Succeeding with your Conversational AI/ Chatbot Implementation – Part 1

Conversational AI/ Chatbots/ Intelligent Virtual Assistants, no matter what you call them, have become common occurrences these days. It’s hard to come across a famous brand that doesn’t have one. Still, the number of chatbots that are failing to achieve business outcomes are on the rise. In an APAC study, Coleman Parkes found that almost 50% of the respondents were frustrated with chatbots and live chats. The failed implementations give the overall technology a bad rap. In our business, we have a saying that rolling out a chatbot is easy but giving it intelligence is tricky. In this blog series, we talk about the common pitfalls organisations face with chatbot implementations and how you can beat the odds.

The number 1 mistake everyone makes at the start of the initiative is assuming what the customer wants. More specifically, choosing wrong automation use cases and solving the wrong problem resulting in the project not delivering the expected business value.

Now comes the critical step, choosing the scope of the initial rollout. Even though the initial showcased customer problem needs to be solved, the chatbot needs to address more use cases to deliver actual value. When it comes to choosing the scope, either the team comes up with a bunch of use cases based on top-of-mind issues or requests top call drivers from the contact centre team or others even choose something based on what is quickly achievable from the chatbot technology selected. Meaning, trying to solve problems that the team deems important without quantifying the problem. This often results in poor engagement rates, negatively impacting customer experience and losing trust in the bot & the brand.

A better approach is to start by analysing your current customer engagement data, like chat interaction history, transcribed call recordings, customer feedback, website search data, knowledge management usage, web analytics, etc. Anything that you can get your hands on, after all, you are trying to solve your customer’s problem, and these interaction data offers that on a silver platter. Loading these data into an analysis tool like Verint Intent Manager, which uses Machine Learning to do language classification of the natural language inputs and categorises the data, will help you uncover insights like,

• What are the different enquiry types?
• How many interactions fall under the different categories?
• How widespread is an issue?
• Queries that hinder your sales/revenue generation opportunities
• Interactions that results in negative sentiment
• Interaction types in which customers value human assistance
• Time-sensitive interactions
• Current customer interaction channel patterns
• Current digital failure points
• Digital interactions resulting in a phone call
• Number of Digital journey abandonments and reasons
• Customer goals i.e. what is the customer trying to achieve when they are facing an issue?
• Your short turn interactions i.e. agents giving 2-3 line answers that solve your customer’s problem
• Agent led interactions i.e. the agent spoke more than the customer
• Journey Patterns i.e. customers facing issue with a step in a particular process also has a higher likelihood of facing issues with another step further down the process
• Conversational Patterns i.e. what do the customers do/ask next? e.g. segment of customers asking for a bill’s due date goes on to enquire about payment extensions

Here is an example of the power of analysis – for one of our clients in the insurance industry, we found that many policyholders were struggling with lost passwords. Across all departments, everyone said they needed better support for password resets. When we started looking at the patterns and the conversation flow, we quickly realised the vast majority of these customers requesting password resets went straight on to renewing their policy. The renewal was a once in a year occurrence, and they did not care about the password. If you think of it, it is cumbersome to find out you don’t know the password, reset your password, and then log back in to go through the policy renewal. So, in addition to building the password reset process, we also created a simple “easy policy renewal” process that took the customer’s basic details like their name and date of birth to renew their policy quickly. This made it simple for the customer and improved the customer experience massively. The uptake on the easy policy renewal process ended up being much higher than the password reset.

Coming back to the analysis, the outputs should also help you get a better understanding of,

• Your high-value use cases
• Help you categorise use cases: the ones that can be solved quickly, partially automatable ones and those that require system integration.
• Help you identify (and prioritise) additional web assets/ processes development required to support them
• Help you define your overall channel strategy to deflect enquiries to digital channels

When defining the initial scope, with limited time and resources, rather than choosing a handful of top x interactions to automate, it is essential to go after a good base of use cases (combination of deep & wide) that can start to deliver returns. The ones you go deep should consist of use cases from the top 10 interactions, high-value interactions and ones that result in high digital failure. The mix you choose to go wide should understand your customer’s common queries and give them some level of assistance even if it’s not a fully automated journey (will cover different approaches for this in a subsequent post). These would be for your top 25 enquiry types, the remaining ones that cause the digital failure, and all the short turn interactions.

Once you identify these, it’s time to do the maths. Quantify the scope based on current interaction volume, assumed engagement levels and an estimated deflection rate to validate if the proposed scope makes sense to move forward. If your ROI doesn’t stack up, revisit your scope to see which ones can be included without adversely impacting project costs. These data points should also be used as a baseline for ongoing tracking and measurement. to help you with this we have created a simple ROI calculator for you to download HERE.

Verint offers the above process as a standalone service called AI Blueprint™. Whether you are starting a chatbot project or looking to improve your existing chatbots, Verint’s AI tools can help you identify the right use cases and de-risk your AI investments with a data led approach. click HERE to contact us

In my next post, I will talk about the brains of your chatbot – intents and the considerations that need to go into when creating them.

– Arvindh Janarthanan
Connect with me on LinkedIn