Service with an Artificial Smile

40 percent of consumers prefer using automated services rather than talking to a human on the phone

AI will lead to enormous cost savings in sales and customer service, because they will be able to answer customers‘ questions faster and more accurately than before. Services that make customers‘ lives easier will generate more customers, providing more training data to make AI systems even smarter. Customer service might finally live up to its name thanks to AI.

„Your call is important to us,“ says a tape voice to resigned customers who are endlessly on hold, waiting to talk to a human agent. This is where AI can help companies improve the quality and consistency of their service and convince customers that their concerns are actually being addressed.

Ocado, a UK online grocery retailer, receives around 10,000 emails from customers every day and uses AI to gauge their mood. This way, the most urgent complaints are first identified and forwarded to agents with the right expertise in the relevant field.

„As with other AI applications, it’s about making people more efficient, not taking them out of the process altogether,“ says Paul Clarke, Ocado’s Chief Technology Officer. At least 40 percent of these interactions will involve an element of AI, according to Gartner market researchers.

AI could change customer service as much as the telephone once did. Before that, companies processed customer inquiries by mail or personal visits. The telephone helped agents become more productive. Thanks to AI, productivity will increase even more dramatically because it can process a large number of customer inquiries faster than humans. This has become even more important as communication channels have multiplied: Customers have more options than ever before – email, mobile messaging applications and social media – and it is always the customer who decides which way to contact the company. And consumers have become accustomed to using automated services. Surveys show that around 40 percent of American Internet users prefer to use digital customer services rather than talking to someone on the phone: self-service, it seems, is perceived as the best form of service.

Virtual agents are now on the rise. Around 30 percent of companies now offer their own „bots“ that can answer questions and solve problems, even if their range is still less than that of a human being. Many of them use AI; they are trained on protocols and transcripts of past customer interactions, and the more data they receive, the better they can solve more complex queries. Such bots allow companies to handle many more requests without having to hire additional staff.

China Merchants Bank, a commercial bank, uses a bot for the popular Chinese app WeChat to process 1.5 to 2 million requests a day, which corresponds to a workload of about 7,000 employees. Caesars, the hotel and casino group, offers a virtual concierge, Ivy, in two of its hotels, who answers guests‘ queries by text, many of them automatically if the query is easy to answer. This has reduced the number of calls to the human concierge desk by 30%.

By their voices you shall know them

AI will also improve the knowledge, performance and speed of account managers. Some companies are experimenting with „voice printing“ – a technology that recognizes customer voices and alerts agents when a caller attempts to impersonate someone else.

This will be particularly useful in financial services. An Australian bank is experimenting with a stand-alone intelligent voice-controlled loudspeaker to listen to their agents‘ credit calls. If the agent forgets something or makes a mistake, colleague Bot steps in. Other companies are using AI to suggest responses to customer inquiries, which a human agent can approve or adjust before sending. Last year, KLM, the Dutch-flagged airline, doubled the number of text-based customer inquiries it handles to 120,000 per week with only a 6 percent increase in agents, says Dmitry Aksenov of Digital Genius, a company that helps automate customer support.

A few companies have already started offering AI-enabled services that listen in on calls to assess agent performance and send them suggestions for improvement in real time.

Cogito: AI with emotion

Cogito, a Boston-based startup founded by a former MIT employee, has developed software that uses natural language processing to perform emotion and mood analysis in call center conversations. It measures energy levels, pace, speech style and other factors in real time to detect and interpret the intentions of the speakers so that they can identify errors and make spontaneous corrections. For example, if an agent speaks too fast, Cogito’s software could suggest that he or she slows down or addresses the client with a question. With the help of AI, Cogito determines an „Empathy Score“ for each call center employee, depending on how good the agent is at conveying empathy to the customer and successfully resolving complaints.

Cogito uses artificial intelligence to enhance human intelligence, especially emotional intelligence. That’s right – we use technology to help people be more human. Almost ironic, isn’t it?

Cogito is acquiring channel-separated audio signals from your phone system. This allows us to isolate both customer and agent speech and call exchanges. During a call, the technician deconstructs the conversation in millisecond steps and analyzes over 200 different vocal, non-verbal signals, such as porosity, tone, tension, speech rate, vocal effort, rotation and mimicry. These signals are then analyzed and correlated with the insights gained from tens of millions of conversations using our proprietary artificial intelligence.

