Step 2 of 5: AI Self-Service Without Compromise – Virtual Agents Need “Guardrails”
This is the second of a five-part blog series that outlines the Five Best Practices for AI Self-Service Without Compromise. Use this guide to automate your contact center and Customer Experience (CX) with AI self-service in voice, chat, and text.
Best Practice #2: Virtual agents need “guardrails”
There is a fine line between what should be automated by conversational AI and what should be transferred to live agents. Many conversational AI automation projects fail due to misunderstanding where that line is and should be. It’s also why SmartAction CX designers spend so much time and effort getting “guardrails” right. You must intimately know what conversational AI is capable of and the thin line between a great CX and a subpar one. It’s a delicate balance of technology expertise and CX best practices.
Beyond understanding what conversational AI can and can’t do, setting the perfect “guardrails” requires access to customer data for insight and unique knowledge of your business to establish handling rules, whether human or virtual agent, at any point in a conversation flow. The first “guardrail” for a virtual agent typically involves the eligibility of a customer to stay within automation for self-service.
Here are examples:
- For TechStyle order management, this means the caller ID can be mapped to their customer account, or the customer is able to authenticate in one of three other ways.
- For AAA emergency roadside assistance, this means first looking up the customer account to ensure there have been three or fewer tows over the last 12 months.
- For a medical supplies company, this means callers are able to authenticate with three out of six known pieces of data about the patient (HIPAA-compliant process).
- For a global appliance manufacturer, this means the caller ID can be mapped to an account that previously called within the last three days.
- For the largest pizza chain rapid reorders, this means:
- A previous order occurred within the last 60 days;
- The price of what they ordered before is within 15% of today’s price;
- The store they ordered from still exists and is currently open;
- Items from the previous order are currently in stock;
- And the delivery will take less than 90 minutes.
For all of the above, the customer stays within automation because they have started their journey on the “happy path,” which is another way of saying a path that follows the same consistent, repetitive process until reaching completion.
“Business rules” establish when and where live agents are needed to take over any given conversation from a virtual agent. Some call types or chats can have many “business rules” associated with them depending on the number of paths a customer could take at each point in a conversation flow. If the path is one that a virtual agent can handle, then keep it in automation. If the customer heads down a path that requires critical thinking or additional customer data that’s not available to the virtual agent, then create a “business rule” to transfer these exceptions to live agents. Occasionally, creating a given “business rule” is just a business decision you make (i.e. “We don’t want to automate that yet.”).
Generally, it’s best to start conservatively by designing “guardrails” that may lead to less call deflection to start, but better overall CX with both automation and live agents. After initial rollout, best practice involves monitoring the #1 biggest reason for live agent transfer, then determining how the application can be tuned to increase containment. This includes changes like language or script edits, as well as expansion of the “guardrails” to allow virtual agents to automate more processes and scenarios. Once the #1 reason has been tackled and improved, rinse and repeat. Increasing containment also involves A/B testing proposed changes to see how customers interact with the system. It is an ever-evolving, iterative process to improve and contain more over time without sacrificing an ounce of CX.
AI-Powered Virtual Agents & What They Can Handle
For the “crawl, walk, run” approach to work, it’s important to understand: we’re not talking about simple chatbots here. Virtual agents are powered by a centralized, cloud-based AI “brain” that can:
- Connect to the same data your live agents see;
- Read and record data just like your live agents;
- Recognize natural language and extract intent over the phone, chat, or text;
- Navigate multi-turn conversations;
- And predict why someone is reaching out.
Unlike simple chatbots that use keywords to reference knowledge articles and generally struggle to truly understand customer intent, these advanced cognitive abilities allow AI-powered virtual agents to mimic live agent behavior. In the words of Gartner, “[Virtual agents] have memory and form a relationship with customers. Chatbots, on the other hand, are often narrow entities that perform a narrow set of tasks.” Thus, you can begin designing automation by examining how agents handle things today and what customer data they use to do it.
Identifying the “Guardrails” For Your Call Types
In Best Practice #1, you identified the call types and chats that are perfect for automation, and now you need to map out how agents are handling these conversations. Here’s what you are likely to find.
Many of the calls follow the same, repetitive process over and over again; humans almost feel like robots doing the job.
A percentage of the calls will mostly follow the same process but have exceptions that require critical thinking. For example, the customer doesn’t yet have an account, so the agent must create a new account for him/her.
Lastly, you’ll see that the remaining calls are each unique and different. The agent had to use judgment or complex critical thinking to get to the root of the caller’s issue and attempt to solve it.
By performing this categorization exercise, you’ve already begun to identify the “guardrails” for the virtual agent where it will provide an experience as good as or better than a live agent. Those handful of exceptions in Category 2 outline the “business rules” for the virtual agent to contain every call that follows the “happy path” and properly transfer those that don’t.
AAA clubs use virtual agents for emergency roadside assistance calls. After authentication, the virtual agent will capture intent using natural language: “How can I help you today?” The virtual agent is listening for one of several intents it can help members with like flat tire, out of gas, dead battery, and tow. If the member says something outside the intents contained within automation, like being locked out of a vehicle, then the virtual agent will simply transfer to a live agent along with all the information it captured and the reason they are calling. This enables the live agent to pick up immediately where the conversation left off without having to re-authenticate the customer, since the virtual agent already did that work and provided the account. This transfer is an example of a business rule. AAA identified seven processes that were perfect for virtual agents; all others should remain with humans for now. The benefit for AAA clubs is that their most common, repeatable processes constitute a massive percentage of overall volume, especially during peak hours, and significant cost savings. It also allows them to up-skill their live agent staff to handle the situations that only a human should handle.
For other organizations, keeping most calls with live agents is still the norm, but they use virtual agents to authenticate, capture intent with natural language, then route to the appropriate live agent. Global brands like Electrolux have reduced customer service representative average handle times (AHT) by nearly two minutes using this approach.
“Guardrails” Help Create a CX Without Compromise
The key takeaway here is that when you introduce virtual agents, you do not need to worry that they will be attempting to contain every single call or chat, especially the difficult ones. In fact, you can be sure that they won’t. By using the “business rules” that you’ve identified and developed, virtual agents will stay in the lane where they can provide a CX that rivals or exceeds a live agent.
A few additional examples of business rules that keep virtual agents in their lane:
- Durable medical equipment orders covered by Medicare have a very complex process, but virtual agents are able to handle it if the order ship date is within 30 days of the call or if the recurring order will be shipped in the next 7 days. Other parameters go to live agents.
- For a global shipping company, virtual agents can schedule package deliveries if the tracking number is either 7-digits or 13-digits. Live agents are needed for exceptions.
- A top education company uses virtual agents for invoice requests when the caller has either the invoice number, order number, or PO number. If not, invoice request calls are transferred to live agents.
At SmartAction, there is an entire team of CX professionals whose job is this one singular function. They identify the optimal “business rules” and “guardrails” during the Implementation & Design process, and make sure the data can be accessed by the virtual agent to follow. Creating the best possible human-to-machine experience requires deep understanding of what a virtual agent can do as well as or better than a live agent. With this knowledge, you can build out the “business rules” with customer data insights to keep the virtual agent in its lane to contain as much call volume as possible without sacrificing an ounce of CX. While there are many exceptions that SmartAction’s conversational AI solution can handle gracefully – such as helping retail customers skip a month of their subscription rather than cancel it altogether, or setting up a payment plan if a customer can’t make the full payment – SmartAction’s CX experts know when the exception in a conversation flow goes from repeatable to completely unique, and help you design the conversation flow and “business rules” accordingly.
Ultimately, this is why companies lean on SmartAction for CX design and management. Instead of selling software licenses then wishing customers luck in building the CX design themselves, SmartAction bundles its proprietary conversational AI with services from a team of CX experts. This enables organizations to simply outsource all of their voice and chat automation needs to specialists who live-and-breath a very iterative process of perpetual improvement.
This Best Practice, “Virtual Agents Need Guardrails,” is just one of the aspects that makes SmartAction different from any other virtual agent solution. “CX without compromise” must be the top priority when transforming your contact center with AI-powered virtual agents.
To hear from customers about SmartAction’s conversational AI technology and CX services, visit Gartner Peer Insights, where SmartAction is both the top-rated and most-reviewed Virtual Customer Assistant solution.