25 May
Using Human Hand Off in Power Virtual Agents for Better Success
Today, AI-enabled chatbots play a significant role in addressing both simple and complex queries of users. Statistics show that 3 in 5 millennials have used chatbots at least once in their lives. In fact, teams that use chatbots to automate conversations are 27% more likely to meet rising customer expectations than those that don’t.
Moreover, by 2022, 70% of white-collar workers will interact with conversational platforms on a daily basis.
While chatbots act as the first line of support, there will be situations when the conversation needs to be handed off to a human agent.
Microsoft customers using Power Virtual Agents (PVA) to build employee/customer-facing chatbots can explore the human agent hand-off functionality for better success.
But, what is PVA? Power Virtual Agents is a low-code app development SaaS platform that allows pro and citizen developers to build and deploy chatbots in a guided, no-code visual interface quickly.
Before delving deep into the nuances of handing over conversations to human agents using PVA, let’s understand why a human-in-the-loop is important?
Why is human-in-the-loop essential?
Chatbots help organizations automate customer and employee support – 24X7, at scale, and at significantly lower costs. They enhance efficiency and productivity and allow users to focus on tasks that add business value. Chatbots answer FAQs, send the required information, and provide smart suggestions based on the customer/employee profile.
Meaning, they augment human effort; however, in case of an escalation or upon the user’s request, the bot must be capable of seamlessly transferring the call to a live agent.
Let’s understand why a human-in-the-loop is critical.
1. User Sentiment
At times, the user is angry, frustrated, or annoyed. In such a scenario, the chatbot must be capable of detecting the mood or tonality of the user and handover the chat to an agent.
2. Critical conversations
Sometimes, the subject of the conversation may be sensitive/critical, the customer may be at a higher risk of churning, or it may be a high-value transaction. In such cases, it is prudent for the bot to hand over the chat to a human agent.
3. Highly-complex issues
Customer queries are diverse and evolving. At times, the chatbot may not be able to handle them and should recommend the “Chat with an agent” option.
4. User request
Users may be in a rush or prefer to talk to a live agent instead of a bot.
5. Training and Feedback
Training, calibrating and explaining AI-enabled systems requires human-in-the-loop architecture
Gartner
A chatbot needs a human-in-the-loop feedback system to constantly learn and become intelligent. Both users and agents can influence and improve its intelligence. In the case of consumer-facing chatbot, even the smallest customer’s feedback like “click here if you are satisfied with the solution” can help improve the machine learning algorithm of chatbots. In addition to consumer training, contact center agents can also classify outliers and exceptions and help modify the chatbot training data and behavior accordingly.
Initiating a handover to a human agent with Power Virtual Agents
Microsoft’s Power Virtual Agents (PVA) allows you to seamlessly hand over critical conversations to live agents.
Power Virtual Agents is a low-code app development SaaS platform that allows pro and citizen developers alike to build and deploy chatbots in a guided, no-code visual interface quickly.
While PVA is a part of Microsoft’s Power Platform, it leverages the Microsoft Bot Framework and Azure Bot Service and Cognitive Services to build chatbots and configure human agent handoff.
Pre-requisites:
- Engagement hub that interacts programmatically using APIs or SDK
- Product license for Omnichannel for Customer Service and a license for Power Virtual Agents
- A chatbot built with Power Virtual Agents.
Types of agent handoff triggers:
- Internal Triggers – The bot is unable to determine the user’s intent as there is no topic or matching option within a subject. Or the user requests an agent handover immediately or in the middle of a conversation. On detecting such an escalation, the bot redirects the user to the Escalate system topic.
- Explicit Triggers – When bot developers create bot content, they may determine that certain topics may require human intervention. Consequently, they may add the “Transfer to agent” node to such topics.
Learn More: Adding transfer to an agent node
Handoff integration models:
The Microsoft Bot Framework supports two models for integration with agent engagement platforms.
- Bot as an agent – This model helps onboard existing bots to the agent hub with minimal effort. In this mode, the bot connects to the agent hub and responds to user requests as if the requests came from any other Bot Framework channel. When the conversation is escalated to a human agent, the bot disengages from the interaction.
- Bot as a proxy – In this method, the user talks directly to the bot. When the bot decides that it needs the assistance of a human agent, the message router component redirects the conversation to the appropriate agent but stays in the loop to collect transcripts, filter messages, or provide additional content to the agent and the user.
Contextual variables transferred upon hand-off:
When a bot hands over a conversation to a human agent, it shares the full context as well as user-defined variables to help agents resume the conversation smoothly. Microsoft has defined the contextual variables available upon handoff, which are as follows –
| Context | Purpose | Example |
| va_Scope | Helps route escalations to a live agent. | “bot” |
| va_LastTopic | Helps route escalations to a live agent and helps ramp-up a live agent. Includes the last topic that was triggered by an utterance from the user. | “Return items” |
| va_Topics | Helps ramp-up a live agent. | [ “Greetings”, “Store Hours”, “Return Item” ] |
| va_LastPhrases | Helps route escalation to a live agent and helps ramp-up a live agent. | “Can I return my item” |
| va_Phrases | Helps ramp-up a live agent. | [“Hi”, “When does store open”, “Can I return my item” ] |
| va_ConversationId | Helps uniquely identify a bot conversation. | GUID |
| va_AgentMessage | Helps ramp-up a live agent. | “Got a gift from: HandoffTest” |
| va_BotId | Helps identify the bot that is handing off a conversation. | GUID |
| va_Language | Helps route escalation to a live agent. | “en-us” |
| All user-defined topic variables | Helps ramp-up a live agent. | @StoreLocation = “Bellevue” |
Additionally, a transcript of the conversation that took place between the customer and the bot is transferred to the live agent.
Learn More: Configure hand off to any generic engagement hub
Learn More: Configure seamless and contextual hand-off to omnichannel for customer service
How can Acuvate help?
As a Microsoft Gold Partner, Acuvate helps clients deploy AI-enabled chatbots with Power Virtual Agents for customer and employee-specific use cases, such as HR support, IT service desk, sales and marketing, etc.
Organizations can then define scenario-specific workflows for seamless transfer to human agents when needed.
To know more about our services, please feel free to schedule a personalized consultation with our Power Virtual Agents experts.
Abhishek is the AI & Automation Practice Head at Acuvate and brings with him 17+ years of strong expertise across the Microsoft stack. He has consulted with clients globally to provide solutions on technologies such as Cognitive Services, Azure, RPA, SharePoint & Office 365. He has worked with clients across multiple industry domains including Retail & FMCG, Government, BFSI, Manufacturing and Telecom.
Abhishek Shanbhag