Suppose we had a call center that nobody called? For well over a decade, companies have been trying to make that happen, first with those annoying “listen to the following menu because the options have changed” prompts, and then moving to online support and chatbots. All of these self-service technologies are aimed at reducing the staff needed to answer support calls (and of course their cost), but it’s AI that’s the recent focus. Now, our phones’ personal assistants may be about to partner with chatbots to put an end to the call center.
Almost everyone uses personal assistant technology like Siri or Alexa, and while there’s always a bit of frustration in terms of how well our assistants understand us or how appropriately they respond, personal assistants have carved out a niche among smartphone and smart home users.
We can’t quite say the same about chatbots – i.e., the automated technology that seems to be handling more and more of our interactions with company sales and support systems. Here, user frustration levels are much higher. Users like “chat” interactions because they don’t have problems understanding the agent and that they can save results for reference later. However, when “chat” means “chatbot,” users are disappointed more often than not, and many abandon support chats to look for a way to reach that elusive and expensive human.
Companies realize this, and this is why AI-based chatbots are among the fastest-growing of all AI business applications. Almost every enterprise says they’re trying them out, and almost every enterprise says that they’ll play a much greater role in their future sales/support interactions. But an increasing number of the chatbot disasters users report involve systems that claim AI capabilities. This just shows just tacking the letters “AI” onto your support system doesn’t make it palatable.
How can generative AI be so hot and chatbot AI so cold? Some experts told me that the reason is that the public generative AI tools are conveying “general answers” to “generic questions” and so their being trained on Internet data qualifies them. However, chatbot missions in sales and support require specialized knowledge. Would OpenAI or Google or Amazon have access to your product and support information, to train their AI? You’d scream “security!” or “copyright!” if they did, but without that source information there’s no chance that a sales/support chatbot could know anything useful to someone who contacts you. Glittering generalities don’t answer support questions, and users and businesses are both at risk to losing faith in AI chatbots.
Companies who offer generative AI tools, most of which are based on open-Internet-trained large language models (LLMs) like OpenAI’s ChatGPT, have been working to extend their own tools to train on company data, and to address security concerns that arise.
So, how do we get AI chatbots more user-centric, or users more tolerant of AI chatbots? This is where personal assistants come into the picture. A recent product tease from Google says a new Bard feature “will make it easier to get personalized help to spark your creativity and manage your tasks by adapting to you.” Let’s face it, a personal assistant has to be personalized to be useful, and that means that it has to understand a specific set of information, just as a support chatbot has to. Yes, there are general questions an AI assistant could be asked (“What time is it?” or “What’s tomorrow’s weather?”) but many more require user-specific data, and that’s taking assistant technology toward specialization.
Microsoft’s Copilot takes “personal” assistants in the worker direction. “Explore how Microsoft is empowering the world to achieve more with AI. See how to amplify human ingenuity, deliver transformational experiences, and safeguard businesses and data with AI tools….” Copilot is a business assistant, something that will let you “[work] smarter, boost productivity and creativity, and stay connected to the people and things around you.” By integrating AI into email composition, document development, and even spreadsheet and presentation work, Microsoft hopes to make its office tools so valuable that they change how we work.
All of this helps AI chatbots in three ways. First, the AI personal assistant is taking generative AI chat technology into the realm of user or company data, which is critical in making it relevant. Second, it’s building user confidence in AI chats overall, which makes them more likely to accept support chatbots down the line. Finally, it’s raising the possibility of integrating support chatbots with personal assistants, and that could be the biggest step of all.
Google’s and Microsoft’s AI applications are focusing on two faces of our lives, personal and business. Much of the chatbot mission, though, is a kind of bridge between these two. If our refrigerator is acting up, we don’t want to build a replacement or craft a presentation on what’s wrong, we want to get an expert to come out and fix the darn thing. The “personal” piece of our life interacts with the business piece of the collective lives of product/service employees we depend on. And guess what; companies want to build that bridge explicitly by linking the technologies so that your personal assistant is working with support resources you need.
Most personal assistants provide a set of APIs that allow third parties to integrate their technologies with the assistant, to provide information, offer control features, or both. Three different insiders in the personal assistant and chatbot spaces tell me that there are already multiple initiatives aimed at linking AI chatbots with AI personal assistants. We can already ask our assistants to control in-home technology, and these new initiatives will extend that by linking assistants directly to support chatbots. The same sort of integration is also contemplated for business services, allowing business AI and even traditional support systems to interact with automated, chatbot-based, systems of service providers, cloud providers, and equipment vendors.
It’s not hard to see how assistant-linked support would work. “Hey Google/Alexa, ask my living room TV what this message means” or “ask my TV’s manufacturer what this message means.” Home/office devices could, via API, expose their own support chatbots directly to the assistant, or identify the right support connection to the manufacturer’s APIs. You can even see how the assistant could, through the device API, gather information it could provide to the support chatbot, and they relay the response back to the user. And, of course, you can see how this could evolve to a situation where your assistant calls their assistant and resolves an issue without any human attention at either end. Nokia demonstrated a prototype of AI-assistant technology in network operations at a 6G summit recently, suggesting that even service providers could potentially adopt it and offer it to customers.
AI assistants talking to each other, leaving humans out of the loop isn’t comforting to some, but this sort of double-ended AI isn’t seen as a threat by most consumers, and even by businesses. One CIO at a big bank gave me the most evocative comment I’d ever heard; “Tom, I don’t want decision support, I just want to be told what to do.” So do we all, and the real goal of AI is to do just that. If that means accepting seemingly far-fetched risks to create a real reduction in costs and errors, most are willing to make the trade.
There’s a boatload of money being invested in developing AI technologies, and this integration of personal and support assistant technologies might be the best place to get a payback. It’s something every user, and every provider, would value. It’s also the death knell for call centers, because it’s the thing that can take all human cost out of support. Many suggest that AI would empower agents, but if AI can do that, why not support the callers directly and save the cost of agents altogether? The combination of the two assistant technologies can take them both where neither might be able to go alone, and that combination is going to happen. So will all the consequences.
We need to be thinking about what the transformation of support from human-driven to AI-assistant-driven will require in the way of technology tools, because that’s where we are going, and you can bet on it because a lot of powerful players already are.
