What's all the chatter about chatbots

Here are four predictions for business.

chatbots chatbot bot
Credit: Thinkstock

We’re hearing more and more about chatbots in the enterprise starting to gain adoption as customer service agents these days. There are two major approaches to chatbots prevalently discussed in AI circles:

  • Generative models, or “open context” frameworks, which try to learn responses and assemble them together based on their learnings;
  • Retrieval-based models, or closed “template-based” systems, which work based on templates of conversations, or confined responses. Most customer server chatbots today work this way, as do most early personal assistants bots like Siri.

As both types of bots grow in sophistication, here are four predictions:

  1. In the short term at least, template-based systems will remain the safest and most implemented approaches—despite their lack of impressive AI and the reality that they can still feel like “brute force” conversational programming. Like open-context frameworks, template-based systems can keep “state” to allow for conversations and not just one-line zingers, and the most advanced template-based systems can make recommendations based on missing conversations. Some can also confirm if not sure, or pass things off to a real-life human if they get confused or beyond their capabilities. Open-context frameworks might sound “cooler,” but their unpredictability has stumped even Microsoft, which launched a Santa bot that turned naughty in 2007 and then followed up last year with a general chatbot so trainable that users converted it into a racist in less than 24 hours. Not ideal if that’s your customer service bot: if Microsoft can’t yet launch a reliable bot with impressive AI, a bank probably shouldn’t try it.
  2. An ecosystem will quickly mature in which specialty players provide chatbots, chatbot services and related pieces ranging from implementation to specialty vendors to user interface and usability experts. The ecosystem has already started to emerge with fintech chatbot startups specific to customer service, wealth and roboadvising. As another example, Layer focuses on the user experience of customer service agents, rather than the underlying technologies, which could be a human, a bot from one of the big enterprise providers like IBM or Microsoft, a specialty bot from a startup, or all of the above. And implementers are focusing on training or templates and the knowledge management involved. No doubt, this ecosystem will continue to grow and very quickly could get commoditized.
  3. Client reception will overwhelmingly drive chatbot adoption as businesses enter the same “better vs. faster” tug of war that they fought with the onshoring/offshoring of customer service. Remember the angry reception when banks moved help desks overseas? Some improved their offshore help desks; some kept their help desks in North America and treated that as a competitive advantage; and some offered tiered service with different help desk locations for different customers. Similarly, some businesses will boast the best chatbot technology; others will compete by maintaining interactions with humans; and still others will use chatbots to feed customer service agents the right information, as already happens today, when it’s not always easy to arm agents with the right responses. Will customers prefer chatbots if they are faster and more accurate than humans at least for the “easy stuff”? Or will chatbots be implemented poorly enough that they annoy customers? Ultimately, chatbot adoption will depend on how picky we humans are in the level of service we’ll accept and who or what we’ll permit to deliver it.
  4. Combined with the correct oversight and capabilities, bots will become a highly successful compliance tool. Chatbots can easily record all interactions, and they can enforce when and where it’s appropriate to buy new financial products, for example, unlike the humans who are understandably human and pushing to make their sales quotas and keep their jobs. Chatbots can stop client reps from offering certain bank products to clients who have been identified as risky, or from offering high-risk products to clients with a low risk tolerance. They can help insurance companies treat claims differently based on risk profiles, and they can help other corporations that need to consider compliance when interacting with third parties. Chatbots make sense for compliance today, and the more they incorporate information and knowledge about relationships, the better they will help businesses enforce rules and regulations. 

Chatbots get a lot of hype, but they also have risks and limitations. To make the most of them, focus on their most helpful capabilities, provide careful oversight and pay close attention to customer response.

This article is published as part of the IDG Contributor Network. Want to Join?