Ten Tips for Deploying Enterprise Virtual Agents
We use artificial intelligence (AI) every day without knowing it. Alexa’s speech recognition, Netflix’s movie recommendations and Gmail’s type-ahead suggestions are all examples of data you share being fed into deep learning AI models to improve your experience.
At work, the same technology is increasingly used to deliver smart services — rooms that book themselves, thermostats that don’t cool empty floors and outages that are restored before anyone knows they exist. Achieving Amazon-quality AI with “small data” from your organization is a more complex technical problem.
To get enterprise AI right at scale requires thinking differently about how applications and services are deployed and managed. In the decade ahead, most enterprise software will be “intelligent,” and most interactions will be guided by virtual agents using natural language.
Designing virtual agents for the enterprise is part art and part science. Ones that succeed will do the following:
1. Continuously Learn: Smart software, by definition, incorporates new data and user feedback into AI models to continuously improve accuracy. Intelligent virtual agents are trained when users ask to escalate an issue or reply with “not helpful” in response to automated recommendations.
2. Hand Off To Live Agents: No user should be penalized because a virtual agent didn’t help. In fact, for virtual agents to be at least as effective as live agents, they must make it easy for users to ask for live help. Make sure live help is easy to access and virtual agents are aware of live agent schedules and the languages they speak.
3. Recommend A Next Best Action: When virtual agents can’t help, make sure your virtual agent enriches tickets with context to help live agents. The virtual agent should hand off details, including who is requesting what, where they are, whether what they need is related to other issues and whatever other data can be used to recommend a “next best action.”
4. Explain Why They Made A Decision: Don’t pretend the bot is human. Always indicate when AI assistance was used to automate the support process. Equally important, make sure your virtual agent provides the ability to explain automated decisions, including why a recommendation was made, how different data would have led to a different decision and what parameters can be tuned to deliver a better outcome next time.
5. Have Configurable Confidence Settings: Provide actionable tips for administrators to make it obvious and easy to configure automated behaviors. For example, your virtual agent should include levers and dials to control how “confident” the AI model must be before making a prediction or recommendation. AI for the enterprise only delivers on the promise of better service when mistakes can be easily corrected by human subject matter experts. While AI is smart, it’s never as smart or as aware of nuance as human process owners.
6. Be Extensible Across Domains: To be adopted by employees as an alternative to phone or email support, your virtual agent must “speak the language” of all employee services. For example, ensure your virtual agent is able to answer IT questions, as well as HR, finance or facilities questions. Alexa wouldn’t be valuable if she only knew the weather. Similarly, your enterprise virtual agent must be fluent across service domains.
7. Support Omnichannel Conversations: Make sure your virtual agent is available wherever “the eyeballs are” at the moment of truth when a problem occurs. Virtual agents only available in special web portals or standalone apps won’t be used. Meet users where they already are to achieve meaningful call deflection. Email, SMS, voice portals and Slack are where they expect to receive service.
8. Scale Elastically As Conversation Volumes Increase: Your virtual agent must be load-tested to guarantee performance at peak times with peak traffic volumes. Validate that your virtual agent platform doesn’t throttle API usage, scales elastically using cloud resources and is integrated with external applications using modern integration techniques.
9. Provide Data Security Settings For Regulatory Compliance: Ensure your strategy doesn’t expose employees’ personally identifiable information (PII). They may ask questions and reveal information that shouldn’t be stored or transmitted unencrypted. Your virtual agent will only be approved by InfoSec if your AI platform is able to detect and obfuscate PII and support both cloud-native and on-prem model training.
10. Offer Predictive Insights: Winning with AI in the enterprise requires harvesting insights from automated interactions. Intelligent platforms should recommend what to do next based on what’s currently trending. For example, spikes in incidents related to Wi-Fi or email may indicate that an unapproved change to a server’s firmware should be rolled back. Select an AI platform that automates service but also suggests how to remediate issues before they occur.
Use the above as a scorecard to assess your AI readiness. If you score a nine or ten, launch with confidence. Score an eight or below, and you’re better off waiting.
Before grading your readiness, align stakeholders — InfoSec, line of business service owners and live agents — around business priorities, then proactively review your strategy by discussing each one.
Conversational AI is the new UI, and enterprise virtual agents are soon to be ubiquitous. According to Gartner, 70% of enterprises will use AI-driven automation for employee service by 2021. The other 30% will try and fail. By understanding how intelligent software is deployed and managed, your organization will be among the leaders.
Note: This article was originally published on Forbes.com on October 2nd, 2019.