Brainstorming: Great FAQ voice app?


#1

Based on a recent post in the Jovo Developer Slack by @rjolayolay, I am curious to hear how you folks would approach building a FAQ voice app. So, let’s say you provide a product or service for which you frequently receive questions, and naturally some of these questions are repetitive, at least in their intent, so that you have a list of answers for the most common questions at hand. Now some questions are:

  1. For which kind of products or services would it make sense to provide their FAQ via a voice app? E.g. would it make sense for a web- or mobile-based service? An online or offline retailer? A restaurant or fitness studio chain? A public institution like a library or university? A video game? A celebrity?
  2. How would you store the content? Probably in a CMS, but which one? Are certain features like collaboration particularly important in such a voice app?
  3. How would you structure the language model? One intent per question? Slot-based questions (like How can I {questionType})? A very open language model in combination with your own NLU?
  4. How would you educate users on what they can ask? Would you give a list of questions in the help intent handler? Or give a short random list of example questions after each answer, like “Opening hours are from 9am to 7pm. What else do you want to know? You can ask me about which flavors of milk shakes we have, or about which payment methods we accept.”?
  5. How would you monitor the Skill to learn which answers were most requested, how helpful the answer were, and what answers are missing?

The idea of this post is to have brainstorming character, so feel free to pick any one aspect you’d feel like sharing an idea or thought about, or to add more questions about possible features. Looking forward to hear what you think!

Disclaimer: This question may sound like we’re about to build an FAQ Skill and want to source ideas, but we honestly aren’t. I’m happy for any great ideas come around for everyone to make use of!


#2

1.One place an FAQ skill would make a lot of sense is products that don’t come with a large instruction manual and have minimalistic packaging. Think [Gadget name]. You could have “Ask You Assistant to talk to [product name]” on the packaging. The skill could contain an FAQ, tips / tricks to use the product, and contact info for support.

If the product is something that would be used at home, the user might find it convenient to answer a product specific question a search result might not quickly find the answer to. For example “Hey Google, ask KitchenGadget if KitchenGadgetXYZ is dishwasher safe”.

4.Users might not care to learn about what you can ask the assistant, if they invoked the action they probably have a question in mind already. It might be more useful to ask if they want a quick product tip after answering the question to increase engagement.


#3

Hi @Marko_Arezina, thanks for your great answer! :+1:

I can absolutely imagine such a badge “Ask your assistant about XYZ” to be standard in the near future, or similarly something that customers would expect. In an optimal world answers would not only come from the supplier, but could also be crowed-sourced (like from the “Customer questions & answers” section of Amazon product detail pages, or Q&A-sites like Quora).

Maybe once such voice apps are standard, customers would also have a clear idea how to formulate their requests in order to get a helpful answer - Similar how it’s a minor skill nowadays to describe a problem such that a Google serach returns a helpful answer. :smile:

I think this option of asking a question directly without the need to educate the user is the perfect solution, but also the one that demands the most effort from the developers beforehand, since they have to do solid user testing to anticipate both the right questions and their formulations.
Also: Good point about the product tips! I bet there are a thousand product people out there that wish their users knew about the little perks and extras they built into their products (I’m thinking of wahsing machines, where most users probably just use 5% of the machine’s capabilites.).