From chatbot to voicebot, that sounds like a no-brainer. We asked Endre Davids, co-founder of Seamly, whether that is actually the case. And: there is no single standard for maximum allowable latency.

 

“Seamly originally started as a consultancy organization. Our advice was embraced, but our clients ran into backlogs in their IT departments. That is why we started developing solutions ourselves that ensure conversational platforms can integrate well with back-end systems,” says Endre Davids, one of the founders of Seamly.

Seamly helps organizations with the steps needed to go from chatbot to voicebot. “We call this 'voiceifying' existing conversational platforms. This is not so much about content – ​​that is already in the conversational platform our clients use for their chatbot. It is, however, about the orchestration layer: the infrastructure needed to make voicebots work in customer contact. One of our key focus areas is: how can an organization best handle the voice channel?”

We believe that new solutions must yield benefits for both the customer and the organization. Those who select only measures that benefit organizations drift away from what consumers want. And if you limit yourself to measures that customers like, the question is how scalable that is.

 

What does 'voicification' entail?

Most organizations with large-scale customer contact have a conversational platform. Think of chatbot solutions that allow customers to have a written conversation. When organizations want to expand that chatbot functionality to voicebot applications—so that customers can talk to your voicebot via the telephone channel, for example—additional provisions are needed. The most obvious are solutions for text-to-speech (TTS) and speech-to-text (STT), for speech recognition and speech generation, respectively.

We work extremely agnostic: every provider has its own strengths, and we can integrate anything with an API. But that doesn't mean you have a connection yet. How do you get the voicebot to connect to your telephony and contact center platform? How do you handle routing? Moreover, processing spoken language is truly different from processing written language. How do you deal with latency?

How do you want to deploy a voice bot?

In addition to those technical issues, more organizational choices also need to be made, Davids explains. “For larger organizations, a common factor is that they have to deal with different languages ​​and customer groups, or with telephony infrastructures that differ per country. The international telecom market is complex and fragmented.”

For the contact center manager, it is particularly relevant how the organization intends to deploy a voicebot. This determines, among other things, the extent to which the existing content of your conversational platform can be reused. When you start with an application such as voice routing – 'state in a few words what you are calling about' – you need to make different information accessible than when your voicebot needs to be able to schedule an appointment, Davids explains.

In that last application, data normalization plays an important role. A caller does not answer literally in the format a back-end system wants to see, such as '22-07-2026'. A caller is more likely to say: 'the day after tomorrow would be best'.

 

You just mentioned latency. Some look primarily at the absolute duration – longer than a second is a 'hurdle'. Others say that the acceptance of delay time varies by cultureWhat is your view on the concept of latency?

“As far as we are concerned, latency is one of the essential components that determines the customer experience. Therefore, you will also need to look at whether your voice solution is good enough from the user's perspective, and whether the latency in your solution could be a problem. In any case, it is far too simplistic to say that it is not a problem or that it will resolve itself as technology improves.”

Reducing latency requires in-depth analysis

Latency is a catch-all term, states Davids. There are all kinds of factors that can cause delays. “We analyze the latency end-to-end and then provide targeted advice. Where in your process – that is, at which conversions or other process steps – does latency occur? And what is the impact of that on the user experience? This involves not only context – what situation the customer is in, what the question is about – but also the stage of the conversation. If I want to have the installment amount adjusted at my energy company, it takes quite a long time as far as I'm concerned if I have to wait five seconds before I am recognized after entering my customer number. But waiting five seconds for a calculation for which the bot has to look into a few things is something else entirely. Especially if the voicebot clearly says beforehand: I am just going to look something up for you, one moment please.”

 

Should we start viewing latency more as a CX variable rather than a technical variable?

Yes, latency determines the customer experience. If processing takes a bit longer, you can indicate that in advance. So I think we need to look much more from a customer journey perspective. That is why you can perfectly well apply different latency standards within a single dialogue.

 

For those wanting to get started with AI, the motto recently has been: start with small steps and do not start with customer-facing AI solutions. Is that advice – start with internal solutions – still valid?

