The best customer care is proactive where possible, solving the issue before the customer is even aware. And where that isn’t possible, reactive care should be timely, precise and successfully solve the customer’s problem as conveniently as possible. Self-care is emerging as the fastest and most convenient approach for a wide variety of customer issues.
If service providers are going to provide better digital experiences than their over-the-top (OTT) digital service competitors, they are going to have to improve customer care for reactive, self-care and, where possible, proactive care. Recent changes in AI-driven conversational interfaces suggest new directions for automating customer care on all of these fronts.
Evolving customer care
Service providers are embracing the self-service trend. Michael Maoz, Research Vice President and Distinguished Analyst for Gartner Research, estimated in a 2018 presentation that by 2022, 48% to 64% of service providers will have adopted some kind of self-care. Leading the change will be developments in machine learning and “conversational agents,” which include customer assistants and chatbots. Also important is the adoption of an omnichannel strategy that not only provides the full complement of communication channels, but ensures that support sessions that have begun on one channel can shift to another channel (e.g. mobile to fixed telephone, or self-care to agent-care) with full continuity.
The biggest developments in this area are around interactive, algorithm-driven software programs that use machine learning and AI to interact with customers through conversationally based human interfaces. Interactive bots provide an ideal interface for customers experiencing common issues that have simple solutions, such as connecting to Wi-Fi, resetting forgotten passwords, or checking on the status of a scheduled technician appointment. These bots are available on demand, eliminating frustrating wait time for customers and allowing intuitive communication using spoken or written words. Known as autonomous customer care, industry analyst firm Analysys Mason, predicts it to be the fastest-growing sub-segment of customer care, with sales forecast to reach 1.486 billion USD by 2020.
Unfortunately, the rapid rollout of virtual assistants, chatbots and other software-driven care modules has so far only addressed very simple scenarios. Their efficacy has been proven, but the current approach isn’t scalable. As the numbers of conversational touchpoints proliferate, coordination between them will prove increasingly difficult to manage and the end-to-end customer experience will inevitably suffer.
To realize the full promise of these technologies, we need a better approach that
- Leverages best practices of care activities from call centers, field technicians and other areas
- Creates re-use across the various conversational touchpoints
- Provides automated responses that are appropriate and more empathetic human conversations
- Continually learns to respond better – both in human conversations and in providing recommendations.
If the task is straightforward enough, automated assistants can be scripted relatively simply. Unfortunately, most service issues don’t have simple solutions. For instance, “my internet is slow” could have multiple causes; thus, troubleshooting is complex. This is often reflected when we capture real-life workflows, where we find that various agents take quite different paths to solving the customer problem. As a result, it often requires a lot of experts with very detailed technical knowledge to improve just one step in a workflow.
A promising approach is to use machine learning inside the human workflow to develop a predictive model for narrowing down what the best probable step or action will be. Here it is critical to leverage the best practices of call centers, field technicians and other areas by capturing historical session information, session context and submitting these to analysis to identify the fastest remediation activities.
To make best use of this kind of predictive model, you also have to create a certain amount of flexibility at key steps in the workflow, so that the predictive model can steer the workflow dynamically. This means under-specifying the execution order of workflow steps where there may be variability. Using recent historical data and what is working best in the given context, the probabilistic model can then steer the workflow appropriately.
The historical and contextual datasets, as well as the probabilistic model, can be re-used across multiple virtual assistants and every other customer care channel. Data from the performance of every customer care touchpoint can then be fed back into the model in real-time.
This dynamic, machine-learning process can also be interfaced with a natural language engine, such as Google’s natural language engine, Amazon’s Alexa, Apple’s Siri or Facebook Messenger. The key here is to supplement the natural language engine with strong analytics to help identify the customer’s intent using sentiment and topic analysis. Over time all interactions should be stored, including date, time, subscriber intent and outcome. This provides a foundation for machine learning, the progress of which can be analyzed at any time by the service provider to evolve their understanding of intent and appropriate remedial actions.
Ultimately, the goal of such a system is for the machine-learning engine or ‘bot’ to take proactive actions to solve the customer’s issue — before they’re even aware they have one or have sought help. But, there will always be some percentage of support calls that will occur. This dynamic machine-learning process can also help to improve self-help interfaces and even guide field and call center agents, augmenting their decision-making with rapid processing of customer intents and suggested follow-up questions or even remediations.
In the highly competitive digital services space, customer care is a key differentiator. Using a machine-learning, AI-driven, omnichannel approach holds great promise for lowering customer support costs. It will also improve the support experience and enable service providers to compete with the digital experience of OTT providers. Ideally, it will proactively eliminate a large percentage of the simplest issues being reported to call centers today. The key is to move beyond a proliferation of single-purpose bots, and create a single probabilistic engine and dynamic workflow model that can both inform and learn from the full spectrum of a service provider’s customer care interactions.