Converting inbound leads into customers at OpenAI

This is part of our series about how OpenAI is building its own solutions on our technology.

When ChatGPT Enterprise and Business each launched, inbound demand surged. Tens of thousands of companies—from early-stage startups to multinational enterprises—were reaching out every month. The demand was remarkable. The strain on our systems was real.

Routing those leads through forms and static workflows couldn’t meet the moment. Too many prospects got an automated reply telling them to sign up online. Too few had their questions answered. The result was missed opportunities and a buying experience that didn’t match the trust customers were placing in us.

The challenge wasn’t just scale. It was quality. Buyers wanted specific answers:

“We were getting thousands of leads a month and only had capacity to talk to a small fraction. Some leads needed a couple of questions answered to really make a great buying experience, but we weren’t able to provide that personalized experience,” says Harsha Chilakamarri, Go-to-Market Innovation.

Traditional automation couldn’t carry that nuance. Hiring linearly wasn’t sustainable. We needed a different approach.

We created an AI-powered inbound sales assistant designed not to replace reps, but to extend their reach—trained and refined with rep feedback.

At its core are our internal connectors. Product documentation, policy libraries, customer stories, and playbooks are pulled into context the model can reason over. The assistant doesn’t guess. It responds with accuracy, in the prospect’s language, directly tied to their question.

That means prospects get a personalized response within minutes, written in their own language, grounded in their actual question.

“This model allows us to engage with and provide every customer a hyper personalized experience,” says Chilakamarri.

This isn’t automation for its own sake. It’s automation that delivers value, right away.

The breakthrough wasn’t just the assistant’s first reply. It was the loop behind it.

When training the model, every draft response went back to sales reps for corrections. Every correction became training data. Accuracy climbed from 60 percent to more than 98 percent within weeks. Instead of generic templates, the assistant started to sound like the best version of our team, codifying judgment and making it available at scale.

For reps, the shift was immediate. Inboxes weren’t clogged with unqualified leads. They opened conversations already in motion, with prospects who had real intent and real questions answered.

The evals also gave leadership confidence. They showed measurable progress, not just anecdotes. They proved the assistant could be scaled responsibly.

The impact was immediate. A small company once lost in the queue submitted questions, got thoughtful answers within hours, and signed an enterprise contract days later. Those stories repeated again and again.

What had been a dead end became one of our strongest growth channels. Within months, multimillions in annual recurring revenue were unlocked.

For reps getting passed qualified leads, the shift was just as valuable. Instead of digging through generic leads, they saw active conversations with clear intent. For the first time, no one felt left behind.

This isn’t only about inbound leads. It points to a broader opportunity: onboarding, renewals, and support can all benefit from trusted, personalized conversations.

The lesson is simple: when you scale the excellence of your best reps through AI, you change what’s possible for the entire team.

As Chilakamarri put it: “Leadership could not be more excited by this. It’s proof that we can build OpenAI on OpenAI and showcase our technology directly to customers.”

Personalizing every lead isn’t a tactic. It’s becoming a better way for all engagement.