60-Min Investigation: Building a Chatbot

Published on: 24 June 2020
Written by: Tridant

Chatbots, also known as Virtual Assistants, are entrenched in the marketplace with many people having interacted with a chatbot in some way or other. It may have been to quickly check the balance of your superannuation, request the service hours of a business, or learn release information of a new product. All engagement comprising an essential part of the customer journey.

Chatbots are useful, fast and available 24/7. Some are sophisticated enough to make it difficult to differentiate whether it is a human responding to queries, or a machine.

I was impressed with IBM's virtual assistant offering, Watson Assistant. At Tridant, we are seeing increased queries from organisations keen to understand virtual assistant technology and witness firsthand the use cases and value that chatbots deliver to organisations.

How does it work?


We know some of what is possible with virtual assistant technology. How much effort is required? 

I say ‘some’ because every few months we see new applications of this powerful technology, demonstrating great innovation, results, and impactful use cases.

A real differentiator of the IBM chatbot offering, over other vendors and open source solutions, is that Watson Assistant is ‘easy’ to implement. We do hear ‘easy’ thrown around loosely by vendors at times, so I thought I would run a test myself.

The test was to see how long it would take to set up a basic chatbot to understand, and be able to genuinely advise or respond to, my customers.

Could a business really get a chatbot up and running quickly using this powerful technology?

A barrier for many looking to get started is getting hold of the software. In this case, it could not have been easier. 

I jumped onto the IBM website, searched for ‘Watson Assistant’ and selected the option for the ‘free version’. Yes, let me repeat that: it is a free version from IBM. Not trial, not time-limited, but free.

Of course, there are restrictions. The major restriction lists a limit of 10,000 messages per month. To be honest, that’s pretty good and likely adequate for many small- to medium-sized businesses.

Getting started


For the first 10 minutes, I tried to free-style my way through the set-up and considerations. Whilst there is a logical workflow to put a chatbot together, I needed some help from my mentor, YouTube.

After a 3-minute video on how to set up Intents (questions people may ask, such as ‘I want to place an order’), and another 3 mins on Entities (the specific pieces of questions, for example the number of items in an order), I found myself needing to watch a 15-minute video on how to then apply the questions into a dialog (the chat itself).

Now that I’m 45 minutes into my build, based on a pretend company selling hand sanitisers, I thought I would show off my handiwork to my wife.

The very first question she asked, “Is it organic?” returned the chatbot answer “I don’t understand. You can try re-phrasing”. Disappointing. Not so clever after all.
In all fairness, I hadn’t prepared my chatbot model for this line of questioning. 

I re-opened Watson Assistant and added the appropriate Intents and Dialogs, and kicked off the automated model training to build in the smarts.  My wife ran through a scenario of basic questions and received appropriate answers. Her customer experience was much improved, with higher response accuracy.

So, there you have it. In under an hour, I registered the software, found a free version, set up a model with a little help from a video tutorial, and deployed it to a website so that a customer could interact with it and gain desired formation.  

I even managed to build in a couple of jokes to ‘liven’ up the conversation.

Conclusion


My 60-minute investigation uncovered that the Watson Assistant proposition of being easy to work with is correct.

Where the effort is though, and we have seen this with customers and their use cases, is that training the model with relevant questions and responses enables a positive user experience, and that it performs with a degree of human understanding.

When utilising chatbots as a digital customer engagement channel, both long-term content planning and optimisation are critical. Keep checking that you are matching customer intents to answers, and if possible, enable live hand off to contact centre agents to help when the chatbot fails to answer adequately.

If you are looking to explore or prototype a chatbot within your organisation, trial how a chatbot could add both value and improved user experience for your customers, or to get your head around some of the fundamentals of AI technology, I highly recommend starting with the free version of IBM Watson Assistant.

Enable better customer experiences.

Zac Anstee | Michelle Susay

Related Articles

Copyright © Tridant Pty Ltd.

Privacy Policy
chevron-down