Virtual agents: not so different from new human agents, actually.

I read a lot of articles and reflections on the world of chatbots, and there’s a strong consensus that 2020 marks the start of the accelerated adoption of virtual agents.  The technology is stable, customer openness is really encouraging, and the ROI is compelling.  It’s exciting to be finally harnessing what VA tech can do to provide consistent, accurate, always on, engaging experiences for customers.

With that acceleration comes a whole new raft of people being asked to stand up a VA for their business, and not knowing where to start – or how to convince the MANY interested stakeholders that they’ve got it under control.  I wanted to share a few pre-launch tips to help you get past that first hurdle, because the more we share the better experiences we create, and that can only help public perception about our digital friends.

We’ve had an employee-facing virtual agent for a number of years now, and Ella is part of our family.

More recently, in line with our new website, we decided to promote our VA to a customer facing role – and it didn’t go flawlessly. I’ll come back to that.

We’re bringing a lot more VAs to life this year for our partners, and one of the things that gives us credibility to guide them through is that we’ve done it to ourselves.  We’ve tried, failed, tested, learned, honed, messed up again, and learned more.  A great VA is not a set and forget process; it’s actually quite a lot like bringing on a human agent.  Let me tell you why.

Virtual Agents have a speed to competency

By far, the thing that makes people nervous about launching their virtual agent is:

What happens if it can’t answer the question?

It makes people twitchy.  Their palms get a bit sweaty.  They think that we should spend more time testing it.  They want us to pretend to be external users and ask things “like a customer would ask”.

We are inherently biased.  We have implicit knowledge we can’t UN-know, and we assume that customers will ask questions that they simply don’t have enough context to ask.  It doesn’t mean you shouldn’t do role playing, and that internal testing is a standard part of our pre-launch process.  It gives us some really valuable info, and picks up really simple bugs.

What is more valuable is using actual web chat transcriptions to make sure your VA can answer all of that.  If you don’t have chat, capturing the enquiries that come through your contact numbers.  If you have email, scouring these for trends of what people are trying to do.  Asking your sales team what their first phone calls with customers cover.  These data points all represent the voice of your users, and your VA should be able to handle all this stuff before it goes live.

The reality is though, a VA launches with what it was taught in the “classroom” – and it then has to learn as it gets more data.  Real customers ask weird things sometimes, and every question that the VA has to hand off to another channel is a learning opportunity.

If your VA can handle more than 70% of enquiries when it launches, you are on track.

Your curation crew should be getting that over 85% within three months; sooner for narrower use cases.

Virtual Agents have a nesting period

What you do in the days after your VA launches is where the real magic happens.  The critical things we monitor daily (and hourly, on the first couple of days) during the “nesting” period are:

  • Safety net: how many questions couldn’t be answered?
  • Engagement: how are people responding to the VA (reviewing both survey responses and tone of discussions)
  • Usage: are people using the VA the way you thought they would?
  • Uptime: any issues with it presenting when and how you wanted it to?
  • Handoff: if you’ve enabled a hand off to human chat agents, review performance of that routing and the flow-on CX.  If you’re sending them to a web form enquiry, check the detail of their need and any feedback they’ve provided.

While the use cases vary, if you’re watching the above, you’ll have a good view of whether your VA is performing well.

And whether you take ACTION based on those results tells the world whether YOU are performing well as the custodians of a newly minted VA.  You should be upskilling the VA with new content every day in the early learning period, and you should be tuning any messaging that isn’t working for your users.

Virtual Agents are part of your team

The users you engage in your VA journey are more than just your customers.

Your employees are represented by this VA – it speaks for your brand.  How do they feel about the persona, the tone, the function?

Your teams who pick up web chats from the VA, or web forms – they’re covering the VA’s gaps. How do they feel about what’s being missed?  Do they have the avenue to ensure that repeat gaps are getting rectified?

Your employees who used to handle these enquiries by phone or chat – how do they feel about having a Virtual team mate handling the repetitive stuff?

Engaging your people to give honest feedback about the VA at launch and ongoing is really important.  The VA can handle the feedback; it’s a bot, after all!

None of us is perfect

You’re going to make changes once you launch.  You’ll find something wrong, or a better way of doing things, or get a request you couldn’t have anticipated.  Take it in stride and incorporate it.  Your VA is an ongoing project, and it’s okay to try things and back them out.

Let me tell you what I got wrong on our most recent one – which happened to be on our own website.

We initially launched our external VA under the same avatar as our internal one, “Ella”.  It seemed like a great way for our people to enrich their connection with the digital colleague.

When we launched though, we found that our people expected the public facing Ella to be able to give them all the information that they normally get in their internal Ella engagements.  But internally, Ella is authenticated – she knows about their leave balance, our business information, their personal details and schedules. Externally….Ella had “forgotten” these things.

We realised that while our people were excited about a VA, we needed to make it clear that our VAs have different permissions and “skill sets”.  It sounds so obvious now…but there you go.  We all make mistakes.

We’ve rebadged our new VA as Ella’s younger sibling, and called it ASH – our Artificial Stellar Helper.  ASH can help with lots of things, but if someone wants to know something from the inner Stellar world, they need to speak with Ella.

You can chat with ASH now on the Stellar website here…and if ASH doesn’t know the answer – you can bet that it will when you come back in a few days!

If you’re launching your VA, you’re not alone in your hurdles, fears and hiccups.  Get ASH to send me a message if you want to connect to chat to a fellow VA advocate – or message me on LinkedIn!

About The Author

Rik Johnson
Rik heads up Stellar’s Intelligent Automation business, where she and her team of digital optimists are building and nurturing bots that bring out the best in people.

Rik has previously worked in product management, solution development, project management and commercial relationship management, and has over 16 years’ operational experience. She’s inherently curious, optimistic and enthusiastic, and brings this energy to each new product or problem she encounters.

Want to learn more about Stellar?

Get in touch with us today.