woman disappearing out of view from a bright white corridor
PHOTO: Alessio Lin

“Man and machine always get a better answer than man alone or machine alone.” ― Ginni Rometty, CEO, IBM

“The robots are coming, the robots are coming!” said my colleague and artificial intelligence expert Kimberly Nevala in a tongue-in-cheek teaser for her new ebook, “Making Sense of AI.” She is right. In fact, in the context of digital transformation and customer experience, artificial intelligence (AI) already has a foot in the door. And that foot is poised to kick the door wide open.

IDC predicts that by 2019, 40 percent of digital transformation initiatives will be supported by some sort of cognitive computing or AI effort. Servion predicts that AI will power 95 percent of all customer interactions by 2025, and it will do it so effectively that customers will not be able to “spot the bot.” Gartner says that 85 percent of customer relationships will take place without human interaction by 2020. And Juniper Research predicts that chatbot conversations will be responsible for cost savings of over $8 billion per year by 2022, up from $20 million in 2017. 

Those are heady predictions.

On the flip side, Elon Musk, Bill Gates and the late Stephen Hawking have voiced concerns about the impacts of ungoverned AI on jobs and within society in general. Last year, Facebook cut back on the use of chatbots in its Messenger application after finding that the bots failed to fulfill 70 percent of user requests when deployed alone, rather than in tandem with a human agent. And we have all experienced the instant frustration (i.e. customer rage) that poorly designed intelligent voice response systems can provoke.

Preventing the AI Experiment From Becoming a CX Disaster

The good news? Despite the dire predictions and well-publicized missteps, AI — when implemented correctly — can be a great boon to customer experiences and digital transformations. AI can improve response time, provide contextually relevant personalized recommendations, incorporate sentiment into responses, eliminate bottlenecks and automate routine inquiries, freeing up humans to deal with more complex problems. In all cases, a blended approach — one that combines man and machine — is the key to avoiding missteps and viral mistakes.

Is there a way to effectively integrate AI into the customer experience (CX)? Here’s a look at several emerging applications that might be able help, along with recommendations for success.

Related Article: Brands Still Haven't Tapped AI's Full Promise

Chatbots and AI-Assisted Agents

Common options for AI in CX include front-end bots that handle transactions from start to finish and AI-assisted setups in which human service reps are supported by AI during transactions.

The front-end bots typically handle first-level queries by providing answers to simple questions and frequently asked questions. Automating responses to such queries allows companies to decrease training time for service reps and reduce the number of people needed to handle highly repetitive service queries. This approach generally results in a decrease in call handling times, an increase in the number of first-call resolutions and a decrease in service costs. In the highly publicized case of China Merchant Bank, front-end bots handle up to 2 million customer conversations a day, most relating to credit card balances and payments.

In setups where human agents rely on AI for assistance, artificial intelligence may be used to route inquiries, interpret incoming messages and develop initial responses that can be edited by the service rep, or find relevant knowledge-based content and deliver it to the rep. This almost always shortens both call wait time and call handling time.

Recommendation: Don’t chase the cost savings so far that you risk lowering customer satisfaction. Relying on bots to handle too many interactions, particularly those involving complex problems, can cause frustration for both customers and the human agents who must eventually deal with them.

Many companies underestimate the effort and resources that go into making a bot-based environment work effectively. Here’s a look at what’s involved: You must develop (and continually augment) knowledge management databases. You must monitor voice of the customer feedback mechanisms so that you’re aware of any problems that arise. You must review chatbot conversations to identify problems that are too complex for the bots and take note of cases where the bot misunderstands the customer.

Cutting resources before fully understanding the types of human touch that will be required can be a big mistake. Before jumping into a chatbot implementation, consider developing a set of customer journey maps to illustrate how customers actually navigate through the organization, and to differentiate the repetitive tasks that chatbots can handle from the more complex issues better resolved by real people or AI-assisted agents.

Related Article: What Data Will You Feed Your Artificial Intelligence?

Visual Engagement, Voice and Text Sentiment Analysis

Visual engagement, voice and text sentiment analysis technologies can gauge emotion and sentiment in various types of communications. Visual engagement technology parses facial expressions in face-to-face or video chat conversations. Augmented natural language processing can use voice biometrics and nuances in tone and modulation in phone calls to understand emotion and authenticate voices. And sentiment analysis can analyze written communications to determine emotion and intent. These technologies help determine how to route communications, identify satisfaction and help justify technology expenditures by allowing companies to tie positive sentiment to higher lifetime values, more profitable orders and more repeat business.

Recommendation: Ensure humans are an integral part of the implementation and ongoing review process for these technologies. While many sentiment analysis technologies come “pretrained” to identify a range of emotions, specific businesses may require modifications. Reviewing the interactions to determine where the AI did not understand the customer or missed an emotion or intent will be an ongoing activity. For example, a customer saying “I would be seriously frustrated if I lost my phone” may be natively tagged as a negative emotion rather than a positive emotion. Having service reps review and tag words, phrases and tones will help machine learning engines continue to learn. That may also lead to new metrics or more granular customer satisfaction metrics.

AI-Augmented Contextual Analytics

Analytically generated recommendations, next best actions and contextually correct product offers delivered in real time to all customer touchpoints are moving beyond simple recommendation engines to more powerful AI and machine learning applications. Spotify’s intelligent playlist, Discover Weekly, is a good example. Custom generated playlists are created every week based on a listener’s past history and likes, and on what others with similar tastes have listened to and liked. Banks are also using this type of AI in the form of offer optimization applications that sift through thousands of offers across multiple channels to find the most appropriate communication based on a customer’s product ownership, historical transactions, browsing and mobile actions and current activity. These AI-driven customer experiences build trust, loyalty and satisfaction for the customer while increasing lifetime value for the company.

Recommendation: Providing this type of AI-based analytics requires an IT infrastructure that can support a comprehensive customer decision hub, deliver recommendations in real time across multiple channels and process large amounts of customer and transaction data in both structured and unstructured formats. Companies struggling to achieve a 360-degree customer view, and those living with aging applications and/or low-quality data, must address infrastructure issues before implementing AI-augmented analytics on a large scale.