The world of conversational AI is certainly growing, and the traditional problem of any client is to compare different solutions offered in the market.
Here there are the 5 key points to orientate yourself in the first evaluation of a chatBOT solution that provides paybacks and image.
- A chatBOT must be a tool that allows you to implement conversational processes in a simple way.
It must be intuitive, rich in templates and micro-functions that can simply be configured and used. In general, who designs BOT services is not a developer, but someone who deals with business, marketing or customer care processes. Being independent in creating your own services is essential.
- “I would like my BOT to understand the natural language” this is more or less the sentence that we have to understand.
The expectation is often betrayed by the result, because typically those who sell these solutions, trust in ever-green “overselling”, except then having to deal with the harsh reality. A good BOT must understand well in its context of the application, so it certainly helps to have a good understanding layer on the specific language, perhaps associated with a wealth of concepts related to specific industries (finance, telco, …). For example, if I speak with my bank’s BOT and I say “card”, first it must understand that I am leaving a credit card, but if I say “card” to my phone company’s BOT, it would be more correct that I’d associate it with the “Telephone card”
- And then… the BOT is online, I understand well. It’s even nice because I gave it a nice character. But, how do you really perform? Can I check its efficiency? A BOT without efficient control tools does not allow me to understand how it is performing, if the services are inefficient or if the customers are unsatisfied. A BOT solution must have a layer of analytics of the highest level, which measures the effectiveness of the conversations, the ability to manage customers independently and an index that measures the effectiveness of the NLP layer
- Looking at the data gives me information about the effectiveness in order to understand. But, the great value is being a click away from marking, improving. I can really see a service, evaluate the adherence to my expectations and understand if something is wrong. I can improve it with a click, review the processes if needed or add intent to the domain of competence that helps to better understand a question.
- Finally, an often overlooked topic to the detriment of the focus on the NLP part: the effectiveness of the management of the different front ends is essential. Today the front ends are many and different, and every 6 months they change. Once you have created a process, it must be easy to export to different front ends (web, mobile, WhatsApp, FB Messenger, Alexa or Google Home, …) without comparing the complexity of an adjustment. An application that is independent in using the capabilities of web interfaces to deliver images, videos, menus to Carosello or buttons, but at the same time, it could guarantee the effectiveness even on text-only interfaces, or only voice, without affecting the service.
Eudata developed Convy AI in 2016. After more than 40 projects for large enterprises, Eudata proposes a platform that, combined with a clear methodology of development and evolution, guarantees paybacks in a short time.
For SMEs, Convy is also available on Amazon AWS directly from the market place.