Questions & Answers – Don White, Satisfi Labs – Voice Tech Podcast ep.037

Don White Satisfi Labs

Episode description

Don White is CEO and Co-Founder of Satisfi Labs, an AI-powered Knowledge Management Platform that offers conversational search and commerce solutions for fan, guest, and customer engagement. Don describes the types of companies and industries most interested in implementing question and answer solutions, and reveals the number #1 question their customers ask.

We cover how these Q&A models work, how scalable they are, and how they are deployed to clients. We also discuss the benefits of voice search, and how ready consumers are for the transition. Don also shares some design tips too, such as how to handle unexpected user queries, the reasons for response bias, and the importance of asking follow up questions. And we answer the most crucial question of all: where to buy Bacon on a Stick.

Highlights from the show

Who uses Satisfi labs product?
  • Companies with high volumes of customer enquiries e.g. retailers, sports stadiums. They put all their customer queries from multiple sources into a indexable database that is suitable for voice search.
What’s the number #1 question customers ask?
  • How do I buy a ticket? There’s a fat head of 20% of common questions, then huge long tail of unusual questions.
How does the system respond when the product/service they ask for isn’t available?
  • The Model can recommend similar products
How does a fan find out where they can buy Bacon on a Stick?
  • With Satisfi, they can just ask.
In what ways are Q&A models scalable?
  • Customers benefit from machine learning models trained on data from non-competing companies in other industries
What is needed for voice commerce to take off?
  • Greater adoption of Apple pay, Google pay, Amazon pay, which help keep financial credentials private when purchasing over voice.
How do you see voice being used cross-industry?
  • It will enable continuity of service over multiple consumer destinations, e.g. sports, entertainment, hospitality, retail, tourism
How ready are we for voice search?
  • Text search has already moved from keyword to full sentence, so only a small change in behaviour is needed to switch to voice search.
When are keyboards and screens more suitable than voice?
  • For options greater than 3 in number, you need to tap/type rather than hear a long list of options.
How are virtual assistants biased in their answers?
  • They often skip the list of options and just recommend a product. But these recommendations are biased by vendors based on user reviews, retail stock levels, product expiry dates etc. To retain the trust of users, the voice-enabled system should also explain why the recommendation is being made.
How does the Q&A system work?
  • The list of customer queries and possible answers are identified. The answers are ‘slots’ to be filled by matching up with possible questions.
How automated is initial training and deployment for a customer?
  • 25% manual, 75% automated. The system can predict slots it hasn’t seen yet. Biggest manual task is getting data into the system, then it gets much easier.
What trade-offs are there?
  • Developing a world model sometimes conflicts with maintaining data privacy. Some customers wall off their data to maintain competitive advantage, so they don’t contribute to the world model e.g. Retail companies tend to be more siloed, and less likely to share. However, they don’t get to benefit from the world model either.
What advantages do smaller startups have over big tech?
  • Startups can develop more specialised models for niche use cases, and can also better protect these models to maintain a competitive advantage.
What initial assumptions did you have to re-examine?
  • The assumption that there’s a perfect answer for every question. However, 20% of the time there’s actually a better answer than the one the system gives. The customer’s question can often offer a non-optimal solution, so to get the best answer, you need follow up questions to dig down into their real needs.
Does the system ask follow up questions unrelated to the initial question, just to build out the world model?
  • Not currently. Although this would be useful, customers are currently cautious about capturing too much user info in the wake of the Facebook scandal.
How to handle unexpected questions from users?
  • At least show that you’ve understood the query, even if you cannot provide an answer to it. This maintains trust.
How many users flirt with your voice assistant?
  • A surprisingly high percentage..

Links from the show

Subscribe to get future episodes

Join the discussion

Support the Voice Tech Podcast

Share this article

What do you think?

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Related Posts

Mike Page Phebi
Petar Ojdrovic Home Service
Top Podcasts

Get notified about new articles