Are AI-Enabled Chatbots Ready for Customer Service?

People like to say that necessity is the mother of invention. But, as often as not, an invention is born ahead of its time, and it gets shuffled around to different foster homes before it’s true necessity comes along to claim it.

While invention and necessity are related, they’re rarely directly related. And more often, an invention is around long before its necessity becomes apparent.

Artificial intelligence (AI) is one such invention. Poorly defined, rarely understood, AI has failed to meet expectations in a range of fields—though it could be said that high expectations were the problem.

Could chatbots be the must-have necessity that AI has been waiting for?

AI: Then and Now

If you’ve engaged in any discussions on the topic of AI, there’s a good chance you didn’t arrive at any solid conclusions. This isn’t your fault. AI discussions often fall victim to the same problem: a failure to begin with a clearly defined term.

People have very different ideas about what counts as AI, and even experts have had to reign in and qualify their definitions more than once.

How was AI originally defined? For that we’ll turn our clocks back to August, 1955 and check in on a Dartmouth summer research program for computer scientists. This program was the first formal attempt to get the AI ball rolling, and it provided an early definition of AI that some still believe in. The program was described like this:

“The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”

Now 60+ years later, it’s only fair to ask: How’d that go? The best answer we’ve seen comes from Gartner Fellow and VP Tom Austin in the report “Smart Machines See Major Breakthroughs After Decades of Failure” (content available to Gartner clients):

“Back in 1955, researchers assumed they could describe the processes that make up human intelligence and automate them, creating an artificial (human) intelligence (AI). They were wrong then, and remain wrong now, but we are moving beyond that into a new era of smart machines, precipitated by a ‘big-bang’ combination of three developments that came together around 2012.”

Those three developments come, respectively, from the fields of IT hardware, software and data. Specialized graphic processing units (GPUs) combined with the powerful new algorithms of deep neural networks gave researchers the tools to brute-force their way around the fact our understanding of human intelligence was lacking.

Massive collections of data, the third component, are what the first two components use to fuel their search for features, patterns and generalizations.

 

Developments That Have Made AI Possible

Developments That Have Made AI Possible

Why We Often Disagree on AI

To be clear, when AI discussions turn heated, it’s often because two different things are being discussed. Most laypeople who think they’re discussing AI are actually thinking of general purpose machine intelligence (GPMI). Also called general AI, this type is still a pipe dream.

“GPMI refers to machines that appear to have capabilities similar to humans in areas such as learning, reasoning, adapting and understanding. If there were a GPMI, it would be applicable to a very broad range of use cases. One machine with GPMI could theoretically stand in for everything a human could do. There will be no GPMI available over the next few decades,” according to “Smart Machines See Major Breakthroughs After Decades of Failure” (content available to Gartner clients).

The other type of AI is officially known as weak AI, though in practice the word “weak” is usually omitted. This is the form of AI that’s in use today, and it’s typically designed to handle a single task. Gartner’s definition is as such:

Artificial intelligence is technology that appears to emulate human performance typically by learning, coming to its own conclusions, appearing to understand complex content, engaging in natural dialogs with people, enhancing human cognitive performance (also known as cognitive computing) or replacing people on execution of nonroutine tasks.”

AI-Powered Chatbot Q&A

Now that we’re all working with the same definitions, let’s turn our focus to the capabilities and role of AI-powered chatbots in a customer service department.

Q: Do customers prefer chatbots over live service?

A: That may not be the best way to ask the question (though that doesn’t stop many businesses from asking it!) Any channel preference customers have—e.g., for phone, email or live chat—will only ever play second fiddle to the primary (and nearly universal) consumer preference for immediate and accurate answers to service questions.

This point is worth repeating, even adding a bit of exaggeration for emphasis. Customers would prefer to receive responses tied to the leg of a pigeon, if that provided the fastest, most reliable resolution.

Any specific service channel should be seen as a means to an end… so service departments can improve the customer’s service experience by focusing on the resolution they deliver, rather than the delivery method.

