Bots Aren’t Born to be Good (or Bad)
Like any relationship, it takes a bit of work to ensure everyone is happy. Building better bots requires more than training a language model and then adding it to a channel. Bots need to be continually evaluated to ensure they meet KPIs; models need to be refreshed as usage increases.
Training a bot to communicate is like teaching a child to talk. As bots communicate with a wider audience (your customers), they encounter more — and different — ways customers ask for help. And information itself is changing. New policies, products pricing strategies and locations are just some of the changes the bot needs to know about.
Bots can only respond with what they know. Without training, the bot won’t know how to respond to these new requests.
Making sure your bots can respond with consistent accuracy leads to wider adoption and use. When bots are optimised, they can answer 100% of the initial interaction as a baseline.
However, delivering an exceptional end-to-end experience requires more than an accurate answer. Bots need to be able to identify when the conversation has shifted beyond their ability to help and do something about it.
Don’t Let Your Bot Get Lonely
Bots that are developed and operate in isolation are doomed to be unloved by customers — and they’ll be actively avoided. It’s inevitable that your customers will test the limits of a bot’s ability. Still, no one wants to hear “I’m sorry. I can’t help you with that. Visit our support site.” This is the equivalent of the phrase “It’s not you, it’s me.” And no one wants to hear that.
Making sure bots can transition a customer to an agent is one way to avoid dead ends. That’s obvious. But that’s where some solutions stop. What’s not as obvious (but offers more value) is to carry that entire conversation throughout the full interaction — with context and history, from the bot to the human agent.
When the customer is handed off to a live agent, they don’t want to repeat themselves or provide the same information again. Forcing customers to repeat themselves creates frustration and makes customers feel that your company isn’t listening to them.
Ensuring your bot is tightly woven into the customer journey requires a high level of orchestration. In the ideal scenario, the bot can listen to the journey as it’s happening, personalise the conversation based on journey context, and then seamlessly transition the conversation to an agent who can pick up and resolve the issue. Too often, however, companies assume a bot is prepared for everything. But there are always surprises. A customer might ask something unexpected or something the bot can’t handle.
Train, Measure, Analyse, Optimise, Repeat
Lovable bots get a lot of love in return. These bots are built through training; they’re measured and analysed; they’re optimised based on performance. And they’re re-trained to ensure they’re continually learning and getting better at understanding customers — and meeting their evolving needs.
It’s hard to show bots the love they need without the right tools and best practices. Here are some things that can help.
Business-Centric Build Tools: Having actual business users create bots — whether directly or side-by-side with conversational AI developers — ensures that those bots have a clear business purpose.
Integrated Interaction Workflow: Instead of creating a bot and then trying to wire it into the interaction flow, it’s easier to create it as part of the interaction flow during the development phase.
Out-of-the-Box Reports: In some cases, bot frameworks only allow you to export data or data logs. Mining and massaging the data takes considerable time. Often, by the time the data is ready for analysis, it’s too late to make any meaningful decisions. It’s more effective to have out-of-the-box, real-time actionable views into performance as well as model accuracy.
Actionable Analytics: Actionable analytics go beyond simple reports; these analytics allow you to use what has happened previously as a guide for what should happen next. This gives you data-driven guidance on where you can make improvements to the bot.
Generative Data: It might be tempting to rely on generative data and avoid the ongoing maintenance that bots require. A good initial data set might get you close to what you need but having a new channel can change customer behaviour. Your generative data might not be specific enough to approximate the behaviour of your unique customer base.
Is Generative Data the Love Potion Your Bot Needs?
Generative AI, the source for generative data, has become a hot topic. Some see it as a magic potion that will replace the need for sourcing training data, which has always been a difficult task for those trying to train bots.
Essentially, generative data is content that’s automatically generated. It can provide a good starting point, but you need to consider the source. Does it represent your customers and your products? Genesys uses actual conversations as the training set for bots, which means your bots are lovable by design.
What Your Bots Need Now
At the end of the day, your bot needs to help real customers with their real issues. We all want bots that can provide help, answer questions and deliver empathy.
We want highly effective bots that are ready to show customers some love. And that starts by handling your bots with love from the very start.
