Proaction Reports 60% Sales Lift Using Codex and GPT-6 Astra

Fleet management startup Proaction reported that deploying OpenAI's Codex has cut engineering workloads and contributed to a 60% increase in sales, according to OpenAI's case study.

Proaction builds software for companies managing fleets of vehicles, including cars, trucks, and construction machinery. To show potential clients how the platform accommodates their specific operations, co-founder and COO Colin Knudsen uses Codex to produce four to six customized, interactive HTML demos each month. Each demo takes 30 to 45 minutes to create, drawing on customer data taken from Granola recordings, email threads, and spreadsheets.

Building equivalent demos manually previously required about 10 hours of engineering work per demo. Knudsen estimates that generating them with Codex prevents 40 to 60 hours of engineering work every month. The demos also serve as visual specifications for developers once deals close, which reduces back-and-forth implementation questions. Knudsen noted that the percentage of prospective deals advancing from initial contact into solution development rather than nurture increased by 50% to 60%.

Codex plugins for tools such as Granola, Gmail, Slack, Linear, GitHub, and HubSpot allow Knudsen to handle daily operational workflows within a single interface. Addressing 15 to 20 distinct tasks each day—such as extracting call transcripts for follow-ups, logging Linear issues, and updating HubSpot opportunities—saves Knudsen 25–33 hours a month, with overall founder savings cited at 33 hours per month.

Proaction also integrates OpenAI models directly into customer-facing operations through a system it calls the Managed Execution Layer. When customers submit vehicle issue reports with photos, ChatGPT-5.6 Sol assists with damage identification. In addition, the company deploys GPT-Live-1 and GPT-6 Astra to run autonomous voice agents. One agent, named Marty, coordinates vehicle maintenance by speaking directly with drivers, placing calls to repair shops, scheduling appointments, and managing estimate approvals. Head of Product Danny O’Halloran stated that computer-use runs in GPT-6 Astra execute more succinctly than runs on GPT-5.6 Sol.