The acceleration of digital transformation was one of the most striking impacts of the pandemic. Video conferencing, contactless payments and conversational AI were just three of the technologies that saw massive adoption rates during 2020 as organisations scrambled to enable employees to work remotely, while helping customers to safely transact in-store and self-serve online.

Contact centre employees in the healthcare, travel, financial, retail and insurance sectors found themselves inundated with calls, while they adjusted to working from home themselves. Many organisations quickly implemented chatbot solutions to address this quandary. According to Gartner, penetration rates of conversational AI increased by 20%-50% in 2020, up from a range of 5%-20% in 2019 and chatbots are projected to see 100% increase in the next two to five years. So how can organisations build on their early successes with bots and use conversational AI as part of their broader digital transformation?

Building on success

From our own customer base, we saw WorldRemit implementing conversational AI to guide new customers on making their first digital money transfers to loved ones overseas as traditional high street outlets were forced to close. The bot conducts around 140,000 interactions a month. Of these, 60% don’t need to speak to an agent, which has reduced service costs, improved customer satisfaction and reduced pressure on customer service staff.

US mortgage provider, Cenlar, created a bot to assist borrowers who might require mortgage forbearance as a result of COVID-19. Offered within Cenlar’s IVR, 12% of borrowers who indicated that they were calling about financial hardship opted to receive an SMS link to the bot, instead of talking to an agent. Spurred by this success, Cenlar then launched a second chatbot to help automate common borrower requests while offering a warm handover to a live chat agent when necessary. In doing so, they achieved a 75% containment rate, alleviating the pressure on agents who could then focus on solving more complex issues.

When calls to its Dublin contact centre increased by 30% overnight, EVO Bank of Ireland Payment Acceptance (BOIPA) set up an option on its IVR to allow merchants to receive an SMS link to a bot that provided COVID-related information. Within twelve weeks, 17% of calls were being deflected to the bot. Once the call deflection solution was launched, BOIPA deployed a bot on its website to handle payment terminal setup and troubleshooting and bank account changes. The bot achieved a 70% engagement rate.

In a typical month, British mobile games developer, Intouch Games, handles 60,000 live chat queries and this number increases when new games are added. The company created a bot to handle routine live chat enquiries. Within 90 days, Intouch Games measured a 28% reduction in the volume of live chat enquiries handled by the contact centre team.

Now we are seeing companies that successfully implemented chatbots to provide a digital self-service option during the early months of the pandemic seeking to build upon that success. There are three clear trends emerging as a result:

Going beyond Q&As and using stories to create better experiences: Organisations want to go beyond using bots for simple Q&As and are looking to use conversational AI to create more engaging, personalised experiences. As an example, banking systems are integrating conversational AI into their digital strategies as a means of solving problems for customers. It’s no longer enough to use a bot to simply inform them of opening hours. Customers are now interacting with digital assistants that can help to transfer funds, cancel payments and resolve complex multi-step issues that involve several workflows. Consider a request like “I need to pay my credit card bill because my card just got declined.” Natural Language Processing (NLP) technology is getting better at understanding user intent, however, these multi-part requests need to be built into the bot conversation so that the right workflow is triggered as the context switches. As a result, we are also seeing stories being woven into bot conversations to make interactions more personable and fluid.

Moving intents into stories and customer journeys is becoming much more important in conversation design. In recognition of this trend, Google launched Dialogflow CX to handle more complex multi-turn conversations, helping to create more human-like conversational experiences. Another example, Rasa, the open-source NLP platform, is being used to go beyond Q&A and build stories to make interactions flow more like real conversations.

There is a growing demand, regardless of sector and use case, for digital assistants to handle more human-like conversations. The pandemic just drove this to new levels. We’re also seeing avatars and sentiment monitoring becoming important in the effort to deliver better experiences. As an example, where a customer expresses a really negative sentiment, the query can be routed straight to a human, rather than a bot resolving it.

Expanding use cases: As a result of lockdown, we’re seeing conversational AI being used in areas outside the typical customer service use cases where it can impact business outcomes, for example, in sales conversion, proactive campaigns, document gathering, onboarding, renewals, complaint handling, and more. The common goal is automation, but mostly with a path to a human to handle the trickier or more sensitive customer issues. 

