Tuesday, September 26, 2023

A Grassroots Initiative to Bridge Practice, Education, and Research.

AI + Human Is the Essential Formula for Customer Engagement

Chetan Dube,
Amelia

As AI systems for customer engagement become ever more prevalent, it is vital that business leaders understand how to design and use them. Chetan Dube explains why the AI + human formula is essential to high-quality, long-term customer engagement.

Consumers rarely grasp the extent to which businesses (as well as government agencies and a variety of organizations) are rapidly embedding artificial intelligence (AI) systems, in particular to increase customer engagement. An Accenture survey of the world’s largest companies found that 75 percent of companies had integrated AI into their strategies.1 This finding is a comfort since I’ve been arguing for decades that future success in business — not to mention the prosperity of our planet — depends on building a workforce enabled by AI to address constantly evolving macroeconomic, societal, and labor issues.

Yet in their urgency to bring AI to customer engagement, companies often miss an important consideration. They rush to take repeatable customer service tasks out of human hands, pass them on to AI (usually in the form of interactive digital agents), unveil the technology to users, and then sit back and wait for what they expect will be positive results. While those are good intentions, and such an approach can be initially successful, it’s easy to overlook how essential a digital + human formula is to high-quality, long-term customer engagement. Utilizing either humans or AI without the other will ultimately prove ineffective.

AI systems and humans must work together as a cohesive team to create personalized, ongoing user engagement that can engender repeat business, brand loyalty, and high customer satisfaction over time. Expecting AI to go it alone on the road to better customer engagement will lead companies to disappointment. Some have already discovered this and gone on to successfully pair AI and human efforts.

The AI + Human Equation

Many companies’ efforts to combine humans and machines to refine customer engagement are in flux. If a future workforce is to thrive, humans should not be primarily involved in repeatable work that AI and automation can do more precisely.2 Likewise, machines can provide data analytics and input into a process, but humans should make the decisions about personalized customer care. Both precepts are essential, but many organizations invest in automation without duly considering the critical role that people, with all of their exceptionally human skills, must play. Essentially, they set it and forget it, relying solely on AI systems to make a real impact on their business.

There is a better approach. It hinges on interdependence between humans and AI, with each working to create the best customer experiences possible, blending their skills, processes, and components. Although each organization uses AI uniquely, certain common policies will help them to build an optimal AI + human balance.

Human Competencies and Skills

For the formula to be effective, a company’s workforce must fully understand its Information Technology (IT) landscape. The IT teams, who are responsible for applying IT systems to securely complete specific tasks, such as transfers between bank accounts, must understand the business processes and workflows so they can ensure success from start to finish. Workers need to learn how AI can create and accelerate processes and outcomes for themselves and external audiences, such as customers or partners, so they can apply those capabilities appropriately.

Most importantly, humans should realize that when they deploy an AI system, they do not abdicate their own responsibility for achieving positive results. AI can independently perform many functions incredibly well, but it does not do so alone. Human workers must monitor, analyze, and study the AI’s performance and outputs, constantly identifying ways to improve it or assign it new jobs, often according to the AI system’s own recommendations.

Finally, for humans to partner successfully with AI, they must relinquish their fear that AI will completely take over their responsibilities, or “steal their jobs.” Yes, AI may take on tasks that were previously done by humans, such as resetting passwords, tracking orders, or scheduling appointments. But then, who among us would want our entire careers to consist of performing such tasks, when far more challenging and rewarding jobs are emerging every day?

Placing repetitive tasks in AI’s hands allows people the time to learn new skills and responsibilities, delivering higher value to their organizations. This newfound freedom empowers people to use their uniquely human abilities such as creativity, insight, inspiration, empathy, and so forth. These abilities are out of the reach of AI systems and are essential to developing innovative products and services. Put another way, humans should eagerly embrace the new work possibilities that AI presents.

