Why CRED Turned to OpenAI for Cleo
Analytics India Magazine (Mohit Pandey)
CRED’s tryst with AI has been years in the making. Ever since the fintech company integrated AI in its CRED Protect feature four years ago to detect fraud and billing errors in credit card statements, the team has steadily stitched a nest of predictive analysis and real-time insights to offer a more premium customer experience.
The result was the Svalbard suite of features released earlier this year. But as its product lines grew, the challenge was no longer scale alone.
For an organisation to turn AI-native, it takes much more than just augmenting customer journeys with embedded AI. And Swamy Seetharaman, CRED co-founder and its head of engineering, knows that.
AI is becoming central to CRED’s functioning, and the executive pitches for AI to make every team member 10x more effective by giving them clarity, judgement, and context that helps with execution.
“With AI, this is no longer an abstract ambition; it is simply how we work,” Seetharaman told AIM. And OpenAI is helping the team shape its customer approach and product experience with its models.
Building Cleo with OpenAI
AI now sits horizontally across many of CRED’s workflows. Seetharaman said repetitive tasks are automated, context is easier to assemble, and teams move ideas into production faster. He frames AI as a steady reasoning layer that reduces friction and closes gaps—essential for growing companies like CRED, which boosted its revenue by 71% in FY24 to Rs 2,397 crore on the back of member engagement and monetisation.
The company developed Cleo as an AI companion to handle simple customer conversations and alleviate customer-facing teams. Built on OpenAI models, including GPT-4.0, GPT-5 and o3, Cleo interprets free-form messages, maps them to standard operating procedures with more than 97% accuracy, and executes them using APIs.
The turning point, Seetharaman said, came when it began handling multimodal voice and text conversations with multiple intents in one thread. This is where traditional bots struggle to push beyond scripted replies.
It began to diagnose, act, validate, and adapt in real time. “This shift, from answering questions to diagnosing, acting, validating, and adapting in real time, transformed Cleo to operate like a true concierge,” he said.
The impact shows up in numbers. According to the company, there has been a 31% decline in session drop-offs, a 14% rise in CSAT (customer satisfaction score), and more multi-intent conversations resolved in a single flow.
CRED also tracks repeat impressions, clarity of responses, and how often users need to add context again. All these markers improved as Cleo’s tone and helped its language flow naturally and more in tune with the company.
Being AI-First
While OpenAI models run customer-facing support, CRED has tapped into Anthropic’s Claude to strengthen its internal processes. Seetharaman described the approach as treating AI as a toolkit. They are picking the right model for the right task.
“Claude strengthens our engineering muscle and helps us ship features faster with higher reliability. We are building internal copilots on Claude that support developers with safer code generation, structured reasoning, and automation workflows,” he explained.
In a blog, Anthropic wrote how developers at CRED now rely on Claude Code to identify incremental solutions for writing, testing, and committing code across both new and existing projects. The team also uses the tool to generate documentation for existing codebases, in addition to breaking down complex problems into manageable steps.
Read: How CRED’s Use of Claude Code Signals the Future of Software Development
CRED rebuilt its customer success stack with three systems: Stark, Thea, and Cleo. Seetharaman describes Stark as a tool that turns complex SOP creation into plain English workflows, optimising manual tasks.
Thea supports agents by giving them full conversation context, insights, and recommended actions. This matters as the company works with more than 20,000 APIs, and Thea saves agents from hunting for information across tools.
Cleo then understands member intent across informational, transactional, and personalised cases, and handles multiple needs within a single flow.
As these AI systems learn with each interaction, they enable CRED to track customer behaviour more closely. “When a conversation moves from AI to a human, the handoff becomes a source of insight that helps us identify new intents and onboard emerging use cases much faster,” Seetharaman explained.
The shift to an AI native organisation runs deeper than customer support. CRED is building foundational platforms that handle access, governance, security, privacy, compliance, and evaluations at scale.
Read: Kunal Shah On Why India Should Treat AI like WhatsApp and LED Bulbs
Looking ahead, CRED plans to integrate Cleo across all business lines. It is also developing tools that detect data dead ends and feed those cases back into the knowledge base. The focus is on raising accuracy and reducing blind spots in real time.
“For us, incorporating OpenAI’s technology has been a true unlock across two of our values: compounding and being fast and right,” Seetharaman said. The early signals have been strong. The next steps are about scaling that impact.
The post Why CRED Turned to OpenAI for Cleo appeared first on Analytics India Magazine.
Generated by RSStT. The copyright belongs to the original author.