ChrisSteinke2
ServiceNow Employee
ServiceNow Employee

I think we’ve all run into it – you are using a chatbot, you type in something you think it should be able to handle and in response you get “Sorry. I don’t understand.” Where to go from there? Maybe you type in your request slightly differently, or you just get frustrated and give up. Personally, the most frustrating thing for me is ending up in a loop where over and over I try to get what I need only to have the bot tell me it doesn’t understand. If you are using ServiceNow’s Virtual Agent and haven’t been keeping up with advancements over the years, this might just be the experience of your own customers.

“I’m working on self-generating computer program. I just need to get the first line done.”

Since August of 2018 I have been a part of the specialist Solution Consulting team that helps to support our clients and account teams with the Virtual Agent (and Performance Analytics, Predictive Intelligence, Machine Learning and more). The evolution of chatbots and how we interact with them has been fascinating to witness. One area that has surprised me is the different ways people interact with chatbots. Although it surprised me at first, it is common to see situations where someone types into the chatbot:

“approval”

While another person will type in:

“So last week I received an approval email and I clicked the approval button which did a reply message with approved but I don’t know if it worked so can you tell me if the approval on new my new hires equipment was approved? Also about two weeks ago I had an issue with email so I need to know if that is fixed.”

Without boring you with details, chatbots are not actually people and they can’t understand everything. Chatbots are pre-built to handle certain situations, which we call Intents. An Intent is something like resetting a password, checking the status of a ticket, helping to debug a VPN session, or resetting a Citrix session (which are all Intents provided free with SN’s Virtual Agent). Chatbots can support hundreds of Intents. However chatbots are not like some magical AI that can dynamically do anything you want on an ad hoc basis. Because a chatbot can for example reset a Citrix session, that doesn’t mean you can ask it to show you the Citrix log from a particular server or user. If you want it to be able to do that the Intent and flow would need to be created beforehand.

This may sound disappointing and trust me, I know. I still remember the moment it was explained to me that chatbots are not super beings and in fact, can only do pre-determined activities. However even though they can’t do everything they provide tremendous value and cost savings to companies by deflecting calls and emails to a Service Desk. A typical medium-large sized customer can expect the Virtual Agent to deflect 10k ‘status updates’ a year, and at an average of $10 per ticket opened by a person on the telephone that results in a cost savings of $100k per year for just a single topic. Wow!

Make a better experience

Knowing that the chatbot can’t handle every scenario, how can we create the best experience when people ask for something that the chatbot can’t do? We call this situation a ‘Fallback’. When the chatbot doesn’t understand, what should the fallback response be? If you are using the default Topic (Name=Fallback Topic.) from 2019 this is what happens:

find_real_file.png

This is not bad as it directs the user to what the bot can do (Show me everything) but it is not the best experience in my opinion as it will happen over an over. What if instead you could have the Virtual Agent leverage something like your existing Knowledge Base, and maybe then if it still can’t find what they need they could talk to a Live Agent? That is actually how the newer, out of the box “Search Fallback Topic.” works:

find_real_file.png

Instead of just saying “I don’t understand” the chatbot goes immediately to search the Knowledge Base. If the user doesn’t find what they need in Knowledge, it allows them the option to chat with a real person. Also note that by default it searches the KB as the logged in user, so if you have access rules or multiple KBs this will all just work. What a vastly improved experience!

Now for the kicker – all you need to do to make this happen is Publish the ‘Search Fallback Topic.’ (if that hasn’t already been done) and then set that as your fallback. You do that by accessing the Default Chat Experience record in the ‘Custom Greetings and Setup’ menu item and updating the related list of Setup Topics.

 find_real_file.png

 That’s it. This is literally a 1 minute change you can make which will vastly improve the experience of your users in situations where they ask for something the chatbot cannot handle. If you are using AI Search already, your instance should already be using the AI Search Fallback Topic in which case you don’t need to do anything.

So what you are you using for your fallback? Have you taken it a step further to show more options? Show less options? Let us know your thoughts in the comments and any ideas for making the experience even better.

5 Comments
Lynda1
Kilo Sage

We customized our Fallback when we first implemented, the out of box fallback in the (version before Quebec, cannot remember where we were visiting then), drove me nuts the bot was giving no response, or that sorry I do not understand. So we had it search the KB articles then catalog if it finds nothing in the KB.

I noticed that over time as the AI gets smarter, if part of a paragraph that a user enters is in the intents, the topic displays, kind of cool!

Chris D
Kilo Sage
Kilo Sage

Absolutely. Great post. First thing to do indeed should be to figure out that fallback.

Here's the dilemma but necessary thinking: the whole point of VA is to deflect SD calls/chats and ideally even prevent incident-creation in the first place. BUT, what about your fallback? Should it include these two options?

For us, the answer is yes. The fallback should be two steps: 1) Search (AI Search if you have it!) then 2) Give the user a choice: live chat, create an incident (using a generic incident-creating topic), or try again.

Sure, it may suck to have a user create a potentially vague and barebones incident - we ask simply for a) app name, b) short description, c) additional details, and d) optional attachments - but the tradeoff is that what is lost in Incident detail (which the agent is now going to need to get) is gained in VA user retention which is critical to keep them coming back and taking advantage of future improvements.

Also key to go along with the fallback is the Anything Else topic. Many of our topics do simply create incidents, but we recognize that some users are not satisfied with that - and in fact, some might have higher priority issues that need to be dealt with asap - so our Anything Else topic doesn't simply let users start a new topic, it directly asks them if they'd like to chat with a live agent for more assistance. Again, the balance is user experience and retention.

Victor Chen
ServiceNow Employee
ServiceNow Employee

The power of genius results!

Tricia Cornish
ServiceNow Employee
ServiceNow Employee

Agreed, Chris - I like to show customers how the 2 fallback options can work together to accomplish this result. AI Search Fallback is only a search, and the Fallback gives the user the choice of what to do next. It didn't occur to me how to edit the Anything else - thanks for the idea! find_real_file.png

ChrisSteinke2
ServiceNow Employee
ServiceNow Employee

Another really cool innovation I have seen recently is adding in the FAQ Knowledge Base article search functionality to the Fallback. In this example, the customer had configured the Fallback to:

1) First search FAQ KB articles

2) Then AI Search

3) Then Contextual Search

 

When done correctly, the user doesn't even realize the FAQ KB articles are being searched unless a close match comes up. Just another neat way to make an even better experience.