AI Competencies and Skills

To maximize the benefit of an AI + human formula, the best route is to use AI that is designed to closely approximate human qualities, but is bolstered by machine speed and the ability to scale for high volumes on demand. We have taken this approach at my company in developing Amelia, a conversational AI tool that automates back-end systems and processes, communicating in the preferred language of users through a human-like digital agent over any channel, including voice, text, mobile apps, and platforms such as Microsoft Teams. (I’ll pause here to proudly note that Amelia is regularly recognized by third-party analysts as the industry-leading digital agent.)3

An effective AI system requires a combination of algorithms, machine learning, analytics, workflows, business rules, integrations, and some method of self-learning, in which the system makes improvements and acquires new skills over time as it completes tasks, in much the same way that a human does. In fact, we’ve taken this approach to modeling Amelia’s “brain,” designing her mind to operate very similarly to a human’s. For example, we have provided her with:

  • Natural Language Understanding through a combination of deep neural networks and natural language data sources so she can contextually understand and interpret simple and complex multi-sentence requests.
  • Semantic Memory to store facts, concepts, and the associations between them, and apply them to conversations.
  • Episodic Memory to recall previous conversations and apply that knowledge to new interactions.
  • Affective Memory and Sentiment Analysis techniques which allow her to model users’ emotions, mood, and personality. Essentially, she can learn how to compose responses that are emotionally resonant within a specific conversation, such as offering comfort to a person who was just involved in a car accident.

How these human and AI capabilities are blended depends on a given company’s ability to invest not just in technology, but in people with specific skills, so it can put an optimal organizational structure in place.

Figure 1: An Organizational Framework for the AI + Human Formula
Figure 1: An Organizational Framework for the AI + Human Formula

How these human and AI capabilities are blended depends on a given company’s ability to invest not just in technology, but in people with specific skills, so it can put an optimal organizational structure in place. Although there are variations between companies, the figure above shows an example of a high-level framework. Note how the human roles are not strictly in IT; some are business and customer service specialists who can use their expertise to influence customer experience design. Additionally, human customer service agents can work in tandem with digital ones; when a digital agent is unable to complete a customer service task, it can hand off the issue, along with information and context to avoid loss of time and data, to a human agent for personalized support. What’s more, robust digital feedback channels allow practically any human involved, employee, customer, human agent, partner, or even casual website visitor or app user, to contribute to constantly improving customer engagement.

Results from the AI + Human Approach


Results from a small but growing number of companies are beginning to demonstrate the power of the AI + human approach in customer engagement. Companies deploy digital agents, armed with conversational AI and automation to the front lines of customer service. In analyzing the experience of 40 clients that have adopted an AI + human approach with Amelia’s digital agent and automation solutions, using actual and estimated metrics and results, I found that several points stand out:

  • Containment: An AI + human approach is performing well in terms of call containment, allowing customers to resolve their issue without escalating to human agents, whose time can then be reserved for the most critical calls.
  • Intent recognition: Digital agents (powered by Amelia) have excellent rates of recognizing intent (average 90 percent), accurately determining what a user/caller wants to achieve by contacting a company by voice, text, or mobile app. This ability contributes to successful containment (average 62 percent), and to high rates of call resolution (average 75 percent).
  • Customer Engagement: An AI + human approach allows companies to bring online new means of engaging with customers that were previously impossible or unfeasible.

The forty clients I analyzed represent a diversity of leading global and regional businesses that generate millions or billions in revenue in various industries including banking, insurance, financial services, technology, and retail. Each required a system that could accurately handle thousands of calls and inquiries each month with excellent customer service.

Containment: Many companies use Interactive Voice Response (IVR) systems for menu-based phone assistance: “Press 1 to speak to a representative.” Users are “contained” within the system to complete interactions without the need for human assistance which can be time-consuming for callers, asked to hold for a human agent, and impractical for companies that may have too few representatives. As anyone who has attempted to navigate these systems can attest, many IVR systems lack the intelligence and information to handle anything more complex than basic FAQs or simple bidirectional transactions (one input in, one output out).

Many of our clients augment their IVR systems with our digital agent platform, armed with natural language and context switching to handle more than one issue at once, with an eye to raising containment rates. One of the immediate benefits of deploying a digital agent platform like Amelia is that customers’ hold times can effectively be cut down to zero by the inherent availability and scalability of the technology. In addition, when looking at actual and estimated results from our forty clients, I found that many have excellent containment rates, with an average of 62 percent, a median of 73 percent, and a range from 12 percent to 100 percent. This result does not mean that digital agents are handling all calls, only that they are playing their assigned role of keeping customers contained within automated systems. Amelia can complete entire interactions so as to reduce call volumes and reduce or eliminate IVR wait times. When a digital agent is unable to complete a task, Amelia seamlessly transfers calls to human agents, ensuring a smooth handoff without any lost data or context, so a human can pick up right where Amelia left off.

