.png)
ING
Banking · Netherlands
20% more customers helped
ING builds GenAI chatbot in 7 weeks that helps 20% more customers immediately
Built with risk stakeholders from day one. Blueprint to scale to 37M customers across 10 markets.
Friction
ING Netherlands handled approximately 85,000 customer contacts per week. The classic chatbot resolved 40 to 45% of chats; the rest needed live agents with wait times and office-hours limitations.
Breakthrough
A bespoke GenAI chatbot with risk stakeholders involved from day one and explicit guardrails (no advice on mortgages or investments). Built for safe operation, not just functionality.
Impact
20% more customers helped within the first 7 weeks compared to the previous solution. Blueprint to scale to 10 markets, potentially reaching 37M+ customers.
Unlock the full analysis with breakthrough, impact, what made it smart and its technical approach bellow!
Problem
ING NL received approximately 85,000 customer contacts per week. The previous chatbot resolved only 40 to 45% of chats. The rest required live agents with wait times and office-hours constraints.
What made it smart
Risk and compliance stakeholders were involved from day one, not added later. This enabled faster governance approval and safe scaling. The multi-step answer pipeline was built for precision, not just fluency.
Technical approach
Multi-step answer pipeline with knowledge retrieval and ranking of possible answers on helpfulness. Disambiguation when multiple plausible answers exist. Guardrails before every response is sent. Practical approach: start with one high-volume intent set, release to 10% of real customers, measure deflection and resolution quality, then scale.
Strategic lesson
The fastest path to GenAI in regulated services is governance-first design, not as a constraint, but as the enabler of speed.
Reflection question
How long would it take your organisation to go from AI idea to live customer interaction? And what does that timeline say about your governance model?