During a live call, Cogito’s low-latency compute engine analyzes hundreds of unique behavioral signals to provide objective behavioral instructions in just a few milliseconds.

Cogito customers include insurance companies such as Humana and MetLife. There, Cogito focuses on the detection of „pity fatigue“. The focus is on details such as the speed at which agents speak and the words used by callers. This enables the AI to recognize emotions and assess whether the interaction is proceeding satisfactorily. If there is a problem, the agents are encouraged to act more sensitively.

Even if the agents know they are being monitored, Cogito says that most of them welcome such measures because they give them valuable feedback. For the company, it’s definitely worthwhile: According to Cogito boss Joshua Feast, call center sales can be increased by between 30 and 40 percent per year.

Marty Lippert, Chief Technology Officer at MetLife, expects the AI to deliver a return on investment (ROI) of around 20 percent in areas such as customer service and human resources. Most companies buy AI services from external providers, but companies with their own technical expertise often prefer to create their own AI. For example, a team from Uber has set up a system for processing AI requests by e-mail, which is offered alongside the traditional telephone number and sends the agent a ranking of options for further action, which can reduce the time needed to process a complaint by around 10 percent.

One hope is that AI will relieve service staff of routine tasks so that they can sell other services to customers and generate new revenue. KLM has generated millions of dollars in additional revenue since it began using AI because agents now have more time to help customers book upgrades and new flights, says Dmitry Aksenov of Digital Genius. However, as he admits, there is a risk that customers will feel disturbed and repulsed by overly intrusive sales attempts.

Other AI systems, such as Unilever’s „Smart Swap,“ can help consumers make healthier nutritional decisions based on their past shopping history. And by using AI, railway companies can improve the punctuality of their trains and avoid train cancellations.

Clever companies will use AI not only to improve existing services, but also to develop new ones. The Metro Group in Germany is testing the use of computerized checkouts: The items in a basket are recorded by cameras and the buyer is billed accordingly. Timo Salzsieder, Chief Information Officer of Metro Group, expects that these new unmanned, vision-assisted checkouts will be able to handle 50 customers per hour, more than twice as many as a manned checkout. Amazon uses a similar technology in a grocery store in Seattle.

Some insurers, including Ping An from China, are using artificial intelligence to record customer claims after a car accident. Instead of calling the insurance company and filling out a lot of forms, customers photograph the dent in their car and submit it via an app to get a quick estimate for repairs.

Building such a tool is a technological challenge, but it’s a good idea to get in early. Services that make customers‘ lives easier will generate more customers, providing more training data to make AI systems smarter. Ping An receives 15 million claims per year and is already processing 30 percent of them via app. „This takes enormous costs out of the system and puts customers in control,“ says Jonathan Larsen, Chief Innovation Officer at Ping An. Such offerings also strengthen the direct relationship between companies and their customers.

Conversely, voice-activated „intelligent speakers“ such as those offered by Amazon, Google, Microsoft and Apple could also come between companies and their goals. Some are just hosts for the apps of other companies. For example, UPS has developed a tool that allows customers to track their packages through Amazon’s Alexa, which they have previously done online or over the phone. The companies fear that they could be disintermediated so that the company that builds the speaker becomes the customer’s primary relationship, says Paul Daugherty of Accenture, a consulting firm and co-author of the new book „Man + Machine“: Reimagining Work in the Age of Artificial Intelligence. Because voice-activated speakers guide customers to a single vendor rather than offering them a variety of companies to interact with, those who cannot or do not want to use these speakers may not even realize they are failing to build a relationship with the customer.

Much will depend on how fast Alexa & Co. spread. Currently, only about one in six adult Americans has such a voice offering, but that’s already more than twice as many as a year ago. And as speech recognition continues to improve, the appeal of the speakers will grow, especially among young people.

Knowing what the customer will want

The robot-supported process automation (RPA) as a gateway technology for intelligent process automation, will also play a huge role in marketing and sales in the future, for example in carrying out credit checks, updating customer data, assigning clerks and personalizing content. Every brand interaction that a consumer has, be it via an e-commerce website, social media or in-store, will in future be tracked and stored by a marketing platform. So-called „predictive“ solutions will then dig deep into these insights to determine the logical next steps and predict actions that customers with specific consumer profiles will take. The result of these analyses can automatically trigger corporate actions through various channels, such as email, mobile devices and the Web.

