Conversation is essential to any retailer’s branding strategy. But in the ongoing shift from brick and mortar to online stores, the channels used for connecting with customers have changed. Promotional mailers have been replaced by email, helpful store clerks by live chat prompts, and customer support phone numbers by omni-channel online communication strategies. The shift to virtual customer engagement presents an exciting opportunity for sales optimization: every customer interaction generates data that can be analyzed to create valuable insights. And as the adage goes, what gets measured gets improved.
Conversational Commerce is growing rapidly as a means of communication between brands and customers. According to the US Census Bureau, most retail growth comes from online sales. In 2016, retail grew 2 percent while e-commerce grew 16 percent.1 To capitalize on the consumer preference to shop online — especially among Millennials — brands need to optimize their live chat strategies. An e-marketer.com survey shows that 63 percent of customers were more likely to return to an online store that offers live chat.2
The data generated by conversational commerce helps companies measure good and bad outcomes, generate insights, and implement strategies that impact revenue, profits, and losses. Measuring the outcomes of a live chat agent’s conversation drives resourcing decisions. If the behavior of an agent impacts conversions, companies should invest in coaching sales representatives. If conversational behavior doesn’t have much impact on sales outcomes, companies could save money by replacing human agents with bots. Furthermore, the outcomes of conversational commerce are highly multi-faceted and nuanced; some companies strive to enhance the customer experience, others attempt to optimize sales conversions or attract repeat customers. Rather than understanding an interaction solely based on whether or not a sale occurred, or simply whether the customer was satisfied with the interaction, agents should understand the trade-offs posed by different KPIs and know which methods of rapport building lead to positive outcomes, providing a roadmap for future interaction.
To deduce whether chat agent behavior affects outcomes, we analyzed the data of an anonymous e-commerce company with more than 200,000 customer visits and $25 million in annual revenue. The company had used agent-based live chat on their website to answer questions and also provide assistance and sales support; a feature that generated over 2.8 million messages.
We then created a model of an agent-customer conversation that took into account factors that influence such dialogue. These included the obvious ones, such as time, day, location and personality type, but also the complexities of human conversation like word choice, grammar, turn-taking, style of speech, and emotion. We identified the most influential variables, and put them in two buckets; one for variables driven by agent behavior, which we considered to be “controllables,” and the other for those variables outside the scope of the chat agent engaged in a conversation, otherwise known and “non-controllables.” We came up with eight significant factors as follows:
- Agent Behavior
- Outside Behaviors
- Agent demographics
- Message topic
- Visitor demographics
- Visitor statistics
Effort consists of the agent’s use of proper grammar, capitalization, and spelling. Emotion encompasses all of the text strategies used to convey the extra-conversational cues that live chat tends to leave out, like exclamation marks, emoticons, and positive emotion words. Friendliness was measured by the agent’s use of sympathy words that establish human connection and communion, such as “I” and “we.” Finally, the chat agent responsiveness category relied on the timing of the text response and the frequency of consecutive messages.
Using an approach called Variance Decomposition, we assigned values to each controllable based on how much it contributed to the overall system of converting leads into customers. If we return to our real-life conversation, Variance Decomposition will allow us to build a model that considers how much the time of day matters in relation to how much the location or other isolated controllables such as effort and responsiveness matters in the model.
We studied four different key performance indicators: customer satisfaction, sales conversions, total order size, and repeat customer purchases. In all three scenarios, the controllables significantly outweighed the non-controllables when we looked at how much explainable variance could be attributed to each. Effort, Emotion, Friendliness, and Responsiveness had a combined influence of 57.7 percent on Chat Scores, 60 percent on Sales Conversions, 58.2 percent on Total Order Size, and 74.9 percent on Customer Retention.