How IVAs Will Transform the Customer Experience as We Know It
Intelligent Virtual Assistants (IVAs) are changing the customer experience game. But what are they, and what does their future hold?
Top Learnings From This Article:
Chatbots tend to operate with a set of fixed, procedural responses, suiting only simple, unambiguous business situations
- NLP allows IVAs to read and understand the customer’s intent -- even when worded differently from the phrasing they trained on
- The more flexible, intention-led approach taken by IVAs can improve the success of the initial triage process with customers, and also allow a more seamless blending of human and machine
- In the future, IVAs will be able to read and deduce multiple user intents, and figure out a specific user intent from having only trained on a similar example.
Two years ago, 73 percent of customers in BT’s global autonomous customer research thought that chatbots would improve customer experience – now that people have experienced them in real life, that figure has dropped to 58 percent.
- Dr. Nicola J. Millard
And why’s that? Because the chatbots most of us know (and hate) are old tech.
The first article in our IVA series looked at the differences between old-generation chatbots and Intelligent Virtual Assistants, or IVAs. We considered the many reasons why IVAs could improve the customer experience, and looked at how traditional, scripted chatbots were falling short.
In this next article in our series, we’re going to look beyond the what and consider just how IVAs improve the customer experience.
Let’s allow Dr. Nicola J. Millard, Principal Innovation Partner for BT, to catch us up:
The issue is that chatbots tend to operate with a set of fixed, procedural responses. This means that they tend to only work well in simple, unambiguous situations – they are effectively ‘interactive text response for digital’.
The worst bots dump customers into a dead end when they can’t understand what they want and force customers to start again on another channel (usually switching from chat to the phone). They don’t understand subtler things such as customer intentions, and they frequently don’t seamlessly integrate with the human contact centre agent.”
IVAs, however, are a game changer in this regard, since they’re programmed with NLP and deep learning. Now, it’s easy enough for us to talk about IVAs and throw around these fancy AI terms, but how do these technologies actually work to improve the customer experience? We chatted with our CSO Jaakko Pasanen, who explained the intricacies of the science behind IVAs in a comprehensive way.
Having NLP in your AI chatbot enables a better customer experience. This much, we know from the first article in this series, Intelligent Virtual Assistants: Far from your average chatbot. But just how does this work? Jaakko explains:
Visitors communicate their issue to the IVA. The IVA now needs to 1) understand what the visitor is asking, and 2) what kinds of answers your brand gives for this type of question.
This is where NLP technology kicks in.
Upon reading the customer question, if the IVA detects a broader topic, it can be trained to ask a few more questions to the visitor about their particular issue.
This is called intent recognition, and according to Jaakko, this technology is already quite advanced.
An old-generation chatbot would need the customer to phrase their intent exactly as it’s worded in the chatbot’s training database, to be able to identify their desired intent.
NLP, however, allows the AI to read and understand the message. This means that even if the customer phrases their intent differently from what’s in the training database, the bot can understand the intent he or she is after.
Let’s consider the example we looked at in the first article.
Shep’s question was: “How does your docking station charge the computer when it’s plugged in?”
The old-generation chatbot, however, replied: “Which computer do you want to buy?” it responded.
This went on for several minutes until Shep left the site to purchase his desired docking station from one of the brand’s competitors.
Had Shep been communicating with an IVA, however, powered with NLP, the results would have been very different. The bot would have been able to understand that the object of the sentence is not the computer, but instead, the docking station, despite both being mentioned in the question. And so on.
The more flexible, intention-led approach taken by IVAs can improve the success of the initial triage process with customers, and also allow a more seamless blending of human and machine. This, too, is appealing to customers, with 81 percent of people in our research thinking that humans needed to be in the loop when complex issues were being dealt with by AI.
- Dr. Nicola Millard
As Dr. Millard points out, IVAs tend to complement human work (that’s why they’re called assistants, after all!). Because of its higher level of intelligence, an IVA can identify when human assistance is required, and, taking advantage of deep backend integrations, can transfer the customer to a human agent seamlessly -- also passing along the full context of the conversation.
How’s that for an improved customer experience?
What’s the future of IVAs?
And this is only the beginning. Just as IVAs stand at the forefront of your customer service, we’re standing at the forefront of IVAs. They’ve come a long way from their old-generation chatbot cousins, and their future looks incredibly bright in terms of improving the customer experience.
We talked earlier about how IVAs are pretty advanced when it comes to intents. Here is how Dr. Millard sees this technology developing:
The challenge for the technology into the future is to be able to detect and act upon multiple intents from customers, e.g. “my broadband is working again, but I’m not sure what this charge is on my most recent bill”. This requires the IVA to work out that these are two different topics (repair and billing), potentially across multiple business silos. A human agent would be able to work this out quickly, but would probably still have to cut through some organisational complexity to solve the business silo issue (which is mostly a CRM problem).
Jaakko, for his part, is looking forward to advances in AI that will make the onboarding process much easier.
At the moment, it takes quite a bit of training to get IVAs online. This is because when it comes to intents, each customer intent usually has just one answer, and it can be phrased in several different ways. Currently, this dataset needs to be more or less customised for each brand. There’s no off-the-shelf, one-size-fits-all solution at the moment.
The next level is to be able to give the AI only one example of an intent -- even a very complex one -- and from that, the AI can extrapolate all the different ways a customer can phrase that intent with far less data. The less data that’s required, the faster the IVA can go live. And that’s going to be transformative. I believe it’s about three to five years until we’re there.
To sum up, IVAs are already starting to turn heads as the new must-have CX tool. Their NLP programming makes them far more flexible and friendlier than their old-generation counterparts, and can work seamlessly with human agents, to give your customers the best possible experience. Down the line, we can look forward to IVAs being better able to detect multiple intents from a complex customer query, and train and go live faster than ever before.
But don’t take our word for it. Let’s give Dr. Millard the final word:
The holy grail of IVAs in customer service is to improve customer experience whilst freeing up human agents to add value to more complex issues. If the right balance can be achieved then IVAs can be of huge benefit to organisations and customers alike.
Dr. Nicola Millard is a presenter, writer & researcher on future trends & customer/employee experience, and Principal Innovation Partner for BT Enterprise.
Jaakko Pasanen is Chief Science Officer & Co-Founder of ultimate.ai.
Stay in the loop
Subscribe to our email newsletter. No spam, just occasional insight from our experts.