An increasing number of businesses are experimenting with artificial intelligence, especially generative AI, to streamline customer self-service and improve customer experiences. However, with estimated moderate deflection rates for contact centers ranging from 20 percent to 50 percent, live agents will remain a fact of life in CX for the foreseeable future. These agents are often driven by repetitive, manual, and largely inscrutable processes, a.k.a. "zombie" processes.
This is particularly true for businesses where customer support involves complex, multi-step processes that span many businesses, partners, and legacy systems. The result is disproportionate deflection costs, efficiency hits, high agent burnout, and risk. Worse, it puts agent-driven support out of step with the rest of the rapidly evolving digitized CX.
What's the problem?
The path to meaningful agent automation is challenging. By definition, every problem presented to an agent is a one-off. In solving problems, agents encounter thousands of ticket types, every type of system, nuanced processes, embedded team knowledge, and dispersed teams, all while customers expect a snappy, quick response. Talk about stress.
But there's some light at the end of the tunnel.
By using AI to identify the underlying agent workflows, it's possible to analyze and re-engineer the agent experience, ultimately creating personalized workflows for each agent at scale.
Agents' lives involve two types of tasks. First, there's one-off research, looking at different systems to gather data to document history and/or provide context; this step is usually manual or moderately automated. Second, there are specific motions to resolve the problem functionally (for example, copying and pasting an address and a bank account number; creating a return label that requires the name and address of a shipper; and so on).
But large enterprises might have 10,000 agents: 1,000 teams of 10 people doing similar things, each team having a slightly different process and nuanced technical architecture. Traditional discovery processes don't work here because it's impossible to scale to so many processes.
While these events by definition can't be deflected, they involve a lot of unnecessary navigation, points, and clicks that could be avoided or shortened. If you can identify problems across processes, you can automate them.
A Three-Step Process
By using AI, we can do away with a lot of steps currently under the supervision of agents. How? Similar to how we're using AI to create self-driving cars.
Step 1: Learn, learn, learn.
Following our car analogy, start by learning the roads. AI can learn what your agents are doing, collecting data to learn more about their response indications. In our car analogy, indications might include looking at the distance maintained from the car in front or the ability to stay within the lanes. Eventually, you can identify similar critical steps in service that have challenged automation.
Step 2: Create automation.
Use your collected data points to start to control the steering wheel under supervision. Agents can course-correct and take over if there's a mistake, while the system improves as it learns.
Step 3: Deliver business transformation.
As time goes on, AI will be smart enough to guess the next event and provide the automation. It will look at every agent process across the team and the company, unearthing nascent process inefficiencies and productivity gains.
Businesses in retail, e-commerce, and services are starting to take advantage of AI for this. A large e-commerce company has used AI to reduce a repetitive multi-click process in its ticketing system. The original process starts in the ticketing system, grabs customer information, finds that customer and data in another system, and traverses multiple systems before coming back to the original system, with agents repetitively copying and pasting the same information along the way. Because AI has learned the workflow, the company can now provide real-time automation for each agent (and experience fewer errors).
Many agent processes can similarly be transformed. A small business (an animal shelter software provider) is using AI to enter its client-intake data into two systems: its CRM system and a proprietary system it needs to maintain. AI can compress this multi-step process into one step.
Workflow transformation promises significant cost savings, particularly in larger companies with many systems and complex, well-ingrained processes. Consider an internet provider. A simple refund order can involve a lot of data, different customer devices, and multiple stitched-together systems. With proper AI automation, such processes can scale easily, mine data, and capture the information that flows from it to enable faster discovery and analysis. Agents spend less time doing the remedial, mind-numbing tasks that frustrate them.
Why now?
Companies have invested heavily in making their customer experiences responsive, efficient, smooth, cordial, and conducive to upselling. Self-service and agent service are both necessary, but the combination is expensive. Agents are always going to be dealing with changing, complex processes, which is why agent churn rates remain unacceptably high for many companies.
Yes, the path to meaningful agent automation is challenging, with agents having to adjust to both new systems and new and legacy processes that are highly manual or inefficient. But few companies appear to be investing in modern automation to help them be more efficient. It's time to break down the complexity that defines agents' lives. AI can make them better, more efficient, and happier at their jobs.
As AI increasingly sets the pace of business, non-automated agent support will bite companies. Businesses need to get to the bones of their customer support processes, like live agent support. AI can smooth the transition.
Boaz Hecht is co-founder and CEO of 8Flow.ai and former vice president of platform at ServiceNow.