What is Intent for Chatbots, and What Does it Mean for Your Business?
Intent is one of the most important touchpoints between customers and bots. We sat down with Paige Twillmann of LivePerson to talk intent strategies and best practices.
April 10, 2019
The following is a transcript of a conversation we had with Paige Twillmann, Global Solution Lead of AI & Bots at LivePerson.
Before we get started, let’s define intent. There are actually two definitions: A customer’s desire to change something to do with your brand from state A to state B, and an ideal unit of management for automations and contact center operations.
Let’s hop into some examples. For instance, person A is coming onto your site, and asking your chatbot some questions, like, “My bill looks different from last month,” or “I'm trying to log into my account and I'm locked out.” They could be asking a number of different things, but ultimately as that contact and as that conversation is being achieved, throughout the course of the back and forth, the ultimate goal is to pay their bill. Right?
Keying in on that intent is where we find value. When you're thinking about that in the context of your business, the intent is the thing that the customer is coming to the brand to talk to them about. When you're thinking about designing a chatbot, there's often a lot of onus put on making the chatbot very conversational and able to have a nice back and forth. Really, customers are trying to get their questions answered and move on.
Creating proper structures
The important question for bot developers is, what is the best way to start going about creating these intent structures that will properly handle customer queries? When you're thinking about the chatbot development process end-to-end, as well as your brand, it’s actually best to work directly with the teams that are having a conversation with your end customers. At LivePerson, we have call center agents that actually weigh in on our chatbot development and give us insight on reaching the end result of any given conversation. They're the ones dealing with those customer queries day in and day out, so getting them actively involved has proven to be super important. Then, making sure that your decisions when prioritizing intent are very data driven is an important aspect as well.
We'll see brands leverage their call transcripts, their messaging transcripts and their email transcripts to really determine what the best intents to automate are. It's also best to start with ones that are not overly complex. If you're a bank, and we used the billing example above, new customer cases may not be the right ones to get started with, because there's a lot of documentation that is involved in that process. You're authenticating users, you're going into their banking details, you're moving money; that's probably one to steer away from at the onset of a project, just because it is so complex. We generally start with ones that are either already automated, or at least very process driven. Things like draft changes or order status for retailers are big ones that brands see a lot of value in. We seem to get many questions like that, and they're pretty mundane if you're thinking about things that can be automated.
Employing human agents
Human representatives will most likely always have a place in customer service. So, when dealing with intent, how do you know when is the best time to have your bot defer a query to a human agent?
Maybe there's a shadow of a doubt of what the customer may be asking, or the bot doesn’t have the conversation script to handle the question. In these cases, they hand them right off to the agent. It’s easy to fall into traps where bots will repeatedly react, react, react, and react again to questions, even if they don’t know the exact answer, and that tends to get customers very frustrated. The bot can ask one qualifying question, but only if they think they may be onto what the intent actually is and direct the user in the right way. After that, it’s definitely time to fall back to the agent.
There are a lot of brands in the market that have chatbots that are live, with no seamless handover to a human if escalation occurs due to the bot not knowing an answer. In many cases, the bot has been programmed to just provide a phone number. As you can imagine, that's a terribly frustrating experience, because customers are having a great back and forth conversation with the bots and they want to stay in that digital environment, and with the presentation of a phone number, customers are thinking, “Wait what, now I have to get my phone, and call a human agent who doesn’t know the history of the conversation I just had with the bot.” Our expertise in these situations is to definitely have your bots replying hand-in-hand with your human agent, so that a seamless handover can take place.
Weathering the storm
With this trial and error period comes an initial storm of issues that can be difficult to overcome. At LivePerson, we've seen brands have a lot of success in developing their bots within really fragile working methods. This means employing your bot in a test environment, with the test environment being maybe 10% of the volume of usual conversations on your site, or displaying the bot for only an hour a day to see your full audience base.
After doing these kinds of tests, examine the details of where the bot failed and improve on it over the course of the first two weeks of the launch period. Allowing that soft launch can help eliminate a lot of early issues. Usually when developments scrums are running, they're either updating the bot every other day or updating it over the course of the first couple of weeks. If you're just going to unleash it to your full audience, and not look at it again for weeks, you're not going to see success in that type of live scenario.
With intent, the name of the game is to try, try, and try again. There will be bumps in the road, but you can use those bumps to inform future strategy and build a better bot. Developing proper conversational tactics to understand customer intent is one of the most important aspects of chatbot implementation, so making sure this is an issue that you focus on will be integral to your success.