Lawn Vision

Overview

For my senior year project, I collaborated with three peers on a sponsored project with Porch.com to build a lawn care calculator that estimates your lawn size based on images of your lot property using Google Maps and Computer Vision.

Why

Porch's business structure estimated the price of a lawn care job by using lot size as a favorable variable. In many cases Porch was over and under-estimating price quotes and thus losing potential customers. I collaborated with members in Porch's Design, Business Analysis, Data Science, and Development teams to understand the infrastructure we were building on and assess the business value of such a project.

Process

Problem Space

Porch priced its lawn care jobs proportionally to a customer's lot size.

Working with Porch's growth manager, Adam Guenther, to understand the strategy behind lawn care services it became apparent that we had to:

My role as a UX product manager: I collaborated with members in Porch’s Design, Business Analysis, Data Science, and Development teams to understand the infrastructure we were building on and assess the business value of such a project.

Project Management

Jason Nutter served as an incredible and agile coach enforcing full scrum. I handled jira tickets, wrote acceptance criteria, managed a prioritized backlog, enabled the development team to pick up tasks, and scheduled weekly sprint planning meetings with our mentors at Porch thereby allowing for a self-organizing team.

Design

Since this project was heavily an iterative CV algorithm research rendition, the design had to supplement and showcase this process in the most effective and seamless manner. I scheduled one-on-ones with Porch's Co-Founder and design director, Jake Cooney, to discuss their messaging, style guide, and design language. I created the front-end UI that is now live at Porch.com and a poster to accompany our showcase booth.


Testing

I worked with our data scientist, Lee Polla, to build a testing data set to accurately evaluate the effectiveness and accuracy rate of our product.

Takeaways

  1. This project was an exploratory research to examine the viability of using computer vision on parcel data.
  2. Shadows from trees made it difficult for us to determine what is lawn and what are leaves. However, our HSV algorithm picked up on the rigidness and blur better.
  3. It is difficult to test something that does not have a predetermined output. Test driven development would possibly produce more accurate results, however our strategy was to adapt our algorithm based on the end numbers we got from running the masked, unmasked, and processed images.
NEXT PROJECT