This kind of impact on containment can produce ancillary benefits in overall customer satisfaction. One telecommunications provider uses Amelia as a live digital customer service agent to handle 100 percent of its mobile call volume.4 A positive Net Promoter Score (NPS) is a critical measure of a provider’s investment value to internal stakeholders. This provider’s digital agent recorded a NPS that was, on average, 16 points higher than that of an approach that used only human agents (see figure 2 titled Mobile), a sizable change in customer sentiment. This data also indicates that human agents, once they are no longer burdened with the most routine calls, can be retrained and tasked with higher-value functions, thus raising their NPS score.

Figure 2: Comparing Digital Agent Vs. Call Center NPS
Figure 2: Comparing Digital Agent Vs. Call Center NPS
Source: Amelia

Intent recognition: A digital agent is a reflection of how it is trained, so a successful AI + human approach starts with AI’s ability to clearly understand a user’s intent and provide exactly what they want or need. This requires, among other things, an ability to understand not just languages but how people actually communicate, whether it be in full sentences, colloquialisms, or short phrases, words, and sounds.

A digital agent which cannot recognize intent is incapable of resolution, and its interactions with users can go nowhere.

A digital agent which cannot recognize intent is incapable of resolution, and its interactions with users can go nowhere. This is why the 90 percent average rate of intent recognition I discovered across our deployments to forty companies is so encouraging. And it has a corresponding impact on resolution rates, measuring the ability of digital agents to successfully complete an interaction. Our customers currently record an average 75 percent resolution rate, with a 81 percent median on a range from 50 percent to 93 percent.

Ability to engage new customers: Our clients have based their AI + human approach on specific business challenges, but there is some commonality in the new methods they used to reach current and new customers. AI solutions can be deployed across channels to match user preferences, so many businesses expand customer service beyond phone-based support to include mobile apps, text, and communications platforms such as Microsoft Teams. This multi-channel ability provides true 24/7 service, because digital agents can handle even complex transactions outside of normal business hours.

Many companies also take advantage of AI’s multilingual ability (Amelia speaks 100 languages) so as not to force users to select one language over another, as may have been previous practice because of technology limitations. A small number of companies, like Resorts World Las Vegas, create their own AI avatar or virtual presence for their digital agent, one that embodies their corporate identity and brand, and can demonstrate their commitment to engaging with users. 5 At the resort, the customer experience is dramatically enhanced. Amelia, renamed Red, handles the company’s call center solo. At any time, guests can speak directly with Red for assistance with anything including making dinner reservations, purchasing entertainment tickets, ordering room service, setting wake-up calls, and much more.

Or consider Visionworks, an Amelia client and one of the largest eye care providers in the US.6 Amelia’s responsibilities there consist mostly of laborious tasks that were previously done by in-store employees such as scheduling appointments and fittings, providing order status and pick-up information, describing safety and COVID-19 protocols to in-store visitors, confirming insurance coverage, and providing location services.

Another example of expanded customer engagement is Amelia’s role at ASICS, the global sports company, where she provides 24/7 support through email, website chat, and social media. Amelia assists ASICS’ customers with order and return status updates, dispute case status, return instructions, and shipping information.

The AI + Human Approach at Work

In a typical execution of an AI + human approach, a company harnesses all of AI’s intrinsic benefits and then makes subsequent alterations to human jobs, with a focus on employees providing new products and high-value customer service. AI, meanwhile, takes on lower-value roles or critical tasks that complement those of human partners. Digital and human agents mentor one another in coordinated customer service, working toward the same objective. The performance of digital agents spurs recommendations while humans alter user services to improve customer engagement in the long term.

One striking success story of the AI + human formula in action comes from an industry you may not expect: legal services. Our client Kenneth S. Nugent PC, Attorneys at Law, one of the US’ largest personal injury practices, uses Amelia to offer AI-enabled, personalized client service by phone, website, or secure client portal.7 The firm also uses a digital agent to address high call volumes and to free its staff and attorneys from rote tasks so they can focus on bringing positive legal outcomes to clients.

The thirty-year-old Georgiabased firm, which employs more than 200 staffers and forty-seven attorneys, fields thousands of calls and inquiries each month. Before AI the road from initial contact to signed contract was far from seamless and sometimes resulted in clients signing with competing firms. The firm’s clients also expressed an increasing interest in conduct business through digital channels, especially after office hours, though the firm had always leaned toward person-to-person interactions.

Beginning in January 2020, the firm piloted our Amelia digital agent.8 Its primary purpose was to use Amelia as a legal assistant, taking on administrative duties for the firm’s intake specialists, receptionists, and legal teams. The law firm programmed and trained its digital agent using a roster of skills, dialogues, and procedures specific to filing personal injury claims. These lessons included basic terminology, the ability to request and record salient information about an accident or injury, and natural language processing to understand user interactions in context, through sentences of varying complexity. Some of this was accomplished by uploading data on claims processing to Amelia, while other, more custom workflows were developed.

