You’re not imagining it. We’re at an AI tipping point.
Half the world is rhapsodizing
The other half is doom mongering about what humanity will lose along the way.
Take chatGPT – wherever you look there are predictions about what it might become capable of.
(And, like most tech predictions at the start of a new era, they’re mostly baloney.)
What you and I are going to do today is simple. We’re going to take a clear-eyed look at one category of AI that has already found its place in customer services – VoiceBots.
In this article, we’ll ask:
- What are VoiceBots, really?
- How do customers feel about using VoiceBots?
- How are businesses using VoiceBots?
- What benefits are businesses trying to unlock?
- What’s the secret sauce for a genuinely impactful VoiceBot deployment?
Can we agree on what “VoiceBot” means?
You can call them VoiceBots, conversational AI, intelligent virtual assistants… I’ve heard every version.
In reality, any VoiceBot is actually a lot of different pieces of tech working together. The key feature, of course, is the “voice” – if it doesn’t respond to you *vocally* it’s not a VoiceBot.
Chatbots are a dime a dozen these days… but they can only work with text.
VoiceBots however mean voice in / voice out. Think Alexa, think Siri.
So what other tech is involved in making a VoiceBot?
VoiceBots need a data set to draw from, called a “training corpus”.
For example, ChatGPT was trained using a fairly substantial part of the internet. When you ask ChatGPT a question, it’s looking at things like Wikipedia to form your answer.
(And an old version of Wikipedia too! Try asking how the career of any recently deceased celebrity is going… You’ll see how confidently it supplies flawed information!)
When a business uses a VoiceBot in customer service, the data set is usually their organizational and customer information.
Here’s an example:
Say you run an insurance company. Your insurance VoiceBot will basically know two “types” of thing:
- who has bought insurance from you
- what businesses processes you use to sell insurance and provide customer service
But that’s not the end of the story.
There’s another area of technology that is absolutely make or break – integration.
This is what separates the fun (but gimmicky) toys from the serious product offerings. Most large businesses have at least a dozen customer service systems. You probably know these call center acronyms: ACD, IVR, CRM, BI, YMCA, etc.
In practice, a VoiceBot is close to useless if it can’t look at that data and act accordingly.
(Keep that integration topic in mind – we’ll be coming back to it…)
ChatGPT can tell you how to bake a muffin. But you wouldn’t trust it with your customers.
And the *reason* you wouldn’t let ChatGPT handle customer service is that it doesn’t know anything about your business.
It’s the exact same reason you train your staff before they start representing you.
This is true of any VoiceBot or AI module on the market. Businesses have large amounts of domain-specific knowledge. You have to train the VoiceBot AI with that before you let it anywhere near a live situation.
Is installing a customer service VoiceBot a ton of work?
Yes it is – but not usually for the business buying it.
The vendor, meanwhile, still has to do a pretty substantial amount behind the scenes.
Put aside the actual process of installinga new piece of software; that’s a smaller challenge that it used to be.
The tougher bits are:
- learning enough about your business to help you build workflows
- training the VoiceBot to handle the interactions it will face
However – vendors do have a few ways to simplify this.
Installing a VoiceBot is easy(ish) because…
…the customer service workflows of many companies have a lot in common.
That’s especially true with different businesses in the same field; two separate eCommerce businesses often have more process similarities than differences.
Once a VoiceBot vendor has completed a few real-world implementations, the challenge shrinks quite a lot.
In fact, the more experience the vendor has, the more likely they are to actually *improve* processes as they go along.
(At least, that’s been my experience running babelforce. We’re not a consultancy, but we do consult… when your team has a few decades of combined experience, you’re going to spot a lot of opportunities to streamline.)
It also helps that most customer queries are about the same handful of topics.
Of the thousands of daily inquiries a medium sized contact center gets, somewhere between 50-90% of them concern the same questions.
(The global average is that 75% of all interactions are about the same 6 queries).
In practice, what does that mean? It means that any business implementing a VoiceBot only needs to focus on a relatively tiny number of interaction types in order to take a pretty huge bite out of volume.
How do customers feel about VoiceBots?
Customer opinions about VoiceBots are… varied.
But let’s dig into the question a bit.
First, ask yourself – how do you feel about riding in a taxi?
(It’s not a random question, I’m going somewhere with this.)