We work with many brands in banking, insurance and finance, helping them to introduce their first bots with a path to extend the technology to other parts of the business. Once they’ve seen initial success from bot implementations, organisations want to either expand the number of use cases and/or add more functionality to their existing solution. For large multinational organisations, that often means developing bots that support multiple languages and provide a consistent experience in different territories.

Conversational AI has an immense range of potential applications. Organisations are looking at how they can apply AI-based bots to support the entire customer lifecycle end-to-end, from customer acquisition, through to operations and retention. Some of our customers use bots for digital onboarding. In the insurance and banking sectors lockdown created difficulties with gathering proof documents, for example in onboarding a new policyholder or complying with Know Your Customer (KYC) regulations.  Identification documents or proof of no claims bonuses for insurance policies were normally sent into a Shared Services centre. When these centres had to close to comply with stay-at-home orders, organisations needed another way to onboard customers. We created a way to send a link to customers’ phones, so that a bot pops up on their phones that welcomes them and requests the relevant documents, such as asking new customers to send a picture of their driver’s license. We’re also seeing bots being used to provide help with insurance quotes, as part of the sales process.

Whatever the use case, the key goal is to automate those interactions that make sense to be handled by a digital assistant rather than a human agent, while still offering a path to escalate to a human. The collaboration of digital and human skills not only reduces the cost to serve but, when done well, it increases both customer and agent satisfaction. 

Scaling up: In addition to broadening use cases, organisations want to scale up their bot deployments to serve larger numbers of users. We’re seeing the emergence of large-scale conversational AI projects and a resulting increase in complexity.

However, adding features and capabilities to a single conversational AI bot can reduce its accuracy and performance. The leading natural language processing (NLP) engines, including Google Dialogflow, Amazon Lex, Microsoft Luis and IBM Watson have different limits when it comes to the number of ‘intents’ that they can handle in a single bot. As organisations take the next steps on their journey to using conversational AI to provide personalised conversations and support end-to-end customer journeys, the volume of intents grows and the intent limits of the underlying NLP can become a barrier.

Avoiding pitfalls

Naturally, if your contact centre has managed to deflect 70% of live chats to conversational AI and you’ve enhanced customer service and grown your business as a result, you’re going to be interested in replicating that success in other parts of your business. However, as described, a larger rollout of the original bot to broader segments and different countries, where more features, languages and capabilities are added, can, over time, result in the bot performance degrading as it becomes overloaded with intents.

Another pitfall is where organisations grow by acquisition and try to merge different bots, built on different NLP engines, into a single bot.

Benefits of a multi-bot orchestration architecture

A multi-bot architecture helps address the issues outlined above. Rather than building a single monolithic bot, we advocate taking a more modular approach where multiple skilled bots are managed and orchestrated by a central Virtual Assistant at the forefront of the conversation. This Virtual Assistant can route conversations, according to intent, to and from different bots, where each has a specific skill. This is similar to the traditional contact centre IVR which routes requests to the appropriate service, or a receptionist in a building who greets visitors and directs them to the correct department.

Using a multi-bot orchestration approach offers many benefits to organisations as they mature their conversational AI strategy:

• It is easier to build, train, and scale each bot service
• Allows new functionality to be added up to 70% faster
• Specific bot content is managed by line of business managers and subject matter experts
• When new functionality is added, it is less likely to impact bot performance
• Easier and up to 40% more cost-effective to maintain the technology
• Common functionality, such as ID verification and authentication, is handled by the virtual assistant. Therefore, there is lower reliance on IT teams when new bots are developed to support new business services
• Customers have a consistent user experience no matter what bot is serving them

Conclusion

As we move beyond the pandemic crisis mode it is time to build on the digital successes of the past year and consider how to use those same technologies across other use cases and areas of the business, driving higher degrees of automation and business success.  For organisations that are expanding their conversational AI deployments, and for new adopters of the technology, a multi-bot architecture enables them to more quickly and easily scale up their AI bot projects without losing the ability to remain agile and responsive to change.

Interested in hearing industry leaders discuss subjects like this? Attend the co-located 5G Expo, IoT Tech Expo, Blockchain Expo, AI & Big Data Expo, and Cyber Security & Cloud Expo World Series with upcoming events in Silicon Valley, London, and Amsterdam.