Recent analyst studies state that one in ten customer interactions will be automated by 2026, representing a dramatic increase from the estimated 1.6% of interactions today. As we pivot to more digital engagement, organizations need to prepare for a chatbot strategy that includes a plan to measure and improve performance in order to meet internal automation goals while offering your customers the same high-touch experience they have come to expect.
The 3 big enterprise chatbot
Enterprise chatbots typically fall into one of three categories
- Customer service bots – to help you take care of your existing customers
- Marketing bots – to help you acquire new customers
- Employee service bots – to help you take care of your team
When managed well, they provide your organization with a competitive edge. But if poorly managed, they quickly turn into a liability that damages customer experience, hurts your brand, your NPS or CSAT scores, and in the extreme, has a negative impact on your bottom line.
In this blog, we will go over each of these bots, how they can help your business, and how to know if they’re delivering what they should. Most importantly, we will show you which KPIs you should be tracking and what it takes to improve them continuously.
Chatbots for customer service
Customer service bots are the oldest and most deployed of the big 3 bots and they can have a dramatic effect on the performance of your contact center. The primary goal of this chatbot is to deliver high-quality service to your existing customers, at the best possible cost, and avoid escalation to a live agent in your contact center.
The data you’re going to want to focus on here are detailed containment metrics.
How to measure service bot performance
The overall containment number that most bot vendors publish doesn’t tell you the full story. What you need is a detailed analysis that takes into account good containment vs. bad types of containment. We consider this “real containment”.
In order to measure “real containment” rate you need to consider the number of times your customer
- Engaged about topics that were assigned as a bot topic, in order to identify which topics your bot will handle vs. the topics live agents will handle.
- Reached a bot endpoint, which can be identified when they are tracked as events within your bot platform.
- Requested a live agent, versus those who didn’t
You then need to remove from this analysis:
- False positives. Surprisingly, many companies have a rate of 20-40% false positives, and don’t even realize it.
- Abandonment while waiting for an agent, especially when it has been escalated by the system. Many customers will not wait more than a few seconds for a live agent.
- Bad containment. These are the topics that the bot was not designed to contain and should have gone directly to a live agent, but your customer was mistakenly routed into a bot journey.
- Customers who leave negative bot feedback. Despite your best efforts, sometimes customers are left unsatisfied even when all other criteria have been met.
By evaluating all of these criteria together, you’ll achieve a much stronger summary of your “real containment”, giving you the insight needed to take appropriate next action.
Chatbots for marketing
Marketing bots are all about generating qualified leads for your sales team, keeping them engaged and curious, ready to convert to a purchase decision. Having a marketing bot working 24/7 qualifying leads and intelligently routing them ensures your prospects are never waiting, so you can maximize your lead generation investments.
How to measure marketing bot performance
You will want to measure all of the typical bot metrics like CSAT, routing efficiency, and rate of natural language understanding (NLU), but you will also want to zero in on contact reasons. Determining why a prospect wanted to engage is key to the success of a marketing bot, and requires a distinct approach called topic modeling.
A topic model is an AI model trained using the data from millions of chat conversations between the prospect and the bot, and prospect and and live agent, to identify the true reasons for contact. So, whereas your conversation journeys may initially be designed around what you think your prospect wants to know, a topic model will tell what topics you might be missing.
A well-trained topic model will let you continuously optimize your lead generation efforts based on exactly what your prospects want to talk about every day, and ensure you’re identifying those additional prospects for your sales team.
Chatbots for employee service
Internal chatbots are a somewhat newer variation of the customer service chatbot. These started as IT helpdesk bots and have evolved to include anything that your employees may need to ask the business about. Typical use cases include:
- HR information
- Document search
- IT helpdesk
These bots are usually made available through internal messaging tools such as Slack, Microsoft Teams, and Google Chat.
The key to measuring the effectiveness of these bots is similar to the customer service bot where you’ll want to measure “real containment”, while also keeping a close eye on employee satisfaction.
How to measure employee service bot performance
To measure ESAT from an internal bot you need to go beyond the typical thumbs up/down or freeform feedback fields. A more holistic model should consider the following:
- Abandonment rate: How often does the employee exit the chat before reaching the intended endpoint.
- Paraphrase rate: How often the employee says the same thing within a given session.
- Ditto rate: How often the bot gives the same answer multiple times or more within a session.
- Conversation sentiment: These models don’t always get it right but are helpful as input within a model.
- Profanity rate: You’d be surprised at how often people curse at bots!
- Number of handoffs, post escalation: If escalation to live agents is available, then what number of transfers between agents are needed to resolve the issue.
- Feedback: Freeform text analysis provides a wealth of information that should be considered in the analysis.
A single view for all of your chatbot performance
The most important thing to consider as you roll out these bots is that you should be managing the big 3 (and more) in one place. Bringing together all your different bot types, regardless of platform or channel, under one consolidated view will give you a clearer picture of where your customers, prospects, and employees are getting value, insights into the real topics they’re engaging with your bots, and where you need to still improve (and bots can always get better).
Getting this right will mean that your team will now have the ability to quickly spot what needs improvement with a clear path to improve performance, meaning they will spend less time dealing with administrative challenges, and more time getting real work done.
How artificial intelligence is advancing chatbot performance
Chatbots are a great asset to organizations, and despite tremendous advancement in the past few years, yet they are still in their infancy compared to what’s coming. Every month that goes by, new technologies and innovations are introduced, so you can expect that your bots will quickly evolve from their current state. Artificial intelligence, for example, offers new opportunities for bot leaders. LIghtning fast text classification and topic clustering, natural language understanding, and sentiment analysis, are all examples of how AI is serving to advance bot technology.
The most important thing to remember is that the performance of your bot is only as good as the improvements you can make. Knowing what to measure is the first step to put you on the path to success, so you’re ready when the next wave of bot innovation presents itself.
Want more Wysdom?
Learn about the 6 key bot metrics every team should be measuring on any chatbot platform in our article.