Building Chatbots for Enterprises: 5 things to keep in mind.
According to a survey conducted by Drift earlier this year, 15% of consumers have communicated with business via a chatbot in the last 12 months, and 47% of consumers would buy something from a chatbot according to Hubspot (2017).
With AI & Machine Learning advancing fast, one of the areas that will experience disruption in Enterprise ecosystem, is the Customer channel — with Chatbots.
With consumers already active on the popular messaging channels like Facebook, Twitter, Skype, Kik, Telegram, Whatsapp etc (depending on the geography), Chatbots are the obvious addition for Enterprises, to the traditional customer interaction channels of Physical Store, Call Center, Web store and the mobile app.
Chatbots not only reduce the cost to serve for a business process but since the conversational medium is more intuitive, they offer greater engagement and ease of use.
But truth be told, despite the obvious benefits, Chatbots are still not mainstream in the Enterprise IT world, partly because there is a learning curve involved, but mainly because most of the Chatbots & Chatbot building platforms are not yet ‘Enterprise Grade’.
What does being ‘Enterprise Grade’ mean for Chatbots / Chatbot Platforms?
Like all things chatbots can vary in complexity from a bot that one can build in 10 mins, which basically just captures contact details, post office hours, to Chatbots that assist Field Force Agents with their entire diagnostics processes.
While each organization has its own business imperatives and IT System standards, in our experience working with Tier 1 and Tier 2 Enterprises, organizations expect the Chatbot solution to meet certain key criterion. There are of course many, but I am listing down 7 key ones, just to get the ball rolling :
Since Customer Experience remains the paramount factor for Enterprise, a high accuracy is a must-have for Enterprises. Nothing is more embarrassing for the Enterprise, and more frustrating for a Consumer than the words ‘Sorry, I didn’t understand that’. And when it goes into a loop, trust me — at that point the user is never coming back to the chat channel again!
Although, earlier this year an AI model for the first time outperformed humans in reading comprehension (The SLQA+ ensemble model from Alibaba recorded an Exact Match score of 82.44 against the human score of 82.304 , on the SQuAD dataset.), NLP solutions are still evolving, mainly because its still very difficult for NLP systems to understand the context. The second hurdle is the absence of labelled data to train models.
So till the time NLP and MRC reach human-level performance in understanding context, an alternate option that offer far higher accuracy compared to NLP led text only bots is the use of Closed Domain AI ( preset context) Rich messages — with cards, buttons and various other UI elements in-chat. The not only provides far higher accuracy, but also dramatically reduces the overall cycle time ( less typing you see, and definitely less dead ends/wrong turns:P).
Two : Leverage the legacy setup — Human + AI hybrid
This is partly related to the point about accuracy, but there is a larger theme here.
All Enterprises of repute, probably are already huge invested & reliant on Contact center agents for Customer & Employee support. Couple that with the fact that Chatbots are also evolving — a step change into a Human + AI hybrid approach, makes more sense rather than an ‘AI only’ path ( atleast not yet).
In all our hashblu.io implementations, we enable an approach of chatbot + human chat user journey. Pre-integrated with most of the key Agent chat platforms like Live Chat, Live Agent, Intercom etc. whenever the user wants they can switch to talking to a human, who can assist them with a particular problem or query, and then hand them back to the chatbot to continue their journey. Another approach, of many such ways, is for the chatbot to engage and gather all the relevant data from the user, and then hand-off to the human to assist with the task in hand.
The hybrid approach helps reduce cost, keeps the user experience optimal and lets Enterprises leverage their customer support & social media support agents most efficiently.
Three: Data Privacy & Security
Consumers and Enterprises are very sensitive about data privacy and data security, and rightly so. Incidents like the Facebook + Cambridge Analytica fiasco, have exposed how vulnerable our personal data could be. Regulations like GDPR ( especially for EU & the UK) should help in ensuring Data privacy and Security, but when designing Chatbots and Chatbot building platforms — data privacy should be a key design parameter rather than an after thought ( and a mad rush to become GDPR compliant, as we are seeing with some of the Chatbot building platforms).
Here the simplest & sensible thing to do its to ensure that by design, your Chatbot platform / chatbot solution does not store any kind of customer private data & commercially sensitive data. Second thing is to provide complete transparency to the Enterprise, through an auditable interface, of all the data points that you are storing & using for your analysis. Plain & Simple.
Four : Supportability — its does not end with launching the Chatbot, it starts from there.
While most Chatbot solutions and platforms taut about the ease of building the bot, and supposed launch time being under 10 mins, as their USP, in my decade & a half of experience with Enterprise IT, I have realized that its not just the go live of an Enterprise system, but the continuous supportability, maintainability and the ability to make quick changes to the system that is equally, if not more important to the Enterprise.
This is where I believe the non-code based platforms, which allow for easy re-configuration, launching/ removal of services through a GUI based interface, rather than being entirely code driven have an advantage. This approach offers high agility ( for market events & campaigns), a higher degree of customization and a greater ownership of the solution from the Business aisle of stakeholders as well.
Five: Enterprise centric design & easy integration with core Enterprise Systems, allowing for richer Use cases
Unfortunately, most of the Chatbot building platforms are focused on enabling, what I think are relatively simpler & shallow use cases of After Office hours support, lead generation and ( structured) FAQ bots for instance. This could be a function of the learning curve & the willingness of Enterprises to gradually ease themselves in, into the Chatbot ecosystem.
But, I suspect, its also partly attributable to the fact that none ( at least the 10–12 I have seen) of the Chatbot building platforms have been design & built from the ground up, keeping the requirements of the Large & Medium Enterprises in mind.
This was the reason I founded hashblu.io, a platform purpose built for Enterprises, in terms of design, features and robustness.
hashblu.io platform, through the use of RESTful APIs and Web services, enables easy integration with backend systems, so that Chatbots can be used to expose and work in conjunction with core Enterprise systems like RPA ( Robotics Process Automation), Service Management, Field Force Management systems and of course CRM. This presents an option to Enterprises to also explore richer use cases, and bring out the same cost reductions & operational efficiencies, but at a much larger scale.
Needless to say, there are many more factors that influence an Enterprise’s decision when selecting a Chatbot solution. If you are keen to understand more, or if you are in the midst of a similar selection process yourself, please do feel free to reach out and my colleagues & I would be more than happy to assist you! ( yup! that was a blatant plug … but you gotta do what you gotta do ..right :P ).
I do hope you found the post useful in getting some basic understanding around building chatbots for Enterprises. As always, do leave your comments & thoughts — including any aspects that I might have missed. I will be more than happy to incorporate them.
Disclaimers: The above post in no way claims any copyright to any of the images ( except for the System & Mobile Screenshots, which actual screenshots of the hashblu.io solution) presented.