There is a version of AI adoption that looks great on a slide deck. You hire a team. They build some bots. The bots work. Your company is “AI-first.” Everyone claps. Numbers improve.
That is not what happened at Cars24.
We had the platform. Voicebots doing 100,000+ minutes a day. Chatbots self-resolving 40% of support. CLIO replacing 200+ manual QA graders. The tech worked.
The business teams were still watching from the outside. They knew AI was happening. They did not feel it was theirs to touch.
So we ran a hackathon. Not for engineers. For the business.
The unlock was not a new model or a better prompt. It was the moment a business leader realized they could build it themselves.
The setup
Same tools our engineers use. AIRO for bots and voicebots. CLIO for post-call analysis. Doc Intelligence for documents. Real data. Real problems they own. Two days.
I went in expecting good energy and one or two prototypes worth productizing later. What I did not expect: working solutions, taken live within weeks. By the business. Without an engineering ticket.
What they shipped
Dealer Operations. A follow-up bot for sourcing leads — work they had been doing by hand for years. One month after go-live, dealer touch rate went from 1.1 to 0.32. A 71% reduction. Zero tickets filed against engineering.
Help and Support. A customer chatbot on AIRO for their top query categories. Support ticket load on those categories dropped 40%. No engineering involvement after platform access.
Lending. A Loan Against Car voicebot campaign. 800 qualified leads a day, with borrower rate signal captured in real time. Prompt to go-live, end to end, theirs.
The pattern, every time: domain owner + self-serve tooling + two days = thing-that-ships.
What we learned
1. The barrier was permission, not capability.
Business teams were not blocked by skill. They were blocked by the belief that AI is someone else’s job. The hackathon was a permission structure first and a build environment second. That mindset shift unlocked more than any technical improvement we had ever made to the platform.
2. Self-serve is a strategy, not a feature.
We had spent months making AIRO no-code. Prompt editors. Live test environments. Frozen eval sets that ops could promote against. That was not engineering convenience. It was the prerequisite for everything that happened that weekend. If teams had to file a ticket to change a prompt, none of this works.
3. Feedback velocity changes the relationship.
The traditional path is problem -> spec -> eng -> QA -> deploy -> monitor. Weeks, sometimes months. The hackathon compressed it to hours. Teams could watch a bot fail in test, fix the prompt, watch it succeed — same afternoon. The technology stopped being abstract.
4. The best use cases live with the people who own the problem.
Engineers are great at solutions. They are often terrible at knowing which problem matters most. Dealer ops knew exactly what was painful about their workflow. Lending knew which part of the lead journey leaked. Domain knowledge plus direct tooling produced use cases our AI team would have taken months to surface on its own.
5. AI adoption is a culture problem with a technology interface.
The hard part is not picking the right model or building the infra. It is convincing people the technology is theirs, that they have agency over it, and that starting is worth the discomfort of not knowing everything. The hackathon addressed all three. Not through training. Not through documentation. Through doing.
If your adoption feels slow
Stop waiting for the perfect tool to land in your team’s lap. Run a hackathon. Give people a real problem they own, a platform they can actually touch, two days. You will learn more about where AI fits inside your business in 48 hours than in six months of roadmap reviews.
The companies that win at AI adoption are not the ones with the best models. They are the ones where the most people, across the most functions, understand how to use the tools.
The companies that win at AI adoption are not the ones with the best models. They are the ones where the most people know how to use the tools.
That is a leadership problem with a leadership solution.
We are still early. But the way teams talk about AI now, the speed at which they build on the platform, the kinds of ideas they bring back to us — it is a different shape of adoption. Not top-down. Not spec-driven. Owned.
That is the version that sticks.