The benefits of implementing a AI Virtual Assistant; V.A., are numerous; 24/7 Customer support, an immediate answer to questions, no wait time, resolves basic requests on the spot, reduction in cost and the list goes on. The questions many companies ask are how they can do it successfully, how do they drive adoption, etc.

Starting your project with a good Knowledge base makes all the difference.

Creating a Good Knowledge baseThe backbone of any good AI V.A. is the knowledge base (KB) behind it. Having a good KB for Go Live ensures that users will keep coming back.

When creating your knowledge base:

1. Define Goals:
One of the key advantages of A.I. is flexibility and customization. At the same time, it is also important to set a goal before starting.
Ask yourself:

a. What is this AI V.A. for?

b. Who will be using it?

2. Identify Opportunities:

Keeping your goals in mind, identify the opportunities for an AI-based AI V.A. to help. Usability drives success. Therefore, understand the needs of your user base is how you will drive adoption. Using past data from your existing resources can give you a good starting point.

A few ways to do this is:

a. Analyzing historical data

b. Reviewing top viewed documentation

c. Run a ticket analysis of the last 3-6 months

d. Look at call/chat trends etc.

3. Knowledge Creation Team:

Decide who is part of the knowledge creation effort.

Including super users; those who submit the highest volume of request or issues, and SMEs (subject matter experts) in your KB creation efforts not only improve your KB content but it also creates ambassadors for your AI V.A. long term.

4. Additional Sources:

Focus groups are a great additional source of information. They can help you flag the important elements from a typical end-use.

You should use these conversations/sessions to identify the top queries and tasks.

Best Practices for KB Creation

Now that you have an idea of what your end-users are looking for and need, it’s time to create your KB.

Here are some guidelines on creating your KB:

1. Q&A Labeling

a. Avoid Question format
Just because it is called a question and answer pair doesn’t mean the label has to be in a question format.

b. Try to be concise but meaningful
You want your question label to be meaningful enough to convey what information they will learn from the answer.

c. Keep your end-user in mind
Make your labels easy for an end-user to identify, yes if I choose this option I will get the answer I need.

d. Use Neutral Language
Try not to use filler words and use neutral language instead. This increases the success of the AI V.A. matching to what the end-users ask

Bad Examples: How to replace my laptop?
Good Examples: Replace a laptop

e. Question Variants
Question Variants are alternate ways users may ask the same question. Therefore, all variant phrases mean the same and should elicit the same response from the bot.

Examples: what's the forecast, how's the weather looking, is it going to rain

Question variants are not words that mean the same semantically across contexts. If the words mean the same within the entire knowledge base, they qualify as synonyms. This includes:

· Abbreviations

· Acronyms

· Typos

· Synonyms

· Compound words with hyphens/spaces etc.

To add synonym, please email your Customer Success Lead.

2. Answer/Response

When deciding how you want your bot to respond to a specific question keep in mind:

a. Conversational Experience
A using a AI V.A. is a conversation experience and not a document. The end-user asks a question and the AI V.A. gives the answer or directs them to an answer.

b. Use Visual Aids
Not everyone likes to read so use visual aids (images, GIFs and videos) where it adds value.

c. Be Concise and Straight to the point
Try to be concise with your answers. There is no need to include unnecessary information.

d. Break it Up
If the answer to a question is long, consider breaking it up into smaller sections or use the multi-step QA option.

e. Accept and Adapt
There are certain questions that end-users will ask that require human interaction. This can’t be avoided.

You don’t want your bot to look like it is unknowledgeable, train it on scenarios that will require a ticket or live chat. With Rezolve you can use our re-usable task to flag quick ticket creation or live chat initiation.

Tip: If you are using an existing knowledge base, don’t just cut and paste

Successful UAT

It is important to review the end-user experience before Going Live with your chatbot.

1. UAT 1

a. Have the same Super Users, SMEs and/or focus group participants use the AI V.A. for a defined period.

b. Take their feedback on gaps, usage etc.

c. You will need to review it, define which would add overall value, categorize immediate most have changes vs those to implement at a later date and implement the immediate changes.

2. UAT 2

a. With a broader group of people start your UAT 2 testing. Have them use the AI V.A. and provide feedback for a defined period.

b. Create an easy way for people to contribute; such as a forum, you can implement during UAT 2 and allow end-users to use also once live.

c. Whichever way you decide to collect feedback, again you will review it, define which would add overall value, categorize immediate most have changes vs those to implement at a later date and implement the immediate changes.

Depending on the results of your UAT 2, you may want to consider doing a UAT 3 with another group of users or go live.

Successful Go Live

1. Announcement/Communication:

a. Announce the new tool to support staff and end-users. Let them know how to use it, what they can use it for and the value it adds to their day-to-day.

b. Reassure them that the AI V.A. will continuously grow and improve. Give them an incentive or a buy-in such as with their help of usage the AI V.A. will improve.

2. Additional Methods:

a. Have support staff include a link to the AI V.A. in their signature

b. Update email templates to include a link to the AI V.A. also

c. Update existing documentation to note that the AI V.A. is an avenue to find the answer

Manage KB in the Long Term

Knowledge management is not a one-and-done process. To keep your AI V.A. successful and to drive adoption you must continuously work on improving it.

1. Monitor:
Initially, you want to monitor the Train section daily. Gradually you can reduce the monitoring to every other day to weekly. We don’t recommend you do less than weekly.

2. Encourage Usage:
To encourage usage, contact users; can be via email etc, whom had an issue and let them know this has been addressed or we are looking into it.

3. Identify AI V.A. super users:
You will now have bot super users. Speak to them to find out what they like/dislike and what they want to see next for the chatbot. These will become your newest ambassadors.

4. Communication:

Anytime there is a new feature or something noteworthy happening with the AI V.A. communicate it. You want end-users to know this is a continuous process and that you hear their needs and are addressing it.