Although AWS makes it very easy to deploy a QnA bot, AWS makes it almost impossible to analyze the data received. Unless of course you love to dig through dozens of CloudWatch logs in un-human type reports.
Seriously, the designers of the UX/UI really need very long vacations as many of the designs are convoluted, and too complex to waste time attempting to figure out. [Ummm just like their newly designed UI console.]
QnA Bot Install Tips
Step 1. Is to deploy via CFT per the AWS post. You may want to dry-run a demo first, and consider using the “public” client, and keep the nodes down to 1. [Practically everything on AWS = $$$$]. Our demo video uses the very basic options.
Step 2. Bookmark your URLs from the output of the CFT as shown in the image. Drill-down on the plain completion title, and go to the outputs tab.
Step 3. Your password and login link will be sent to the email address you provided during deployment so don’t use a placebo.
Step 4. [Important!] Create a test question, and test the bot’s response first! This will generate the required output log for the next section. You must use the QnA client to test, not the designer UI testing tool.
A51 AWS QnA Reports
aMiSTACX to the rescue! Yes, we eat our own dog food, and look for ways to improve A51 all of the time. Since we have been using AWS QnA bots for a few years, and when AWS changed their reporting into a big mess, we decided to develop a human friendly reporting tool in A51.
Step 5. Add the Reports ARN and the Designer output link to your A51 user profile under your specific A51 account.
ARN can be found in CloudWatch >> Log Group >> Fulfillment
Drill-down to get the ARN.
Step 6. Just set a date filter and go! You’ll poll questions asked and answers, and all sorted by date and time. We even included a handy export feature, and quick-link to your designer dashboard.
A very cool tool to save you a lot of time, and keep your Bot productive!!
Introducing MaceyBot, our bot that makes use of AWS and A51.
Real World Example
How do we use our Bot? When you ask a question that our bot may not have an answer, we will review the CloudWatch data from A51, and look for unanswered questions, or logic response issues. [Every 24 hrs.]
We will then make use of the link in A51 to update the utterance Q and A. For example, a customer was attempting to gauge how smart our bot was, and asked “How many people in China?”. For privacy, our bot is not interfaced with Alexa, so we manually updated our bot’s response in 24hrs.
In this way, we have a bot-human connection to our customers without the ugliness of email, and also with the ability to answer questions in a one-to-many format for future customer usage.