September 26, 2022
Conversational AI

Key insights for successfully implementing conversational AI

Experts share their thoughts on the practicalities, pitfalls and best practices for deploying chat and voice bots at this year's Conversational AI Summit

This year’s ReWork Conversational AI Summit brought together a who’s who of vendors, innovators and companies from the conversational AI space to London for two days of networking, thought leadership and panel discussions.

Among the various talks that took place across both conference days, was a sponsor of the panel “Successfully Implementing Conversational AI” focusing on the practicalities, pitfalls and best practices of building and deploying a conversational AI solution.

Hosted by VUX World’s Kane Simms, panelists included Tripta Kumari a Senior Product Designer at PlayStation, Sonia Ingram a Data Science Engineering Lead at Lloyds Banking Group, and’s own Jakob Rudbeck Mølgaard, Head of Customer Onboarding.

We’ve gathered some key takeaways from the near-hour-long discussion to share with you below:

On defining successful conversational AI use cases

Tripta: Gathering data is key to understanding where user demand is coming from - where people aren’t understanding something or where they have queries. Even if you don’t have that data, there are ways to gather it. In the past, we created an app to ask questions and discover new use cases.
Jakob: The first thing to consider is ‘What problem are we trying to solve and how does [a chatbot] create value for our organization’. The use case should be something that creates value for the company and also for the end-user.
Sonia: The simplest customer journey can be the most beneficial. Something simple like checking payments, account balance or address change - those may be low hanging fruit but can be very valuable to customers at the end of the day.

On what to do once you’ve established a use case

Sonia: When it comes to AI, you’re there for the long haul. You need a team to keep the infrastructure up to date, and make sure none of the models go stale. During the scoping phase, it’s important to work with colleagues that are on the front lines
Jakob: When you’re implementing a chat or voice bot for a customer service centre, then we would be interested in building a team from people who have experience in this field. You don’t want data scientists that can technically train the model the best, you want technology-curious people that actually understand the end users best and how to best phrase the answers and training data so that you create the best possible customer experience.

On the challenges and pitfalls of implementing conversational AI

Tripta: Voice [as a channel] is difficult. It’s constantly evolving and changing and you have to keep up to date and adapt with the market. You cannot design an application and then forget about it. As language changes you have to continually update what you’re doing.
Sonia: [Natural Language Processing] has its own set of challenges. Thinking about conversational banking, [a challenge is] bringing together the common data fabric so that you have all streams about a particular customer coming in together and accessible at the point of contact for the conversational interaction. Having that data infrastructure is a massive challenge, particularly for corporations that have gone through a lot of mergers and acquisitions, and are dealing with silos and legacy systems.

On how to measure the impact of conversational AI 

Jakob: We have three pillars of success at The first is automation, which is a mix of resolution rate and adoption. Then there’s commercial value or upsales created through the conversational AI channel. And finally there’s customer satisfaction.
Sonia: If you’re able to transcribe calls [using conversational AI] you can help draw out different metrics and create dashboards on where problems are occurring in close to real-time.

On the Build vs. Buy debate

Tripta: If you’ve got the cash [to build your own conversational AI solution], then go for it. Otherwise, start small - especially with voice as you need to gather data and make use of it. The technology is out there now to do it yourself.
Sonia: [There is an] importance in taking an objective view to third parties. A portion of our work goes into evaluating [different] third-party providers and understanding how they actually work with our data. Have they been tested on your compression rates or [for voice] the kind of terminology and accents your customers use. It is important to have a framework to evaluate these things.
Jakob: Whether you go third-party or build-it-yourself, stay involved in the project because in the end you are automating interactions with your customers. I would be cautious to outsource that completely - it’s important to have a say in how this digital channel communicates. From establishing tone of voice to phrasing, that needs to be in-house, or at least monitored closely because we are replacing a core part of dialogue with customers.

Learn more about how can strengthen your customer service is a leader in conversational AI platforms for enterprise organizations. Our technology integrates with existing technologies and platforms, making it easy to scale up your customer service technology when and where you need it most. Our solution can help your company keep up with customer demand and provide personalized, efficient, and memorable customer service experiences.. 

Learn more about how our innovative technology can improve customer service by exploring the following resources:

Want to learn more about how conversational AI can transform your customer service for your business? Connect with one of our experts today!

Related blogs

Ready to get started?

Learn how you can improve customer satisfaction, cut costs and increase revenue with