Personal Collection: The Surprising Results of a Small Adaptation

We believe in automation. But we also strongly believe in personal human interaction. That is why we did not build a chatbot. We built something better: AI that supports the human agent. We save the agent time-consuming searches and redundant typing, allowing him to do what humans still do best: feeling what the customer needs and dealing with emotions and outside-the-box requests. Based on years of data, our AI comes up with the best possible answer. The AI suggests, the agent decides.

Automation and personal style

But here is the thing: is the best possible answer also the best personal answer? If the answers the AI suggests are too generic, the agent might not feel at ease with the suggestions. One agent might say: ‘‘Hi, how can I help you," another agent will say: ‘Goodmorning, what can I do for you?’’ We may only want to recommend the phrase that suits the specific agent. If not, the conversation might still end up feeling generic and robotic. Many agents indeed wanted our autocomplete and reply suggestions to reflect their personal style more when asked.

So, could we help them?

Our mission is to assist agents in writing text as easily as possible, so we decided to give it a try and help them with personalization.

First, we adjusted the algorithms, with our AI giving more weight to the given answers by the agent in question. We compared our new models in an A/B test. Indeed, agents chose the new model with improved personalization more often. So we made the adaptation for all agents.

Our surprise with a new feature

Encouraged by the results of personalizations, we added another small and simple feature for one of our customers: Personal Collection. This allowed the agent to manage all his or her own standard responses, giving even more sense of control. It looks like this:

This time, when we released our updated version, something amazing happened. Agents started to use Deepdesk more. A lot more. The chart below shows the use of Deepdesk by agents both in absolute terms (blue columns) and expressed as a percentage (white line). We released our new version at the end of December. We saw a rapid increase in use:

Sense of control and personal influence

We are excited over the results of course. And we learned two important lessons:

1. Clearly, having a sense of control and personal influence does motivate agents to use Deepdesk more. Since we are working with humans, in hindsight, that should not have come as a surprise.
2. We work hard every day to improve our AI. But we didn't expect the outcome of this relatively simple new feature. As it turns out, sometimes, small adaptations have the biggest impact.

Read more about what motivates - and demoralizes - contact center agents, and feel free to check out our other features.