Amelia is now deployed on a 24/7 client portal where she communicates via voice or web chat. She is integrated with the firm’s practice management and customer relationship management (CRM) systems for creating and tracking cases and for contract processing. Custom-built intake tools also allow her to securely record accident and injury details and upload documents and photos.

In addition to the portal for existing clients, Amelia interacts with new clients via web chat on the firm’s homepage, and is the first point-of-contact by phone, collecting initial information before transferring callers to the appropriate case manager. In some instances, a case manager isn’t even necessary; for new clients, Amelia can take down initial information, answer questions, and independently initiate a signed client contract.

The firm’s founder reports that staff members are very positive about the technology; they do not view Amelia as subsuming their jobs, only improving them.

Case information is stored in clients’ digital files, and the digital agent (which introduces itself as a digital receptionist) is programmed to ask for updated information when a client returns to the portal. Clients continue to communicate with the digital agent after their initial contact. The firm’s founder reports that staff members are very positive about the technology; they do not view Amelia as subsuming their jobs, only improving them.

After the January 2020 launch, Amelia was fully up and running by May. On average, Amelia handles some 15,000-20,000 operator calls, 2,000-4,000 website chats, and 2,500 client portal visits each month. Amelia processes 65 percent of phone calls independently, without referring them to a human operator. Perhaps most impressive, in two years the digital agent has produced almost 3,000 signed client contracts, without any initial human intervention (human case managers continue to perform contract work). Before Amelia was deployed, clients would often contact their attorney or case manager to speak at length about case-related issues such as health records, doctor’s appointments, and so forth. Since Amelia is now responsible for those conversations, the firm’s teams can focus on finding the best way to litigate a case, and clients still feel valued.

Amelia has virtually eliminated the firm’s concerns about losing potential business due to unanswered calls. Before Amelia, about 5 percent of calls – or some 1,000 a month – were not answered. Most firms have a 10 to 20 percent unanswered call rate. Now the firm reports that no calls go unanswered, and Amelia has returned almost 200 hours a month to human operators.

Clients, it turns out, enjoy having a digital agent that is always ready to discuss their case.

Amelia’s ability to record and answer questions via web chat also provides users with an effective channel which lies within their comfort zone. Before Amelia, an average web visitor would spend about 1.5 minutes on the portal; that number has increased to 7.5 minutes for both first-time and returning visitors, which the firm attributes to Amelia’s 24/7 availability. Clients, it turns out, enjoy having a digital agent that is always ready to discuss their case.

The firm initially used the Amelia avatar for its website chat and client portal. Now, Nugent has gone one step further, working with Amelia to create a “digital twin” in his own image. Clients who interact with the firm through web chat or mobile see his likeness on their devices. Nugent has thus become the virtual personification of his own business In some ways, this approach to AI + human engagement parallels the currently limited use of AI in higher education.9 To achieve a true AI + human approach we must expand these roles considerably. Just as AI is upending how many businesses view customer-centric care, it can also drive higher education to move from current models of faculty-centered education to learner-centered ones. Interactions between teachers and students can be supplemented with 24/7 digital/ human teams. These would pair virtual agents which can converse and interact with students using machine-speed access to curricula, syllabi, research, and other information, with human mentors to provide critical context and help students learn.

Using AI for customer engagement, however, does require designers in different industries to address different considerations. A digital agent’s interactions with a banking client or an insurance policyholder would be markedly different than its interactions with a college student or a patient scheduling a medical appointment. A banking customer would expect exchanges to be prompt and business-like, mostly transactional, and with a degree of empathy appropriate to the issue at hand (e.g. a lost credit card). A person trying to schedule a surgical appointment with a specialist, on the other hand, might be anxious and nervous. That engagement should therefore begin with empathy and comfort before the transaction of setting the appointment occurs. Companies must also be sure to respect various ethical considerations such as the sharing of financial information, health records, or student information.

Preparing for the AI + Human Future

A holistic approach to customer engagement will fuel a company’s ability to survive financially and competitively. Businesses that adopt an AI + human method, one that places equal weight to both sides of the equation, will thrive; those that continue to think in more limited terms will see their revenues decrease, their margins shrink, and the confidence of their customers erode.

