Article Overview
Most AI adoption stories in enterprise settings follow a predictable pattern: a team deploys an AI tool, usage grows slowly, and results are modest. Cars24's story is different in three ways that make it worth examining carefully.
First, the scale is real and specific — over one million conversation minutes handled by AI agents every month, across the full lifecycle of buying and selling a car in India's complex, manual-heavy automotive market. Second, the results address problems that genuinely mattered: a 50% increase in customer support resolution rates, an 80% reduction in turnaround time across key workflows, and the recovery of 12% of seller leads that had previously been written off after ten days of silence. Third, and most unexpectedly, the tool that was supposed to make engineers faster — Codex — spread to finance, legal, operations, and investor relations before the engineering team had finished their own rollout.
This article covers how Cars24 built AI agents across the buyer and seller journey, what changed in their software development process, how Codex moved into parts of the organization that were never part of the original plan, and what the 85-to-90% daily active usage rate across 600 central employees actually looks like in practice.
Introduction
Buying a car in India is nothing like buying one online in a market with mature e-commerce infrastructure. Most transactions remain manual, regulated, and fragmented. The process unfolds over days or weeks — calls between buyer and dealer, document checks that require in-person visits, financing conversations that depend on which branch you reach, and follow-ups that can quietly die without anyone noticing.
Cars24 operates one of the world's largest AI-native automotive ecosystems for buying and selling pre-owned cars, with its primary market in India and additional operations in the UAE and Australia. The company covers the full ownership journey — from the moment someone considers buying or selling through financing, inspection, transaction, and post-purchase service. In a market where most of this still happens through phone calls and physical visits, the challenge of scaling that experience consistently is significant.
The company's deployment of OpenAI's technology — the API for building voice and chat agents, ChatGPT Enterprise for internal operations, and Codex for development workflows — tells a specific story about what enterprise AI adoption looks like when it moves beyond pilots into production.
Quick Summary
| Metric | Result |
|---|---|
| Monthly AI conversation minutes | 1,000,000+ |
| Customer support resolution rate change | +50% |
| Turnaround time reduction across key workflows | 80% |
| Previously lost seller leads recovered | 12% |
| Employees using ChatGPT Enterprise | ~600 |
| Daily active usage rate | 85–90% |
| Teams reached by Codex | Engineering, product, finance, legal, marketing, operations |
The Problem: Scale Without Proportional Headcount
Unlike most digital marketplaces, Cars24 cannot close a sale in a single browsing session. A buyer in India making a decision about a pre-owned car needs to compare options across their actual budget and use case, book and attend a test drive, arrange financing, complete documentation, and often revisit the decision multiple times before committing. The process happens significantly outside the app — across phone calls, WhatsApp conversations, and in-person visits.
As Cars24 scaled, this created a structural tension. Delivering a consistent, high-quality experience across millions of these multi-step, multi-day interactions required either growing the human team at the same rate as the customer base — which is operationally expensive and logistically complex — or finding a way to handle the conversation volume without proportional headcount growth.
The answer was building AI agents using the OpenAI API, designed to carry customers through each stage of both the buying and selling journey without replacing the human relationships that close the deal.
The Buyer Journey: Four Stages, One Continuous Experience
The agents Cars24 built cover the buyer's path from first contact through post-purchase support in four distinct stages, each picking up where the last left off.
Stage one begins with the initial call. When a buyer reaches out, an AI agent collects the information that shapes everything that follows: their budget, family size, how they commute, and what type of car they have in mind. Using that context, the agent searches the Cars24 catalog and recommends specific cars that fit, books a test drive appointment, and opens the financing conversation for customers who need it. This is not a scripted FAQ response — it is a consultative conversation that adapts to what the customer actually says.
Stage two happens before the test drive. The agent follows up to confirm the appointment, surface alternative cars if the buyer's preferences have shifted since the initial conversation, and collect additional details that will be needed for financing. Appointments that would previously require a human agent to chase are now handled automatically, with a follow-up that is consistent every time regardless of which day or hour the visit is scheduled.
Stage three happens after the visit. The agent checks in to understand where the customer stands — whether they want to move forward with the purchase, book a second visit to look at another car, or take more time. This is the moment in the buying process where leads historically went cold most often, and the moment where a timely, contextually appropriate follow-up matters most.
Stage four continues through and after the purchase. Once a customer has bought, agents remain available for feedback, warranty questions, returns, and after-sales service — maintaining the relationship rather than treating the transaction as the endpoint.
Vikram Chopra, Builder at Cars24, describes the operational shift this represents: "Buying a car in India is a journey, not a transaction. For years, the experience depended on who picked up the phone. AI changes that. Today, we handle over a million conversation minutes a month through AI, giving every customer a high-quality experience at any scale."
The Seller Journey: From Inquiry to Inspection to Re-Engagement
For people selling cars, the AI agent workflow follows a parallel structure. From the initial inquiry, an agent collects vehicle details, schedules an inspection, and sends reminders. When appointments are missed — as they frequently are in a market where scheduling is informal — the agent helps customers reschedule rather than treating the missed appointment as a lost lead. If a customer has sold their car elsewhere, agents gather competitive information that feeds back into Cars24's understanding of the market.
The most commercially significant capability is re-engagement. Seller leads used to drop out of the funnel after ten days of inactivity — a pattern common in any high-consideration transaction where a customer might be waiting for a better price, comparing offers, or simply not ready to commit. Cars24's AI agents now re-engage those customers after the ten-day window, qualify whether their intent has renewed, and return them to the active funnel when Cars24 can offer the price the seller is looking for.
