Enterprises are continuously looking to dip their toes into the ever-expanding universe of conversational AI.
However, for ING - a Dutch multinational banking and financial services company - the mantra "nail it, then scale it" could offer an approach for others to follow.
Speaking to CX Today, Ayush Mittal, IT Chapter Lead at ING, said: "ING’s approach has been to "first nail it and then scale it."
"We are continuously monitoring the results of their conversational AI offering, with real-time feedback loops that help them identify sources of improvement.
A safe, secure, and controlled approach is our target for scaling conversational AI across other channels and regions.
ING’s contact center leverages a combination of vendor solutions and custom ING APIs, with Twilio providing voice, chat, and video capabilities.
Google Dialogflow serves as ING’s conversational AI provider.
ING Has Embraced the Next-Generation of Conversational AI
Conventional chatbots rely on natural language understanding (NLU) that matches queries with predefined intents.
As such, these bots can match customer queries to scripted answers.
However, if queries are open-ended or unexpected, or if important context needs to be maintained across multiple conversations, then these previous-generation tools will fail.
Thankfully, large language models (LLMs) are now enabling bots to scour trusted knowledge and data sources to resolve many queries autonomously.
He told CX Today: "Virtual agents have the potential to answer any question and provide a more natural conversational flow."
What’s more, provided the necessary guardrails and regression testing are in place. Mittal added:
AI-based chatbots can evolve dynamically, allowing responses to change based on new training data, fine-tuning, or parameter updates.
However, just because virtual agents can always give an answer, it doesn't mean that they always should, especially when there's little knowledge to drink from.
Cue the inevitable discussion around "guardrails".
Establishing Guardrails: How ING Does It.
While moving to dynamically generated conversations provides superior customer experiences, it also introduces risks related to accuracy, bias, security, and compliance.
This risk extends from the generated answers - alongside the data that the bot feeds from.
To that end, ING has introduced a range of AI guardrails, including model explainability, real-time monitoring and auditing, as well as confidence-based escalation.
"If a low-confidence AI response is given, then human intervention is triggered," added Mittal.
