Very much like the term hacking, our notion of software bots used to be a wholly negative thing. But, just as hacking became the new label to denote the actions of cool nerd hackers building our software-driven future, bots graduated from malevolent virus spewing botnets to become a new type of helpful bot designed to make our lives better. As a shortening of ‘software robot’, a bot is simply a chunk of code used to perform a functional service task, usually online.
Throughout this decade we have seen major technology vendors embrace the idea of helpful bots that can work to manage, channel and even clean the data inside our software systems. Using technologies that harness so-called Natural Language Understanding (NLU), we have subsequently been able to build bots that can disassemble and parse human-generated text into categorised semantic meaning. These are bots that can understand us and talk back. These are the chatbots.
Chatbot brain development
As the chatbot rapidly speeds through its adolescence, we (the humans) are dedicating ourselves to ensuring that these software brains are capable of learning. As we start to apply new layers of cognition and understanding to the chatbot’s talkback response mechanisms, we can then use chatbot technology at the core of our customer service processes to significantly improve customer experiences at all levels. Applied carefully, the chatbot is your new front line and it’s good news for the bottom line.
Applied carefully, the chatbot is your new front line and it’s good news for the bottom line
But Artificial Intelligence (AI) education for chatbots is no plug-and-play affair. We cannot simply flip an ON switch and expect these machines to learn. As the number of chatbot and virtual customer service agents grow, so do the number of failed projects and frustrated customers. Cast your mind back to the ‘hello caller, I think you said’ telephone-based speech recognition booking services that existed before the millennium. We can’t afford to let smart chatbots be that dumb.
Complex questions, smarter chatbots
Chatbots only communicate as a result of the information that we program them with at birth. Smarter AI chatbots only learn and train by exposure to the diversity of information in the data pool that they are exposed to. A shallow pool of similar data events creates a comparatively more narrow-minded chatbot.
Too many AI projects fail when they are developed and deployed in isolation from a firm’s contact centre
Many simple customer service tasks lend themselves well to automation, but as customers ask increasingly complex questions, how do we ensure that we deliver the empathy and expertise required to maintain high levels of customer satisfaction? Too many AI projects fail when they are developed and deployed in isolation from a firm’s contact centre. This is the hub that forms the heartbeat of human interaction, extending as it does from inside the business, outwards to customers and partners. So exactly what is required to deliver a seamless integration of automated and assisted customer service?
The four cornerstones of bot
Enterprise applications software company IFS-mplsystems provides four cornerstones for bot creation designed to produce AI-driven customer service services that work effectively and accurately:
- Narrow, then broaden: Any level of bot development should ‘start narrow and then broaden later’. No brain (human, or computer) is good at drinking from a firehose on day one and both brains need to be given appropriate learning space.
- Shared chatbot-human DNA: Chatbot virtual assistant AI should never be considered as a wholly alternative replacement channel to human customer service. Rather, it should be embedded in every part of the customer service channel to deliver intelligence at the ‘front end’ of every conversation or interaction.
- Seamless transfer: When a request is too complex for the chatbot, it should seamlessly transfer the conversation to an appropriately skilled human agent. But, crucially, each of these more complex query resolutions should then be fed back to the chatbot in the form of data for onward learning.
- Be iterative, continuously: Make bot development a truly iterative ongoing process that develops organically inline with the businesses’ commercial goals and market strategy. Further here, bots should not just answer questions, they also need to process requests, in just the same way that we want human agents to actually solve our problem not just advise us when we contact a call centre. It’s not just about AI, but also about solutions that combine Robotic Process Automation (RPA) to automate tasks.
So, like a good customer service bot, let’s stop and ask ourselves what we’ve learned so far. We know that successful modern commerce is driven by emotionally-rich interactions with companies who understand the value of ground level collaborative engagement with customers. We also know that chatbots can form an invaluable ‘front end’ interfacing layer that positively enhances customer experiences, if engineered and built conscientiously.
That might sound like a strange term to apply to software programming, but conscientious chatbot development is the careful application of AI as part of a multi-channel customer journey, not added on as a separate silo or an alternative channel. Conscientious chatbot development also understands that chatbots need time to learn on the job; and they need a staged approach to integration into business operations, as they get smarter. Too many contact centres have got ahead of themselves by creating quick-fix chatbot software without conscientious engineering.
“Every year the volume of interactions coming from customers increases by almost 10% and every year the number of channels that customers use grows. A few years ago it was phone calls, email and webchat -- now it’s Facebook, Twitter and messaging channels such as WeChat and WhatsApp. For a business to continually grow the contact centre to soak up this volume and complexity is not sustainable. Businesses have no choice but to introduce AI in the form of chatbots and virtual assistants to help out. But to deliver an excellent customer experience, not just control costs, theses chatbots must not be viewed as an alternative channel, they need to be an integral part of every other channel, providing intelligence at the front end and seamless proactive handoff to an empathetic human when it makes sense,” said Susannah Richardson, marketing director, IFS-mplsystems.
10,000 hours of training
While chatbot development success will be largely governed by the realities and best practices thus far discussed, no virtual assistant will ever be very capable until it has a rich, deep and varied pool of data upon which to draw experience from.
No virtual assistant will ever be very capable until it has a rich, deep and varied pool of data upon which to draw experience from
Social philosophy author Malcolm Gladwell suggested that humans require somewhere near 10,000 hours (just over a year of continuous time) to become an expert in something. IFS-mplsystems Richardson echoes this assertion by saying that Even for something as simple as requesting a balance of your account there can be over 1000 ways to ask that question and that equates to a huge amount of sample data require to train your bot to answer a single simple request.
Chatbots, just like humans, need training. They also need carefully applied and intelligently graduated levels of experience while being fed on rich streams of quantified and qualified data. If we can give our new virtual team members this kind of education and work experience, then they can emerge from adolescence as truly productive and positive members of society. Let’s never forget, an effective chatbot is a happy bot… and it won’t be long before they are reminding of us the fact.
For more information on implementing chatbots in your contact centre, download the report here