“Every organization has its own challenges. So, promoting a preferred approach is difficult at the moment anyway. However, starting small is a good route, because that allows you to gain experience and add value quickly. But not everything you do with a voicebot needs to be AI-driven. You can also organize reporting a move via a voicebot on a rule-based basis – even after first classifying the subject or intent using AI. A customer usually doesn't realize which part of the process is handled by AI and which isn't.”

 

Implementing a voicebot makes it possible to get rid of your IVR. Is that recommended?

Replacing the IVR can solve many challenges. It is quite disappointing that we have come to accept the IVR-like way of communicating in this field. There is no one who, when I ask 'do you have a tip for a nice holiday destination?', says to me: 'if you are thinking of Europe, say 1; if you are thinking of outside Europe, say 2'. Over the past twenty years, we have come to accept all kinds of technical solutions like this, primarily because they are convenient for organizations. Anyone who replaces their IVR with a voicebot can not only deliver the call to the right team but also provide it with information, so that the agent can start the conversation prepared. That works better than a customer who simply presses 1 five times in the hope of reaching the right employee. And a voicebot works faster than an IVR.

 

Why, then, is the IVR still commonplace?

“As customer contact professionals, we sometimes let ourselves be tempted a little too much by new and attractive applications and pay too little attention to optimizing what is already in-house. As a result, we generally pay too little attention to the customer experience associated with what is currently live. My advice: go and experience it for yourself; call your own customer service instead of looking at a PowerPoint slide or a process flow. Those kinds of diagrams show how you have set it up, but not what the customer experience is.”

 

Speech is generally considered the best form of interaction when we want to interact with systems: faster and better than a keyboard. Are there companies that have already elevated this insight to an art form?

“Of course, we have implementations we are very proud of, absolutely. But I must say that the boundaries of what is possible shift every week, so to speak. In addition, you have to ask yourself whether organizations can keep up with the pace at which new possibilities are emerging. Is it realistic to ask your delivery driver if they can come by two hours later today? At Posten in the Nordics, we did succeed in serving customers in four different languages ​​using a voicebot. That has had all kinds of effects: the volume of voice conversations has increased by a factor of 14, the first-time resolution rate stands at 74%, and customer satisfaction has not decreased, but increased. There are still many steps to be taken, and that is how Posten views it as well: it is an ongoing process rather than a one-off project.”

 

And what comes after 'we automate the simple questions'?

Of course, a large portion of contacts remains that is not worth automating, for example, because the volume per type is too small. But as an organization, you can also turn it around. Which contacts do we definitely not want to automate? You can quantify this as well, to prevent going on endlessly. You can also do this in a budget-driven way: allocate your budget as smartly as possible between people and automation. On the other hand, the complex questions that remain can also cost the organization a lot of money. And another question you can ask regarding automation is: do we want to subject our customer to this?

 

And finally, the future. The number of companies actively thinking about the machine customer is starting to grow. Do you have any tips for companies on how to prepare for agent-to-agent (A2A) communication?

“One of the first things that comes to mind is the volume issue. Currently, your telephony platform can handle a thousand simultaneous calls – or perhaps five or ten times as many. There are also organizations that grind to a halt if fifty calls need to be handled at the same time. With machine customers, it is possible that as many as 50.000 agents are calling simultaneously. Can your infrastructure handle that? And by extension: some of these calls may need to be escalated, meaning forwarded to a human employee. Do you have the capacity for 2.500 extra calls a day, even if they come from machine customers? Or do you need to configure the handover differently? Especially if it becomes much more than the usual fluctuation – at just five percent more, many contact centers are already starting to creak and groan.”

A consumer's personal agent is also just a customer.

At some point, you will be faced with the choice of simply treating your end customer's personal agent as an end customer. Even if that personal assistant pushes for a phone inquiry about your lost package, every single hour. I can also imagine that rules will be introduced to set limits on how consumer AI agents are allowed to burden your organization. In any case, we have a responsibility in that regard as consumers.

 

When does this become relevant? Or is A2A just science fiction?

I expect that within a few years, a few percent of the Dutch population will be using these types of applications. A small group will explore this and learn a great deal. Organizations, in turn, can learn a lot from this, so that we are collectively ready to offer this on a larger scale.

 

That is about the early adopters. Do you belong to that group yourself?

“I think so, yes.”

 

(Ziptone/Erik Bouwer)

Featured, Knowledge partners, Technology Seamly