Chatbots and How We Stumbled upon an Award Winning Opportunity
We never set out to make our presence felt in chatbots. We just stumbled upon it in the middle of 2017.
We were not the pioneers of this field, and honestly we were pretty much the last to enter this scene. What helped us was that we had a client who wanted to explore chatbots and we were eager to see them through.
We were new to chatbots and hell, we were even new to technology back then. We were a bunch of consultants trying to help our clients solve business projects. But we always believed that any problem could be solved with common sense and customer focus. And so we set out to take it step by step. We first asked our customer to shut-down what was delivered by our predecessor. It was a hard choice to make. Shutting down something that was live, would open doors for other chatbot players in the market. Which it did. However it was a wise choice since the bot was not helping neither the customers nor the agents with what they wanted to do. It was adding workload to everyone. If I had a chance to go back in time, I would still have recommended to shut the bot. Yes, it was our baby, but we were ready to sacrifice it for our customer.
After shutting down, we re-looked at what were the major challenges. And funnily enough, the challenges were not chatbot at all. The bot was getting things wrong sometimes, which is expected. What was not thought about was the fact that the agents could see it in real-time and wanted to help the customers. So they stepped in to help the customers and the bot would also keep responding at the same time. Which made the whole conversation messy. The first work-around was to shut the bot when it sees that an agent was talking to the customer. A simple work-around which saved many conversations. We also included ways for customer to shut the bot and talk to an agent directly; and also for the bot to shut itself. With these simple work-arounds, which had nothing to do with bots but with process, we recommended to go-live again. And so we launched again in Sep 2017.
It was a simple English menu driven bot with basic integrations to billing and CRM databases. Even then we learnt a lot supporting and measuring the performance of the bot. Some of the key parameters that we use to measure our bot are
- Bot-only conversations — We made many types of graphs and parameters. Of all the things what stood out was the bots ability to handle customers on its own. The bot could only handle 6.3% of the conversations as you can see below.
I agree that a 6.3% of bot-only conversations doesn’t sound like much, but in the bigger scheme of things, being able to measure ourselves was a huge leap. We could set out next version targets and also work on how to improve these. And also knowing what parameter is paramount in defining bots was also something that we realized through this.
A 6.3% of bot-only conversations doesn’t sound like much, but in the bigger scheme of things, being able to measure ourselves was a huge leap.
- Conversational ability — Most people generally like being listened to than being bombarded with. We realized that our bot was dumping information and the customer was displaying a typical TL;DR behavior. So we set out to measure and work on improving our conversational ability.
We were hitting a 10:1 ratio which is way higher than having a meaningful conversation. In our opinion, a healthy ratio is 2:1 and I hear that we are hitting that number with our recent launch.
As you can see from the stats, the bot has a lot to learn and a lot of improvements are required. But nobody ever starts with a perfect product and we are very grateful to be recognized for all the hard work. Cheers to the team.
Please drop your suggestions and feedback in the comments. Thank you for reading.
