As the need for automation increases, are chatbots living up to their potential? Not yet but they will…
Kaveer Beharee, CEO at Ubiquity AI, kicks off a series of insight posts on the use of AI chatbots in our communications. Here he outlines what is holding chatbot technology back and what they need to succeed. The answer might surprise you…
I am sure that everybody remembers the day they realised that the Covid19 outbreak would change their lives. But I am not so sure that people (me included) realised the extent to which the virus would disrupt our daily lives.
A few weeks ago, I contacted my bank to query a financial transaction. In SouthAfrica (where I reside), banks and other financial services providers were classified as ‘essential services’ during our country’s lockdown, so presumably it created the expectation that it would be business as usual.
Far from it. The call centre was operating on skeleton staff and most queries were directed to an online chatbot, which is quite functional for addressing basic banking queries, but then connects you to a live operator for more complex queries. Having been quickly shuffled to an operator, I was then informed that I was number 1176 in the queue. Unbelievably, having lodged the query mid-afternoon, I received a ping at 7am the next morning. I had a minute or two to respond, in which time I was unable to and subsequently lost the conversation thread, which was then closed.
Why are so many chatbots so bad?
I find it amazing and deeply depressing in equal measures that banks – who have been striving for more than two decades to reduce costs (as the sector is highly competitive) are not able to operate more efficiently in automating customer engagement.
If any industry in the world is capable of developing viable and working technology – in this case chatbots – you would expect it to be South Africa’s highly innovative banking sector, which scoops international awards and accolades year after year for global banking innovation. This is largely due to the industry’s incredible response to the many challenges relating to opening up the sector to much of the country’s population after being largely excluded from the formal banking sector during Apartheid. In short, banking in South Africa had to become simpler and more efficient as banks jostled for market share.
So the question now as the new Covid-19 reality is exposing is: why are banks and other consumer-facing industries so bad at developing functional chatbots that would obviously be highly beneficial for business, but more importantly, enable continuing engagement with consumers despite current and possible future restrictions on movement?
The missing skills in chatbot creation
A research paper published by Oracle, entitled ‘Can Virtual Experiences Replace Reality’, reveals that four out of five companies globally plan to have some kind of chatbot by 2020. This suggests that companies understand the value proposition of automation via chatbots, it is now a question about how to get it right.
The reality is that advances in artificial intelligence have improved chatbot capability in leaps and bounds since Alan Turing proposed the question in 1950: Can computers think?
In reality, however, there are mainly two technologies required to build a functional chatbot: natural language processing (NLP) and machinelearning. Herein lies the problem.
Machine learning and development skills are readily available. But natural language processing and linguistics skills are in extremely short supply. And the truth of the matter is that machine learning and artificial intelligence account for nothing if a chatbot does not understand natural language inputs, is capable of structuring and storing that data and is not able to execute commands via conversational inputs.
How can chatbots improve their reputation?
A few years ago, my previous company Ubiquity Consulting conducted the most comprehensive study ever conducted on South Africa’s banking industry by mining every snippet of conditional online data generated by customers. The study, driven by NLP to structure the data, opened the door to meeting with most of the country’s banks who were all keen to understand our methodology. To our surprise, not a single data scientist or business intelligence analyst in those meetings had a proficient understanding of NLP or any such skills.
While universities in South Africa, Europe, the US and other markets are producing NLP graduates at a reasonable pace, I suspect that the demand for these (and related linguistics) skills will significantly outstrip supply over the short and medium term.
In the interim, one of my main concerns, is that chatbot technology as a viable engagement platform, runs the risk of becoming commercially unviable for companies as poor adoption and retention rates gives the tech a bad rep.
For the technology to succeed chatbots must not only respond robustly to Turing’s ‘Can computers think’ question, which they can, but chatbots must also mimic the way humans communicate. I want to text a chatbot that communicates similar to the way I do and I want a chatbot to fulfil a query, or escalate it someone who can address it quickly, without me having to exert more effort. Herein lies the opportunity for chatbots to shine.
This will be a significant competitive advantage for those companies that get it right.
In my next article in this series, I will look at the objective and subjective factors for evaluating chatbots for commercialisation.
This blog was first published on Firehead’s website. Firehead is the the go-to place for European recruitment, consulting and training in AI and digital communications.
Leave a Reply