Beyond customer engagement, the potential benefits of an AI + human approach will be evolutionary, not revolutionary. As its influence grows over time, business leaders should set their short- and long-term expectations accordingly. For example, we expect that our base of some 200 direct clients will reap cost efficiencies of 5 to 30 percent in the first year and additional savings in subsequent years as they continue to optimize their business processes. These savings depend in part on whether these companies expand their use of AI and automation. As rote process work is handed off to digital agents, human employees can be reassigned and retrained for higher-value tasks, creating a workforce engaged in challenging and satisfying pursuits, and thus to high employee satisfaction and retention.

One of the best indicators of beneficial customer engagement with a digital agent is volume, the number of conversations the digital agent conducts in all channels. Strong performance in these sessions offers a window into a digital agent’s efficacy, whether it understands what users are asking, and adaptability, whether it can deal with a wide range of customer inquiries and learn more over time. It also demonstrates users’ overall acceptance of the agent, their satisfaction with its abilities and their willingness to engage with the digital agent again.

Looking across thirty-nine clients, I found that the average increase in sessions with Amelia during a recent six-month period was 21 percent and the median 52 percent. This metric was strongly influenced by how long the client had been using the system and thus how well established it was. Some had installed the platform right at the start of the timeframe. Others were longer-term clients reporting month by month increases in sessions. One insurance company reported that, since going live six months earlier, conversations with Amelia had increased from less than 200 per month to more than 50,000 per month. A financial services company that had deployed Amelia two years previous was still seeing substantial gains in user conversations with Amelia, from 190,000 to 260,000 per month, while maintaining a 90 percent call containment rate.

Implementing the AI + human approach is not risk-free, particularly if a company fails to prepare for how it will profoundly change corporate operations. Leaders who expect AI to be a panacea that will magically counter years of inaction in process or customer service improvement are likely to be disappointed. Similarly, those whose main purpose is to appear innovative, or to react to competitive market forces, are likely to fail in planning AI’s proper use or in due diligence more broadly. Bolting AI elements to digital channels or the back end of already inefficient processes is often a pointless exercise. Consider the recent customer service turmoil in the airline industry to see how digital is irrelevant if companies do not address underlying issues.10

Despite risks, though, it’s imperative that business leaders make the attempt, stepping out of their comfort zones and having the courage to experiment so they can make this model a reality for their companies. With millions or billions dollars of capital investment at stake, this step is not nearly as straightforward as it may sound. It requires that leaders thoroughly rethink customer engagement, working from a desire to meet customers where and when they are most comfortable.

Finally, companies must properly prepare their workforces for the changes that this model will produce. This preparation requires that leaders involve employees in the planning, designing and deployment of AI + human projects. Employees who participate in the process tend to feel a sense of ownership over the model, recognizing its potential to improve not just the business but their own jobs and livelihoods. They know that they are an essential part of the AI + human approach.

It is therefore incumbent upon companies to trust the technology and their employees in equal measure so that the AI + human formula works harmoniously, producing impactful customer engagement. Any strategy that doesn’t account for the profound contributions of both is out of tune and out of touch with where the market is going.

Author Bio

Chetan Dube

Chetan Dube is the CEO and Founder of Amelia (formerly IPsoft), launched in 1998. A mathematician, Chetan was previously an assistant professor at New York University. He is a widely recognized speaker on automation, autonomics, cognitive computing, conversational AI and the future digital workforce. He serves on the boards of numerous IT-related institutions. In 2017, Forbes listed him as one of the Nine Greatest AI Minds. He is also a member of the Forbes Technology Council.

Endnotes

  1. https://www.accenture.com/in-en/insights/artificial-intelligence/ai-maturityand-transformation
  2. https://www.mckinsey.com/capabilities/operations/our-insights/the-imperativesfor-automation-success
  3. You can learn more about Amelia here: https://amelia.ai/conversational-ai/
  4. https://amelia.ai/customer-story/ameliahandles-100-of-mobile-phone-traffic-fortelefonica/
  5. https://amelia.ai/customer-story/ameliaenhances-guest-and-employee-experiencesat-resorts-world-las-vegas/
  6. https://amelia.ai/press-release/visionworkshires-amelia-to-enhance-customer-serviceand-reduce-costs/
  7. https://www.attorneykennugent.com/
  8. https://amelia.ai/video/the-amelia-story/
  9. https://www.insidehighered.com/digitallearning/blogs/online-trending-now/artificialintelligence-assist-tutor-teach-and-assess
  10. https://www.wsj.com/articles/airlinecustomer-service-wait-times-11650999776