Your answer depends on the context. If you’re ordering an Uber in a familiar city, you probably feel positive or neutral.
If you’re on foreign shores, clambering into an unlicensed cab at midnight… well, you probably have some misgivings.
It’s the same story with VoiceBots.
Voice commands are the new normal
Do you own an Alexa? Do you talk to Siri? How about a smart TV?
Issuing voice commands to devices is normal, which is really a natural progression for technology.
I mean, think about how old you were when you learned to talk, versus how old you were when you learned to type.
The thing I keep in mind if Bruce Tempkin’s model for great customer experience:
- Success – Does the tool allow the customer to successfully complete their task?
- Effort – Is it relatively easy to do?
- Emotion – How do they feel afterwards? Impressed, annoyed, or neutral?
There is certainly a hardcore of consumers who strictly will not engage with any kind of automation or self-service. (They tend to be the older generation.)
That’s why it’s so important to provide options; the goal is not to alienate anyone.
But the proportion of engaged VoiceBot users is large, and growing.
And, of course – the ChatGPT hype is your friend when it comes to generating customer enthusiasm.
How are businesses using VoiceBots?
Let’s say you’ve got a VoiceBot up and running.
It knows your standardized contact center processes.
It’s integrated with your core systems.
It can “understand” customer questions.
It can learn.
Time to fire all those expensive humans, right?
No. Definitely not.
The key distinction I make is between full and partial automation. It’s the difference between interactions that are handled entirely by the VoiceBot and interactions that benefit from a VoiceBot but still involve an agent.
If you’ve seen our demo video already you’ll have seen a classic example.
Overall, the standard uses include…
- FAQs and self-service
- User identification and authentication
- 24/7 (out of hours) service
- Data and context gathering
- Customer research and feedback
But maybe the better question is – what are businesses and contact centers actually trying to achieve with their VoiceBots?
What benefits are businesses getting from VoiceBots?
This is basically your classic “full automation” use case. Call containment is generally one of the major metrics that all contact centers use. If you can increase, it’s a win-win.
Amtrak successfully used VoiceBot tech to increase their call containment to a staggering 54% by finding the most automable and often-repeated call drivers.
Increase availability (and shorter call queues!)
It follows that the increase in call containment increases the availability of agents to handle calls. That means shorter queue times for incoming callers, among a range of other benefits to the agent and customer experience.
NYC 311 boosted their agent availability by 25% by keeping low-urgency calls in self-service.
Better customer experience
This is always the one that causes the raised eyebrows – are customers really that keen on talking to a VoiceBot?
Anecdotally I can say that quite a few customers are either positive or neutral about VoiceBot automation for service, provided:
- It genuinely can help them, rather than just delaying their route to an agent
- It’s easy to use and makes the available options clear
- It has some obvious benefit – like being faster, or more private
Here’s another question to keep in mind: what is the VoiceBot actually replacing? Sure, sometimes it’s an agent interaction.
But a lot of the time it’s replacing a traditional IVR, and doing a much better job.
With that in mind, it’s not so surprising that Amtrak’s VoiceBot deployment lifted CSat by – drumroll – over 50%.
Saving impressive amounts of money
Depending on your outlook, this might well be the bottom line issue. Calls handled fully in automation are obviously cheaper, by a spectacular margin.
Calls handled with partial automation speed up every part of service, and unlock very real business benefits.
We’ve completed enough implementations to have solid numbers on this: businesses see between €400K and €5 million per 50 agents.
(There’s a big range because some clients have a greater focus on high value sales and retention activities.)
It’s all about integration
Let’s end on a topic I’ve mentioned several times already – integration.
VoiceBot success isn’t complicated. If a VoiceBot can genuinely solve customer problems, customers will use it.
What does a useful VoiceBot have?
- Detailed contextual information
- Customer interaction histories
- Access to customer data
- Channel switching
- Agent availability
- The ability to “warm transfer” to an agent
In other words – integration. I really can’t stress this enough. It is entirely possible (or even easy!) to link up virtually any collection of systems (with APIs).
A lot of people think they’re buying a standalone product, a box that does the voiceBot. But, if you think about it, 90% of what makes a VoiceBot or a chatBot useful in your business is how much it speaks to all your other data sets.
Any time a business tries to bring a VoiceBot into their customer service mix without meaningfully integrating the CRM, Helpdesk, SoR systems and channels, well… they might as well give up on it.