Fishbowl’s team of experts has been working with chatbots, specifically Oracle Digital Assistant (ODA) since its inception in 2017. Our expertise has been used to create chatbots with conversational design in mind to enhance the overall ODA experience. Through personalized branding, channels, integrations with Oracle Cloud HCM and Oracle Service Cloud, as well as other 3rd party integrations like Zendesk, we have been able to build and go-live on several successful chatbots with customers in the North American market.
That being said, our success has been earned through some bruised elbows and scraped knees. We have seen first-hand chatbot projects that fail. While such projects fail for different reasons, we have learned that it really comes down to 4 key factors. Factors that occur before, during, and after the project is “complete”. And chatbot projects are never really complete – see number 4 below. Instead, you should expect and want to improve your chatbot incrementally over time. Let’s get started with our list:
1. You can’t just turn a chatbot on and have it work.
· Conversational design is a skill and requires consideration and effort. Natural Language Processing requires knowledge, repetition, and multiple iterations to implement a successful chatbot.
· Each business and their use cases are different, so even out-of-the-box solutions will need some degree of customization. Without customization and understanding the verbiage and business language used by end users, the chatbot will miss intents and misunderstand users.
2. You don’t have good data.
· The chatbot isn’t smart enough to go and find answers on its own. (Well, it is but that’s not necessarily what you want, e.g. Microsoft’s Tay.) The chatbot is dependent on the data you give it, whether it’s good quality or bad.
· Not all businesses have access to the data and metrics necessary for a successful chatbot. Determining how users’ interactions with the bot went, if they finished the interaction, they quit out of frustration, the bot isn’t understanding what’s being asked or is missing colloquial phrases, are all data points necessary to improve the bot. Without them, you won’t know why or if users are even taking advantage of the bot.
3. People expect more from artificial intelligence than it’s currently capable of doing.
· End users seem to think that artificial intelligence is like having another employee that knows everything. It’s not. It’s only as good as you train it to be. Read the complete article here.
For regular information become a member in the Developer Partner Community please register here.