Q: Then why would customers choose to use a chatbot?

A: Chatbots mimic live chat in two important ways:

  1. Both are text-based applications
  2. Both allow for real-time messaging

We’ve seen these two qualities before, as they are behind the reasons consumers give for preferring live chat over voice or email. And those reasons apply to chatbots just as well as they apply to live chat.

The bottom line here is that consumers will prefer a chatbot over chatting with a live agent, if that chatbot is available immediately and the agent is not. If the chatbot is integrated with a wide selection of knowledgebase articles and can suggest the right article to the right customer, while phone agents are limited to their script, then the chatbot would be preferred.

Additionally, consumers can often be encouraged to use chatbots out of interest or curiosity, depending on the service it offers. Clever and creative chatbots can engage customers, aside from any clear customer service need. Customers, for example, probably wouldn’t call up the headquarters of Whole Foods to ask for a recipe suggestion, but they do enjoy asking its chatbot for ideas.

Q: What options do SMBs have for deploying chatbots?

A: There are currently two options, practically speaking. The first is to use the services of a third-party chatbot development company, such as [24]7, AgentBot, interactions, kasisto and Nuance.

These companies can handle everything from designing the initial concept, to programming and training the bot, to integration with your company’s existing IT.

The second is the DIY option. This a good place to start for most smaller companies—they can get their feet wet and start learning the concepts behind the training and deployment of chatbots and virtual customer assistants (VCAs).

These DIY options mostly rely on third-party technology, available from a handful of big-name tech leaders (discussed here), as well as some others offering their own API, such as api.ai, a company Google acquired in 2016. Integrating a chatbot on the front- and backends is a task that can be handled by the average company’s IT staff.

Q: Will customers know they’re using a chatbot instead of chatting with a live agent?

A: Probably. But you know what? They probably won’t care. Referring back to our first answer above, customers will always care more about the end result than the means or channel by which it’s delivered.

Again, exaggerating for emphasis, customers will prefer a response that’s stuffed inside a flaming bag and left on their doorstep if that’s the fastest and most accurate method for delivering a response or resolution.

Conversely, customers will not abide chatbots that don’t deliver results. Chatbots that are employed as a simple delay tactic (to relieve pressure from live support) will only increase frustration, before handing those freshly frustrated customers over to an unsuspecting live agent.

Chatbots From Around the Web

Now that we’ve cleared up some confusion about the term AI and answered some common questions about its use to power chatbots, let’s turn to the question that many want to ask: How intelligent do the chatbots actually seem?

Well, it wouldn’t really be fair for us to answer that. Like beauty, intelligence is in the eye of the beholder, so we’ll let you be the judge. Scroll through some examples of real-life chatbots below that would love to show you exactly how smart they are:

Sure Bot
Sure Bot
Slack Bot
Slack Bot
Whole Foods Recipe Bot
Whole Foods Recipe Bot

While true general purpose AI may still be decades away, so-called “weak AI” has already arrived, and it’s already been deployed in a range of customer service contexts.

Is better customer service the “necessity” that AI needed to be relevant? Perhaps. More likely, enterprising businesses will continue to experiment with AI applications in customer service. It’s an area worth watching for any business that handles customer service.

The best AI inventions are yet to come, and there’s tremendous opportunity for the businesses that discover and implement them first.

Realistically, and in the near term, chatbots will only play the role of a hardworking team member… they’re not going to run their own shows. This is fine and is to be expected because, as we’ve said before, customer service should be a team sport!

Improve your own team’s performance while preparing it for the addition of some AI-powered team members by finding the customer service platform that best meets your team’s playing style and strategy.

You may also like

Guide: How SMBs Can Improve Customer Service With Virtual Customer Assistants

5 Cost-Saving Benefits of a Customer Self-Service Knowledge Base

Beating Your Competition With Better Mobile Customer Support

Compare Customer Service Software