There are now powerful forecasting tools such as Salesforce.com’s Einstein or Atomic Reach that can help you understand what content will appeal to a target audience.  Concured, which sees itself as a „listening tool“, uses artificial intelligence to analyze consumer behavior in relation to content and to shape future content marketing. Such tools are crucial in creating and executing personalized marketing strategies. In order to create truly personalized consumer shopping experiences, marketers need up-to-date insights into each audience member, and must then be able to generate and deliver unique content accordingly. When consumers receive this customized content in near real-time, the efficiency of marketing teams‘ work is greatly increased.

Showing the customer what they want

The Chinese Alibaba Group, the country’s largest trading company, operates e-commerce portals such as AliExpress and Taobao which have long been using powerful AI systems to help customers find the right product. Each time a customer comes to the website, a selection of items is generated based on their known preference patterns. This is done by watching closely what the customer does, and the selection is adjusted in real time to increase the likelihood that a sale will be closed. On the Taobao portal, the company uses reinforcement learning to train the AI. However, because the risk would be too great if the machine were to make decisions completely without supervision, they have created a kind of „virtual Taobao“ where customer behavior is simulated using hundreds of thousands of hours of recorded data. Due to the large amount of data, the system can be trained with a variety of different customer behavior patterns in a much shorter time. The findings are then fed into the „real“ trading platform.

Alibaba also has its own chatbot called Dian Xiaomi, which means „sales assistant“, which helps to channel and process customer enquiries. Customers who call the call center without visiting the website can be redirected to the chatbot one by one. Only if the chatbot does not know what to do, is the customer redirected to a flesh-and-blood consultant via a call-back service. The chatbot also serves as a marketing instrument, for example to answer simple questions before the purchase.

According to Alibaba’s product manager Liu Jianrong, the AI sales consultant is primarily used to relieve the company’s customer service centres at peak times. „We have noticed that companies are short of staff during the day, especially at busy times, but also in the evening when the call centers are less busy. Even special days such as 11 November – in China the Singles Day, which is celebrated with bargains and special offers in the shops – are easier to manage this way.

In marketing, various types of forward-looking solutions are possible. Predictive E-Mail solutions can provide consumers with targeted recommendations via e-mail by analyzing their previous shopping history, e-mail usage rates and the historical affinities of similar consumers. Other types of recommendation emails include cancelled shopping cart campaigns, post-purchase campaigns and cancelled browse campaigns. When a consumer visits an online website, predictive analytics views can be used to create a personalized shopping or browsing experience by placing certain recommended products over the fold or generating a sidebar of „frequently purchased“ items.

Looking to the future of marketing

In the meantime, the buzzword „predictive marketing“ is beginning to gain acceptance among experts. Anyone who knows today the customers of tomorrow will of course have the chance to use his advertising budget more efficiently. Knowing when a customer is ready to make a purchase will be decisive for marketing success in the future. But this requires a sufficiently large database of customer and sales data. A decisive factor here is the intelligent combination of data from different data sources to identify sales opportunities and to use the findings from data analysis to identify target groups that are ready to close.

Predictive marketing is primarily about creating so-called personas that have statistical characteristics that are relevant to the closing. The data-driven identification of the closing potential based on historical data enables advertisers to achieve a much more precise target group description and identification. Data must be collected and bundled using various channels, taking data protection into account. This enables conclusions to be drawn about potential target groups in the individual channels.

The prerequisite for successful predictive marketing is a well-thought-out strategy. This strategy must not only include the usual demographic characteristics, but also things like time of day and user behavior. This is the only way to create an overall picture of the customer and to get answers to key questions such as which channels should be used to address the statistical personas, how marketing processes can be adapted to personas and which algorithms are necessary for prediction.

As always in AI, the key is a sufficiently large and well-maintained database. But where to get it from, if in the past customer data has not been systematically collected and stored in a retrievable way? In such a case, AI-based personalization platforms offer a simple and cost-effective way to enable individual customer experiences across all contact points such as website, app, email, POS system, IoT device and call center. Providers such as Nosto or Dynamic Yield use machine learning and AI to personalize and optimize the so-called „customer journey“, i.e. the journey of each individual customer to the product in real time. Based on self-learning algorithms, brands can directly test what works well and what doesn’t. Customers then receive personalized offers in real time or purchase recommendations by e-mail. „Personalization will soon be omnipresent,“ says Liad Agmon, founder and CEO of Dynamic Yield, „because it is an absolute must in marketing!