OpenAI faced a common but critical problem: thousands of interested companies and very few human reps to handle them. Instead of sending generic forms, they built an inbound sales assistant powered by ChatGPT that replies quickly, in the prospect's language and with real product context. They published the experience on September 29, 2025. (openai.com)
How the sales assistant works
It's not magic. It's engineering applied to the company's knowledge. Product documentation, policies, customer stories and playbooks are connected to the model so it can reason about concrete questions without guessing.
The result: personalized answers in minutes. A hospital receives compliance details from the very first exchange. A company in Tokyo gets the explanation in Japanese, not an automatic English form letter. If the conversation is clearly commercial, the thread is transferred to a representative with all the context intact. (openai.com)
Trained with reps, not to replace them
The most interesting part wasn't just the first automated reply, but the feedback loop. Every draft the model produced went back to sales reps for corrections. Those corrections were used as training data.
That strategy boosted the assistant's accuracy from around 60 percent to over 98 percent in a few weeks. In other words, the system learned to sound like the best version of the sales team and to apply human judgment at scale. (openai.com)
"We built a very complex evaluation system... once we automated those evaluations, we went from 60% to 90% and now to 98% on the first emails." (translation). (openai.com)
Real results: from lost leads to measurable growth
What used to be a dead queue turned into a growth channel. Repeated cases showed that small companies could receive useful answers within hours and close enterprise contracts in days. In a matter of months, this unlocked millions of dollars in annual recurring revenue. For the sales team, inboxes stopped filling with low-quality leads and conversations began with real intent. (openai.com)
What this brings to your company
Wondering whether this applies outside OpenAI? Yes — and not just for sales. Onboarding, renewals and support can benefit from fast, personalized conversations based on verified documentation.
If you're at a company with spikes of inbound demand, consider this:
- Identify the sources of truth your assistant will need: documentation, policies, use cases.
- Design a review cycle with your commercial team to convert feedback into training.
- Define clear criteria for when to escalate the conversation to a human.
OpenAI shows that scaling the excellence of the best salesperson with AI changes what's possible for the whole team. (openai.com)
The final lesson is simple and practical: automation has value when it preserves human judgment and delivers verifiable, fast answers. Wouldn't you like the first contacts with your product to feel like that?
