Scaling Behavior Change Across 1,000 People
Combining machine learning and cognitive psychology to drive behavior change at scale.
Originally published on the Cresta Blog.
“Run a mile in under 6 minutes.”
“Meditate once a day.”
“Eat avocado toast every morning.”
These are some of the responses when we asked engineers at Cresta about their goals. However, adopting new habits is challenging, even when we know it’s the right thing to do.
The same challenge applies to the 3.4 million contact center agents in the United States. Zooming into the floor of a contact center, we meet Bob, an agent helping customers find the right internet service at Cox. Bob’s goals include “Increase my conversion rate” and “Make customers happy.”
He takes around 100 chats a day, and sometimes he brings the deal home, but other times the customer leaves the chat — often for reasons he doesn’t know. How can we help Bob better guide customers to the right solutions and increase his conversion rate?
Enter Cresta’s Behavioral Engine, a machine learning system that drives behavior change at scale. Combining state-of-the-art machine learning models, cognitive psychology, and human-centered coaching, the Behavioral Engine helps contact center agents grow into experts with every chat they take. The 2-part system, described below, not only helps an agent like Bob achieve his goals, but ultimately drives success for his manager and company.
1. Hints: Automated, Real-time Coaching for Every Chat
An agent receives a Hint based on what an expert would do in this situation.
Today, Cresta powers thousands of agents handling over 40,000 chats every day. To provide custom coaching tailored to every agent, our machine learning models (BERT, GPT-2, and T5) diligently observe each chat in real-time and determine the best thing to do at any given moment.
During key moments that shape the trajectory of a conversation — for example, a customer asking about the right package for their needs, or objecting to the price of a service — Cresta’s Behavioral Engine steps in and guides agents with best practices to handle these critical moments.
Hints prompt agents to seize an opportunity and positive feedback reinforces the learning when the behavior is performed. Over time, agents learn to ask the right discovery questions and address objections with empathy and ease.
In the past year, by advancing our AI models, we have scaled the volume of hints by a factor of 5, while simultaneously reducing cost per hint. As of today, we have sent over 2 million hints — providing agents with 2 million custom coaching touchpoints for their personalized growth.
2. Agent Progression: Automatically Track Behavior Change Over Time
Just like we are motivated by seeing our mile times go down and our meditation streaks go up, agents and their managers find value in similar measures of progress. How has an agent’s empathy or customer discovery skills changed over time? When agents receive a Hint, how often are they adopting the suggested behavior?
To visualize behavior change, the second component of Cresta’s Behavior Engine is Agent Progression: a comprehensive analytics engine that tracks the behaviors performed in every chat. With progression, it’s simple to drill into a particular agent or particular behavior and see progress over time. Access to these insights allows agents to understand their personal improvement and prompts growth-oriented conversations with their managers.
Our latest edition of Agent Progression is powered by our AI models at every turn. Previously, we provided insights on a sample of chats per agent. Now, insights are available on every chat an agent takes. With our current volume of 40,000 chats per day, this means over 40,000 insights are aggregated into a few key metrics for agents and their managers.
Additionally, we introduce an hourly view to complement the standard weekly view. As our AI models generate insights dynamically, managers can see real-time data on the behaviors agents exhibit at any given moment.
Measurable Results
To date, Cresta’s Behavior Engine has driven the adoption of hundreds of expert behaviors across thousands of agents in diverse domains (including telecom, retail, and software), resulting in measurable value for our customers, ranging from 12% revenue growth to 19.1% increase in NPS (for more, see case studies).
Behavior change is challenging, whether you’re an agent trying to adopt an expert behavior or just someone trying to eat avocado toast every morning. With Cresta, it can be visible, measurable, and scalable.