human help was necessary.
Inbound demand still needs data collection and a response
At one point Rosenthal and four sales representatives received about 10,000 inbound requests a day. The volume was too large to handle manually, but the signup process also captured too little information for later follow-up. In hindsight, she would have added more fields, including an optional phone number, so the team or an AI system could contact qualified people later. She also would have sent an automated response to every company, even when the enterprise product was not ready. That message could acknowledge the request, explain that the product was on a waitlist, and ask what the company needed. When OpenAI eventually launched, many prospects had already moved on.
A buyer should not have to leave the sales process to do unpaid work
Rosenthal tells companies to remove work from the buyer's side of the deal. Instead of asking a prospect to complete five tasks, find another stakeholder, or return with answers, the seller should do as much as possible with them. That can mean running a hackathon in the customer's office or setting up a live Zoom session. Sending the buyer away creates delays and gives the seller less control over the sales cycle. She also recommends fewer pricing choices, simpler paperwork, and an easier approval process. Her broader point is that every extra task gives a buyer another reason to stop or postpone the purchase.
Pilots should be reserved for the biggest opportunities
Rosenthal describes pilot work as a second sales process that companies often run for free. The seller must win approval for the pilot, provide substantial support, and then begin another sales process when the pilot ends. She recommends using pilots only for the largest and most valuable deals. Alternatives include a reference call with another customer, an evaluation on part of the buyer's data, or a custom-data demo over Zoom without handing over product access. A startup can also offer a signed contract with a 90-day opt-out clause. That puts a time limit on the customer's validation process while avoiding an open-ended proof of concept.
Security work can move into a self-serve trust process
Security reviews are a common place for deals to stall. Rosenthal recommends trust portals where buyers can sign an NDA, retrieve penetration-test and security documents, and find answers without waiting for a call. AI tools can fill in security questionnaires automatically. When a company asks for a two-hour security meeting, she suggests directing it to the portal first and asking the buyer to identify what is missing. This changes the call from a standard requirement into a discussion about specific gaps. Rosenthal's advice is to automate as much of the security process as possible because much of the work does not require a live sales or security interaction.
Lowering the entry price helped usage spread across the company
ChatGPT Enterprise initially cost $60 per user per month. Rosenthal says OpenAI set that price based on its own serving costs and later saw that it was too high compared with Microsoft Copilot, Gemini, and Anthropic products. The company changed to a lower base license fee with usage-based pricing. More companies then signed contracts, and usage increased as the initial barrier fell. At the old price, buyers limited access to developers, investors, or a small subset of employees. With a lower threshold, adoption spread more widely. Rosenthal also suggests spending caps and employee-level limits to address concerns about usage costs rising unexpectedly.
"When we finally did launch our enterprise product 9 months later, almost every company went out and talked to you said, "Well, you never got back to me." And I went out and I brought Microsoft Copilot."05:20
Who should watch
Founders deciding whether to start with self-serve, enterprise sales, or a mix of both.
Go-to-market leaders whose teams are overloaded with inbound requests, pilots, or security reviews.
AI companies choosing their first enterprise customers and deciding when to hire salespeople or forward deployed engineers.
