Vodafone were keen to explore how AI could level-up their online sales. TOBi was already a known brand within its network for customer assistance. TOBi sales provides rich interactive experiences inside the chat to streamline purchasing plans through conversation.

What shipped

• TOBi Sales chatbot
  • Conversational assistant to aid customers in purchasing, powered by IBM Watson
  • Realtime conversation analytics provided by IBM platform
  • Model training configuring entitities, intents and decisions

Impact

• ↑ Conversion: As much as 3x increase in conversion rate from basket to checkout vs website
• Faster purchases: 47% reduction in time to checkout vs website
• Time-to-market: First end-to-end sales chatbot in Telecomms
• Data-driven: Conversation analysis to retrain flows
• Award-winning: TOBi has won 3 major industry awards
  • AI Project of the Year (Computing Digital Tech Leaders Awards)
  • Best Custmer Experience (Marketing Week Masters)
  • Most Effective use of AI for Business Optimisation (Campaign Tech Awards)

My role & scope

• Tech Lead on a 3-person delivery team
• Designed and implemented the architecture (AWS + IBM Watson + React + Node) to host the chatbot at scale for Vodafone
• Configured the entities, intents and decisions in IBM Watson
• Monitored and improved conversation branches to steadily improve the chatbot in production as it was used by thousands

Key decisions & trade-offs

• Sticky sessions in AWS: Users retained their conversation to an individual server in MVP to reduce time to market but increase scaling complexity.
• Limiting actions in conversation: If you were mid-checkout, we opted to limit what you could say in conversation to avoid setting up customers for failure - you could either proceed or cancel.
• Deciding scope of bot: We focused solely on sales for our conversation decisions and had to balance it carefully to avoid customer frustration.

See it in action