AI Micro-service to improve Search
Background: Search on Smartbox App isn't the best as search is mostly location oriented and if our audience enter any other parameter, we show up max 3 recommendations and the result is a jumbled list of irrelevant results.
Problem statement: How might we improve search and results displayed on the App?
Team: Android lead developer, API Tech lead, Integrations developer, Digital content specialist, UX
Stats
- Smartbox's top 3 markets are France, Italy, Spain
- Around 11% of App usr base use 'Search' feature
- From 'Search' on the App, 45% users go to the list page, of which 22% head on to the product details page
- These users spend an average of 15minutes search on the app, and drive 5% conversions of all Smartbox purchases via the app
- Adults typically type 40 words per minute, people using 2 thumbs can type 38 words per minute, people speak 110-150 words per minute
- There are 100k more users, 300k more sessions and 400k more conversions on Android than on iOS
- Most Android & iOS devices have default microphones within their keyboards that can be shown/hid as required (dev needed)
Opportunity
According to ContentSquare, the opportunity is significant.
- Converting just 1% more of overall App user base who use Search could reel in approximately 6k in revenue/year
- Converting 2% of overall App user base who use Search could reel in approximately 13.5k in revenue/year
- Converting 10% of overall App user base who use Search could reel in approximately 68k in revenue/year etc
Potential KPIs
- Increased customer satisfaction
- Increased App re-engagement
- Increased conversion rate
- Encourage repeat Buyers
- Accessibility for users with speech or fine motor disability
- Saving costs: If new products are added to our catalogue, AI should be able to find all the relevant existing and new/latest gift boxes, thereby (potentially) reducing time and manual effort of Search & Optimization team
The strategic, North star vision
A step by step approach to achieving a personalised recommendation engine. The first goal is to break the location specific search barrier.
The UX happy path
Workflow realities with backend API for Search micro-service
High-level Tech workflow
Detailed Tech workflow
To make this happen, we looked at numerous Speech to text services and finalised on Google API. Whilst IBM Watson seemed to have most of the languages we need, Google won the match as we had all used Google Translate, and voice search and the responses were more accurate than relying on a service we were not familiar with.
UX: Bringing workflows to life
Whilst the Tech were figuring out how to build the micro-service, the wireframes began to come to life. At this point, there were conversations, discussions, debates and problem solving every 5 minutes
Smartie
And that is how Smartie, the (in-house) AI Search Assistant, was born.
Prototype vs Live demo
A click-through prototype brought much needed front-end usability clarity. The team pulled off a spectacular live demo in French, which won us the Popular Vote.
UI prototype
Real time demo
--x-- --x-- --x--