Launched a proprietary AI chatbot — $580K saved annually
How replacing a third-party chatbot vendor with a custom AI product saved $580K a year and hit 100K monthly active users in 3 months.
The problem
One of India's largest stockbroking platforms was paying a third-party vendor for a chatbot that couldn't handle the nuance of financial queries. It misrouted customers, gave generic responses, and was costing INR 50M ($580K) annually — for a product that frustrated users more than it helped them.
The business needed a solution that actually understood financial context, could handle high query volumes, and didn't come with a per-query vendor bill.
What we built
We owned the end-to-end product — from defining what the chatbot should handle, to building the NLP layer, to designing the escalation paths for queries it couldn't resolve. We ran structured A/B tests to measure containment rate (queries resolved without human agent) and iterated fast.
The chatbot launched in 3 months. It handled account queries, trading FAQs, onboarding flows, and basic support — all without routing to a human agent.
“Replacing a vendor with a proprietary system is always a product problem first, a technology problem second. The hard part was defining what 'good' looked like — what queries should be handled automatically, what should escalate, and how to measure success without alienating users during the transition.”
Have a similar problem?
Book a free 30-minute call — we'll map out exactly what's possible for your business.