After 60 years of broken promises, AI is finally delivering. Backed up by the massive parallel processing capabilities of the cloud, and advances in algorithm design, machine learning and Natural Language Processing, we are at the point where machines can understand – and sensibly respond to – around 95% of our everyday speech.
And that’s good enough to see products based on AI and cognitive computing break out of the tech labs and into our workplaces, pockets and kitchens. Amazon Echo – the smart speaker with the Alexa AI on board – is now the most popular item on Amazon and is already embedded in over eight million homes. Siri and Google Now, of course, are on hundreds of millions of smartphones.
Automation is inevitable
If consumers are ready for AI products and services, what opportunities does this represent for businesses? BI Intelligence estimates that 29% could be shaved off the cost of customer care with only moderate levels of automation – while actually improving service.
Gartner expects 85% of customer interactions to be automated by 2020, while Oracle suggests that 80% of brands will be using chatbots by the same year.
Out of 1,000 UK consumers interviewed by research company myclever 46% thought chatbots would unlock “immediacy” – getting a response to a question or issue at once – and increase the convenience of online services. 40% said they would use chatbots as a way of connecting to the right customer service agent for their enquiry.
Speed and impatience are two factors driving companies to digital transformation, with 71% of customers demanding the ability to solve most service issues instantly. The chatbots that will become most successful will be those that resolve customers’ issues accurately while meeting their needs for speed and convenience.
The challenges of deploying chatbots
Even though it is a relatively new sector, the ecosystem for chatbots is already quite complex, with many options to choose from. These include service-specific platforms such as Facebook’s wit.ai that only lets you build a bot for Messenger. Other platforms from Microsoft, Google – and our own recently launched IMIbot.ai – enable you to build bots for multiple services at once.
It’s important to differentiate between services that allow almost anyone to build and deploy a chatbot, and solutions that are enterprise grade. Only the latter have all the things that a business needs if it is going to rely on a chatbot to take a frontline role with its customers. These include the ability to scale, operational management tools to make changes quickly, detailed reporting and analytics, and integration with other business systems and processes.
We suggest using the following criteria for evaluating chatbot platforms before you start with any implementation:
1. The ability to understand natural language.
There’s nothing customers find more frustrating than getting answers to the wrong questions. A chatbot that doesn’t get it right first time, or is unable to clarify something it doesn’t understand, leads to a disruptive service experience and lower customer satisfaction scores.
2. A solid process for dealing with questions it can’t understand.
It’s inevitable that bots will sometimes get things wrong. What’s needed is the ability to ask questions to clarify meaning. Simple conversation-tree style bots don’t really have this ability. Those based on machine learning and NLP do, but to varying extents.
3. A level of empathy/humanisation including intersection design.
People still like a human element, even if they know they are talking to a robot. On the other hand, it is possible to push this too far and create a bot that sounds like it’s trying too hard to be human – for example, by asking irrelevant, personal or jokey questions to try to seem friendly. Studies have consistently shown that this has the opposite to the intended effect and puts people off.
4. The ability to integrate with other business processes, software and platforms.
In all cases, the ability to quickly hand over to a live agent before frustration sets in is essential. For orchestration purposes, the integration of the bot service with the company’s full suite of channels becomes crucial at this point. Ensure that the chosen solution allows you to make the integrations necessary to deploy a robust, enterprise-grade solution.
5. The ability to configure and make changes speedily.
A chatbot is only as good as the data and knowledge it has to work with, no matter how fancy its machine learning algorithms. Without having ingested the whole of Wikipedia, and millions of other sources, IBM’s Watson would never have been able to win at Jeopardy. Particularly in the beginning, you will need to make constant adaptations and upgrades to your chatbots, changing their decisions trees, scripted responses, and knowledge bases.
6. Clear set-up and on-going maintenance costs.
Once deployed, a single chatbot works 24/7 without breaks and can handle unlimited interactions simultaneously. This is what makes them significantly more cost-effective that live agents. But that doesn’t make them free. Depending on the provider you choose, there will be set-up (such as bot development) and maintenance costs (service management and operational support), telecommunication costs (e.g. text massaging), third-party API costs (number of times your service calls an API), and so on. Understanding all the costs involved is crucial to evaluating ROI.
7. Flexibility beyond supporting single messaging channels or AI engines.
Many chatbot platforms support proprietary services, or just a few messaging channels. For example, creating and deploying a skill for Alexa is easy enough with Amazon’s own developer kit, but it won’t work anywhere else. Chatbots built on multi-service platforms can be deployed across many different applications and channels.
Finally, do something this year
While we don’t expect large volumes of customer interactions to move over to chatbots within the next year or so, it’s important to start dipping your toe in the water. Most companies know they need to be using them sooner or later, but they haven’t yet discovered how they slot into their business model.
The key is to begin the learning process as quickly as possible. For example, this year you could identify one or two simple use cases, such as routine enquiries, scheduling appointments, or handling repeat orders, and deploy a bot to automatically handle them.
While you should probably ensure there is always a hand-off to a live agent option, anything you can do to free up agents from handling routine queries allows them to focus on solving thorny problems or delivering genuine added value for customers.
By running a few small pilots this year, you can gather information on performance, costs, savings, and the impact on customer experience measured by NPS, CSAT, customer effort or whatever metrics you employ.