The result of that re-engagement workflow is the 12% lead recovery figure — seller leads that would previously have been written off are now being converted into sales. In a marketplace where each transaction represents a significant ticket value, recovering one in eight previously lost leads through automated re-engagement represents meaningful additional revenue without adding to the cost of lead acquisition.
Codex Inside the Development Workflow
Alongside the customer-facing agents, Cars24 deployed Codex across its software development lifecycle. The framing from the beginning was deliberate: Codex is treated as a participant in day-to-day work rather than a standalone coding tool that developers consult when they need specific help.
In practice, this means Codex is woven into the existing workflow rather than running parallel to it. Product managers use Codex to draft and refine tickets in Linear — the project management tool Cars24 reoriented its workflows around in a matter of weeks specifically to create a cleaner integration path. Engineering teams tag Codex directly into bug reports, allowing it to pick up clearly defined tasks from the ticket itself. Codex summarizes work across GitHub and posts status updates for teams, reducing the frequency of standups needed to keep cross-team work coordinated.
The scope that Codex covers inside the development cycle now runs from ticket creation through task grooming, implementation, bug resolution, and progress reporting. It is not replacing any one of these activities so much as connecting them — moving work forward across stages that previously required multiple handoffs.
What No One Expected: Codex Spreading Beyond Engineering
The story inside Cars24 that Jayesh Gupta, Builder of AI and Innovation, identifies as the most significant is not the developer productivity improvement. It is that Codex spread.
In finance and investor relations, teams use Codex to pull numbers directly from their systems of record, run analysis on those numbers, and prepare investor reporting workflows — eliminating the process of manually collecting inputs from multiple business heads each time a report needs to go out. The work that previously required chasing approvals and waiting on data from across the organization now happens through an automated workflow that compiles, analyzes, and formats the information.
A separate finance workflow uses Codex to review purchase requests and purchase orders above a defined threshold. The workflow checks for anomalies, flags concerns that require human review, and automatically approves requests where no issues are found. The combination of automated review and automatic approval for clear cases dramatically reduces the turnaround time on routine procurement decisions — contributing to the 80% reduction in turnaround time across key service workflows.
Some teams went further. Employees built what they describe as "chief of staff" agents — AI systems that connect Slack, Gmail, WhatsApp, and other communication tools to manage scheduling, communication coordination, hiring workflows, and follow-ups across those channels. These were not built by the engineering team on behalf of other departments. They were built by the departments themselves.
This is the shift Gupta captures in his observation: "I thought Codex would make our engineers faster. What surprised me was how quickly it spread beyond engineering. Product managers, finance teams, and even day-to-day workflows started changing. That is when I realised we had not just changed how we write code but had changed how the entire company thinks about getting work done."
ChatGPT Enterprise: The Internal Operating Layer
Alongside the customer-facing agents and the Codex deployment, Cars24 rolled out ChatGPT Enterprise to approximately 600 employees across its central organization. The daily active usage rate sits between 85% and 90% — a figure that is unusually high for enterprise software of any kind, let alone an AI tool that requires teams to actively change how they approach their work.
The teams using it span engineering, finance, legal, marketing, and operations. The use cases range from building custom workflows and automating recurring tasks to moving work across systems more efficiently. The consistent pattern across departments is that employees are using ChatGPT Enterprise to solve problems they previously either solved manually or waited for engineering support to address.
Together, the three products — OpenAI API agents for customer conversations, Codex for development and operations, and ChatGPT Enterprise for internal workflows — have created what Cars24 describes as both a customer engagement layer and an internal operating layer.
What Makes This Case Study Meaningful
Several things about Cars24's deployment stand out from typical enterprise AI adoption narratives.
The starting context was genuinely difficult. India's pre-owned car market is not a streamlined digital environment where automation is straightforward. It is fragmented, manual, and relationship-dependent. Building AI agents that handle million-minute monthly conversation volumes in that context is a harder engineering and product challenge than doing the same in a more digitized market.
The metrics are operational and specific. A 50% improvement in customer support resolution rates, an 80% reduction in turnaround time, and a 12% recovery of previously lost leads are numbers tied to specific workflows rather than general productivity claims. Each one maps back to a concrete business problem that the AI deployment was designed to address.
The spread of Codex beyond engineering is the most instructive data point. Organizations that deploy AI and see it stay confined to the team it was initially targeted at are the norm. Organizations where other teams see what engineering is doing and start building their own workflows are significantly less common. The 85-to-90% daily active usage rate across 600 employees in functions as different as finance, legal, and marketing suggests that the tools became genuinely useful across contexts rather than remaining a specialized engineering capability.
Final Takeaway
Cars24's deployment of OpenAI's technology across customer-facing agents and internal operations represents what enterprise AI adoption looks like when it moves from experimentation to genuine operational integration. The one million monthly conversation minutes are handled by agents covering the complete buyer and seller journey — not a single-use-case chatbot but a multi-stage, context-aware system that follows customers through weeks of consideration and decision-making.
The internal story is equally significant. Codex reached finance teams, investor relations, operations, and individual employees building their own productivity tools — not because of a top-down mandate but because it proved useful enough that people without engineering backgrounds started adopting it for their own work. That pattern of organic spread into non-technical teams is the signal most organizations are looking for when they try to assess whether an AI deployment is succeeding.
For companies operating in complex, high-consideration, relationship-driven markets — where the customer journey extends well beyond a single session and where scale creates a genuine tension with consistency — the Cars24 story offers a practical picture of what AI agents can and cannot replace.
