Automation is becoming less of a novelty and more of everyday business for many brands. According to a McKinsey Global Report, 45 percent of all work duties can be potentially replaced with AI and machine learning. Over 65 percent of companies will experiment with customer-facing chatbots (AKA bots) this year, with the aim to provide 24/7 customer support and reduce cost.
So how can firms make the most of their bot implementation for CX success? As people rush to use the technology, they often overlook important human and process elements resulting in poor execution and a frustrating customer experience. Here are six common mistakes, and how to avoid them.
Issue 1: Not built for true conversation
AI’s natural language understanding (NLU) has come a long way. By 2020, Gartner predicts that half of customer queries may involve natural language processing along with search and voice methods.
The challenge lies not only in programming bots to understand what a customer is asking, but what the right answer is to their question. Today’s bots are highly scripted, built to follow a simple flow based on predefined customer “intents” or tasks. But customers don’t follow scripts when engaging with companies; they tend to ask multiple questions at once or divert to another topic mid-conversation, which can confuse a bot. For example, if someone asks about health insurance open enrollment and adding a dependent to their account in separate inquiries, the bot may be able to answer just fine. But put those queries together in the same question and the bot can get stumped, serving up unhelpful information as it tries to decipher multiple requests within one query.
Solution: Task-oriented bots can become easily overwhelmed by the unstructured ways people talk. Technologies leveraging recurrent neural networks (RNN) allow bots to be trained on conversations, not tasks, and handle multiple intents within a conversation. When training bots on conversations, leverage historical associate and customer conversations to understand not only the way customers ask questions, but also the types of inquiries they may group together.
In addition, connecting the bot to an integrated knowledgebase allows it to access information and formulate responses for situations it hasn’t yet experienced. For example, if a bot was being used to take restaurant orders, it would not be trained on every possible order combination, but providing access to the full menu and custom order options could train it to place valid orders based on unscripted customer interactions.
Issue 2: Set up to fail
According to a recent CMO Council study, 47 percent of consumers say if the brands they love provide a frustrating experience, they will stop doing business with them, while 45 percent say they also will intentionally seek out the competition to spend their money where they believe it’s more valued. Many bots are deployed to take in all customer inquiries, when in reality they are only programmed to handle very simple ones. They operate in a silo of their own information, without the ability to connect to other company systems. This causes frustration and friction, leading customers to “bot out” and escalate to human associates. It eliminates the cost-reduction benefit companies were looking for in the first place and increases the likelihood of customers souring on the brand.
Solution: Identify the bot’s role in the customer journey and only funnel interaction types to the bot that it can successfully handle, ensuring appropriate integrations have been made to support them. For those interactions a bot cannot handle, seamlessly escalate to a human at the onset. For instance, if a bot is going to receive order tracking inquiries, it must connect to a customer’s order history and the company’s fulfillment system to accurately respond.
It’s about setting bots up for success, not providing a kitchen sink’s worth of tasks for them to handle. Bot misplacement can set it up for failure from the start.
Issue 3: Inconsistent experiences across channels
Bots are often deployed in specific channels (i.e., Facebook Messenger, web chat), and are trained based on interactions in that channel alone. This leads to siloed insight and inconsistent responses to the same question across channels, including what a customer may encounter when interacting with a human customer service representative.
If bots are trained inconsistently in an increasingly omnichannel world, as a customer switches between channels or escalates to a human and receives different information or responses, it can damage trust with the brand and erode the relationship.
Solution: Deploy and train bots for an omnichannel experience, using a single training platform that is consistent with how human associates are trained.
By employing machine learning that is shared across all channels, bots will continuously improve experience consistency while also leveraging lessons learned from live associates.
Issue 4: Schizophrenic bots
Just like human employees, bots are an extension of a brand. But many times, bots are created by different conversational designers and trained by different data scientists or developers—be it across business units or even around different interaction types within a single business unit. As a result, the style and tone may vary from interaction to interaction, making the bot seem schizophrenic in tone and style and widely inconsistent with the brand—veering from casual colloquialisms to Queen’s English, depending on who trained it for a given interaction.
Solution: Giving bots a defined brand persona and developing clear guidelines around that persona, tone, and style can help ensure consistency no matter what task it is performing. Just like advertising agencies follow brand standards and guidelines when developing marketing collateral, conversational developers should follow bot persona guidelines to ensure bots are aligned with the desired tone, style, and personality of the brand at all times.
Issue 5: Can’t seamlessly escalate
In theory, bots should be able to provide fast and efficient service for customers who want simple answers. According to a 2017 Ubisend report, 69 percent of customers are willing to use a chatbot before a human associate it if it means they can get instant answers. But what if the customer has had enough of the bot and wants to talk to a human?
Many bots cannot directly escalate to a human; they provide a 1-800 number as part of an escalation script. And even if they do escalate directly to an associate, few bots pass on context of the conversation, forcing customers to start over with the associate. This can be an incredibly frustrating experience that wastes time and energy. And it ultimately defeats the whole purpose of always-on bots as smooth and frictionless assistants.
Solution: Ensure seamless integration by integrating the bot with the same chat tool your associates use, as well as your CRM desktop, passing on the context of the conversation so that when a chat or call is transferred, the associate can pick up where the bot left off. CRM integration is essential for not only allowing associates to review bot chat transcripts, but also for allowing the full case to be appended to the customer’s record. In that same vein, enable the bot to disposition inquiries it handles to resolution to ensure complete customer case histories, and allow the ability to report on bot case handling consistently with that of human associates.
Issue 6: Bots are in over their electronic heads
Bots can do very well with simple, repeatable requests, like tier 0 and even tier 1 simple interactions whose answers can be scripted. But some questions are simply beyond their understanding.
If a bot performs poorly, customers will not give it another chance, and the result will be angry customers calling human associates.
Solution: There must be proper due diligence in assessing which inquiries lend themselves to automation. Start small and add over time. Don’t handle all inquiries completely with bots. Implementing a visual IVR with the option to initiate a conversation with the bot or immediately transfer to a human associate saves the hassle of starting a problem the bot cannot finish. Pick the right conversations to automate to avoid customers questioning why they were connected to a bot at all.