IGOR KALENNYY - Pronto

Overview

"Pronto" is a chat bot which could be built into Yelp's app to help find places to eat. It uses AI Deep Learning algorithms and Data Mining to interrogate the data inside Yelp's user reviews and, with a user's permission, it can leverage social networks data to incorporate their friends’ recommendations. All that would drastically cut down on the time-consuming process to suit a user's tastes and needs in finding food quickly.

Note:

This design prompt was specified as a group project assignment in a graduate design course that I took at Indiana University, Bloomington

Problem space and definition

What

Yelp wants to add a Conversational UI element to its platform to improve the experience of the users and also exploit Yelp’s new Machine Learning algorithm which can extract ratings and comments from Yelp sites to help people make an informed and trustworthy decision to quickly find what they want.

Why

Yelp wants to explore alternatives to their current method: Users have hard time finding the food they want when they are hungry because they need to sift through a lot of information, including reviews. They need to either know the specific restaurant they want or search by different criteria. It is labor intensive and time-consuming.

Example screenshots of a typical process the users have to go through

Long lists of places

Many options to consider

Long reviews to read

Users experience hard time finding their food, especially when hungry, because they need to process much information, including the reviews to simply suit their needs. It adds much of mental load and can create a shopper anxiety which would ultimately lead to poor and uninformed choices at the expense of quality and their time.



Roles and Responsibilities


Project Duration

  • Two weeks
  • September 2016

Team

  • Igor Kalenyy
  • Diandian Cao
  • Cecilia Gutknecht
  • Ryan Griggs

My role

  • Research
  • Concept co-generation
  • Ideation
  • Co-Sketching
  • User Testing
  • Documentation

Users and Audience

University Student


Headshot from: Creative Commons. https://www.pexels.com/photo/photography-of-a-man-holding-longboard-792529/
(Click on the images to enlarge)

Process

Primary Research / Interviews

Figure 1. Primary interviews

Results:

  • Students tend to trust their friends’ recommendations to find new and recommended food places
  • When viewing results, they prefer a few options to choose from rather than from a long list.
  • "Word of mouth” greatly influences their decision to choose a restaurant versus reading different reviews

Research Insights and Product Goals

General breakdown

Intelligent Assistance

Provide a quick and easy-to-use intelligent assistance to find food places of interest

Brief but Meaningful

Present a manageable amount of meaningful information to make an informed choice

Smart and Relevant

Deep learning algorithm will infer the specifics to suggest tailored and relevant info

Detailed breakdown

Data Mining / Services

AI Deep Learning

What do your friends suggest?

Leverage social networks data to incorporate your friends’ recommendation of dishes or restaurants.

Learn about the user

Suggest by learning about user’s tastes and preferences based on their own social network posts and “likes” related to food choices

Overcrowding and long lines

Use Google Services to watch nearby restaurant’s peak-times to avoid long wait times

Reduce input error and simplify

Provide the user with the preset choice selectors to ease the typing of responses and reduce error input

Smart chatbot

The 'chat bot' will constantly learn from the user's responses to increase its understanding of their habits and preferences

The power of personality

Make the chat bot's conversation quick, relevant, polite and human-like


Relevant, helpful and timely

  • Look for local food places' daily specials, free food events and food trucks
  • Filter on local sport or cultural events involving congestion and overcrowding


Teamwork and big questions

Sketches / User Journeys / Design questions

  • Where would Pronto 'bot' reside in the interface?
  • How would a user find the Con-UI?
  • Should it be re-active or pro-active or a mix of both?
  • How many questions to ask before the results shows up?
  • What is the appropriate number of options per question?
  • How many items should be in a result?
  • How does the user navigate through the results?
  • What are the consequences of mining of social networks data?

User Testing

Figure 2. User testing

What we found:

  • People prefer up to 3 options in the results (displayed as a row)
  • They wanted to swipe to the right or left to see the results
  • They wanted to see the ‘extended’ quick info view of the result
  • Integration of social networks into the search was found as valuable
  • Allow a chat bot to be easily accessible on-demand inside an original Yelp's search screen

Concept demo

See a quick video presentation demonstrating the final concept


Video

Figure 3. Concept's video


Our goal was to make the 'bot' not only analyze the data and serve the information but also attribute it with some human-like conversation characteristics when 'conversing' with a user so that it would seem less 'robotic' but more human

Outcomes and Lessons


Future product's strategies based on the user-testing feedback

  Quick feedback

Do it quickly right from the Con-UI ( also allows Yelp to get more data for future analysis!)

Dislike feature

Indicate to Con-UI to exclude a place in future results

Message Center

A Con-UI “bot” can send messages or recommendations for a later retrieval