How CRED is tapping AI to deliver premium customer experiences
CRED serves India's most creditworthy consumers-people who expect trust, security, and exceptional design in every interaction. As the product suite grew, so did the need to keep context tight and decisions fast. The solution: go AI-first without compromising quality, compliance, or brand experience.
As Swamy Seetharaman puts it, "The question we ask is, how do we make every single member in every function be 10X? For us, AI has become a huge unlock in that journey."
Why CRED went AI-first
More users, more teams, more data-the compounding effect made it harder to stay fast and accurate. AI became the layer that preserves context at scale. The goal wasn't to replace people; it was to make every function-support, operations, engineering-far more effective.
That mindset shift matters for support leaders: measure AI by how it improves precision, empathy, and speed across the entire support system, not just ticket deflection.
The stack: Cleo, Thea, and Stark
CRED built three AI tools to cover customers, agents, and operations. Each one removes friction in a specific workflow, but they work best together.
Cleo: AI conversational companion (powered by OpenAI models like GPT-4.0, GPT-5, o3)
- Informational: "What is CRED Cash?"
- Contextual: "Am I eligible for CRED Cash?"
- Transactional: "Can I refund to my wallet or original payment method?"
Cleo diagnoses the issue, classifies intent, maps to the right SOP, and responds with accurate, contextual guidance. It's built to handle multi-intent conversations without losing the thread.
Thea: Agent co-pilot
- Summarizes multi-format conversations (text, voice, Hinglish).
- Surfaces next best actions aligned to SOPs and policy.
- Reduces manual switching and context rebuilding for agents.
Stark: SOP engine for operations
- Creates or updates SOPs in minutes instead of days.
- Closes gaps between policy and practice by pushing updates where agents work.
- Keeps Cleo and Thea aligned with the latest rules.
What changed: outcomes that matter to support
- CSAT: +14 percentage points. If CSAT is your north star, this is the signal you want. For context on benchmarks and framing, see this primer on CSAT scoring.
- Resolution accuracy (Cleo): 98% within three months of launch.
- Multi-intent resolution: +18% improvement.
- Average handling time: Down across Cleo, Thea, and Stark.
- Session drop-offs: Down 31%-fewer abandoned experiences.
This is the pattern to aim for: higher accuracy and empathy, lower effort for customers and agents.
What surprised the team
Initial skepticism faded once people saw results from CRED's internal evaluation framework (which also uses OpenAI models). Adoption followed outcomes. The real unlock was psychological-teams realized they could move faster and stay right.
What's next at CRED
- Expand Cleo across all business lines with domain-aware logic.
- Detect "data dead-ends"-when the system can't answer-and loop them into the knowledge base automatically.
- Make every function 10x more efficient across engineering, QA, infra, and compliance by reducing context loss and decision lag.
"With the help of AI-we're moving closer to our goal of creating a true concierge experience built on trust, reliability, security, and exceptional design."
A practical playbook for support leaders
- Start with intent clarity: Map top queries into informational, contextual, and transactional. Build prompts and SOPs per intent.
- Close the loop: Use an internal evaluator to score responses for accuracy, empathy, and compliance. Feed misses back into training and SOP updates.
- Support the agent, not just the customer: Ship an agent co-pilot that summarizes context and suggests next steps. Time saved here lifts CSAT and FCR.
- Treat SOPs as code: Version them, update fast, and propagate changes to every surface (bot, agent, help center).
- Measure the full system: Track CSAT, first-contact resolution, multi-intent success, AHT, transfers, and drop-offs.
- Respect multilingual reality: If you operate in mixed languages, train for it (text, voice, transliteration, tone).
- Align with values: Make security, reliability, and transparency non-negotiable. Efficiency means nothing if trust erodes.
Advice for hesitant teams
Decide what matters most-efficiency, effectiveness, or both-and build around that. Pick a high-volume slice, define success, and ship a small system that closes its own gaps. Show results, then scale.
That's how you get buy-in without a 6-month business case.
Key takeaways you can use this quarter
- Segment intents and wire SOPs to each type. Let your assistant route with confidence.
- Give agents a co-pilot for summaries and next steps. Reduce cognitive load.
- Automate "dead-end" detection and turn them into knowledge updates weekly.
- Obsess over accuracy and empathy before deflection. CSAT follows.
Want to upskill your support org on AI?
Explore curated learning paths by role and skill to accelerate rollout and adoption: AI courses by job. If you're just scanning options, start here: Latest AI courses.
Your membership also unlocks:
