Questions To Ask Vendors To Determine If Their Virtual Agents Are Conversational
As with any emerging category, there couldn’t be more ways to describe AI-based employee self-help: chatbots, virtual assistants, virtual employee assistants, virtual agents, virtual support agents . . . and the list goes on. This is a time when prospective buyers need clarity, but the variety of terms describing this space does just the opposite.
Now there is a new phrase being used by some vendors that could further complicate the landscape: “Advanced Conversational AI,” described as “a dynamic conversation engine designed for IT support.”
I agree completely that your virtual agent needs to have conversational AI, but what characteristics are required to consider it truly conversational?
What Makes A Virtual Agent Conversational?
The most common way for vendors to define “conversational” relative to their virtual agents is to simply state that they understand employee language. Don’t get me wrong. Building a virtual agent with natural language processing (NLP) that understands the human language is not a simple feat, and any solution that does that well should certainly be considered.
But it goes beyond that. The term “conversational” implies that a virtual agent can engage with an employee to get additional clarity on an interaction or information for a request. This is particularly important for a number of use cases, including when employee interactions are not clear.
Here are a few questions to ask vendors to determine whether their virtual agents are truly conversational.
QUESTION #1: Does your virtual agent distinguish between specific and vague questions and then respond accordingly?
Let’s assume an employee has an issue with VPN. The employee could provide a contextually relevant interaction by saying, “I want to setup 2 factor authentication for VPN.” That is so specific that any virtual agent should immediately take action or provide an appropriate response.
But what happens when an employee simply states, “VPN” or “help with Cisco AnyConnect.” For virtual agents that act like “searchbots” and rely on knowledge articles, this would lead to a frustrating experience for the employee because the virtual agent would show all knowledge articles that are related to VPN – and there are likely many of those.
Espressive Approach: At Espressive, we have experienced millions of employee interactions across our customer base. Because of that, we know that employees are not always specific with their questions. We designed Espressive Barista, our virtual support agent, to be able to recognize vague interactions and provide contextually relevant options based on the employee or organization to ensure the appropriate help is given.
We also recognize that employees will use the words that most resonate with them. They may want email, and not know the magic words that a chatbot would recognize as email. It is important that virtual agents be designed in a way that lets employees interact using their natural language. No decoder ring required!
In the above example, the virtual agent recognized the employee had a question or issue related to VPN, which for that organization means Cisco AnyConnect. The virtual agent then provides the most common topics asked by employees within that organization.
QUESTION #2: Can your virtual agent conduct problem resolution with guided tours?
If an employee asks the virtual agent, “Why is my VDI not working?” This is the type of problem where a service desk agent would generally walk the employee through a series of questions in order to pinpoint the actual issue.
So how would your virtual agent respond? Would it provide the employee with all knowledge articles related to VDI issues? Would it leave the employee to try to manually walk through the steps in the articles, and likely require them to understand a highly technical document?
If the vendor relies on knowledge articles for answers, there is a high chance that a number of potentially relevant articles would be provided to the employee since the question was vague. But this means your employees are on their own to figure out which document or steps are right for them.
Espressive Approach: One of our customers was dealing with a large number of issues related to virtual desktops. Their help desk agents would walk an employee through a complex workflow to eventually end up at a resolution. The frequency of these issues was so high that the customer wanted to automate the process.
Leveraging Barista, the customer recreated the complex workflow to engage employees and launch the needed automation to resolve the issue, which included remotely rebooting the employee’s virtual desktop.
More importantly, when employees provided a more specific statement, such as “I need my VDI to be rebooted,” it would skip the steps to that part of the workflow. In this example, the customer is using the Barista integration framework to initiate orchestration to reset the employee’s VDI.
QUESTION #3: Does Your virtual agent create service requests and ask all relevant questions related to those requests – or are employees sent to fill out forms themselves?
Anyone that has built a service request in an ITSM platform knows that these “forms” generally include a number of fields that employees need to pick from in order to submit their requests. However, experience reveals that employees rarely ever understand these forms and so they revert to email and phone to submit requests. According to Gartner, only 14% of tickets are generated by employees submitting them on their own. And when they do submit them on their own, they generally do it incorrectly. That means that help desk agents end up acting as human middleware between employees and forms.
Some virtual agents are capable of periodically pulling information from your ITSM tool and “scraping” your service requests while making them available in their virtual agents. That certainly is helpful, but that means your employees need to leave the agent conversation to fill out the form, creating a negative employee experience.
This approach has two additional downsides. First, most service requests do not have accompanying documentation, which means the virtual agent has no context on the service request. Therefore, employees must know the magic code words to invoke the service request. Further, since this will bring your users to your self-service portal, it means you must continue to fund the development of your portal.
Espressive Approach: What if employees could simply ask, “How can I enroll my new phone,” and instead of being sent to fill out a form, the virtual agent asked them the required questions in real-time? What if it could even initiate an automation workflow connected to the service request? With Barista, you can simply import service requests from your ITSM tool (e.g., ServiceNow) which means your employees submit their own requests without human intervention, and they don’t even know they did it.
QUESTION #4: Does your virtual agent help employees with only IT questions? Or through a conversation will your virtual agent understand when the question is for another department in the enterprise and help them based on that?
Virtual agents that rely on knowledge articles have a difficult time understanding the nuance of a given phrase or topic, which means they may not know where to open a ticket if one is needed. For example, Workday authentication or access issues should end up going to the IT team (or HRIS). However, if the employee is asking about their dependencies within Workday, it would clearly end up in the Benefits team.
Predicting the routing of an incident or a request is an area where virtual agents that rely on knowledge articles fall flat. Knowledge articles typically do not have organizational context embedded useful for “ticket routing rules”, which means that the service desk team needs to work with the vendor and define the rules based on specific topics (or knowledge articles). This can be a very arduous and never-ending process.
Espressive Approach: Barista understands over 1.3B phrases for employee language across the enterprise. This means the onus is on our virtual agent to ensure we are correctly handling every ticket across all enterprise departments. Take the following examples:
- “I cannot access Workday” – Barista routes this to IT
- “How do I change my dependents in Workday?” – Barista routes this to HR
- “It’s my workday and the elevator is locked” – Barista routes this to facilities
Of course, if someone simply said “Workday,” our virtual agent would need to dig in to really understand the issue before making a determination on where to potentially route a ticket. Oh, and it’s important to note that Barista understands these questions across 9 languages and growing.
QUESTION #5: Can your virtual agent collect data for leadership analysis while triggering specific actions in real-time that help employees?
Let’s say you are going through a significant technology change, like moving from Office 365 to G Suite. Can you quickly take the pulse of your employees to determine whether they are able to adopt the new technology and remain productive, or whether they are struggling and need help? If they are struggling, can your virtual agent get them the specific help they need on the spot?
With most vendors, designing an interactive conversation for the purposes of collecting and acting on information is not possible, forcing you to find another tool to engage with your employees.
Espressive Approach: We recently launched Barista Conversational Surveys, the first AI-based surveys that inform decision making while triggering actions in real time. These interactive surveys can be deployed to proactively gauge an employee’s well-being during any time of change, from crisis to a corporate-wide technology transition. For more information, you can check out the demo here.
Make Sure That Your Virtual Agent Knows How To Carry On A Conversation
Any organization that is looking at an AI-based virtual agent for employee self-help should make sure that the solution they are considering permits them to build interactive conversations. Focusing on searchbots that can only provide responses based on existing knowledge articles, that send employees to fill out their own tickets, and that can only understand IT requests will severely limit your ability to engage with your employees and maximize your call deflection.