Increasing the value of Shoppar's Facial Analytics Platform



Shoppar.io, London – UK

August + September 2018

As part of Pivigo's highly competitive data science fellowship program (S2DS), I worked with Shoppar – an early stage retail analytics startup, as a full-stack data scientist.


 

Introducing Shoppar


Shoppar.ai is an early start-up retail analytics company, developing real-time data analytics technology and software for retailers. They work closely with major brands – including Samsung, Hugo Boss, Sky, Western Union and Heineken, and use AI to leverage data captured from in-store digital marketing displays.

Shoppar’s goal is to empower their clients and enable them to create more engaging and meaningful branded relationships with their consumers. Interactions that are truly customer-centric and go beyond traditional passive advertising displays. At the core of their platform, is sophisticated hardware and software that captures age, gender, mood and engagement in a split second, The data is then delivered to custom built client dashboards.


Our Work


Shoppar’s goal is to provide clients with a simple, yet revolutionary, data platform that delivers real-time data insights. Insights that clients can act on in real-time and provide them with a competitive edge. By July 2018, Shoppar had completed a rigorous test phase and validated their custom designed hardware and facial recognition software – underpinned by a deep learning pipeline built with Tensorflow). They had developed their dashboard application and they had begun working with a small number of high-profile clients, resulting in a rapidly growing data store. While their client dashboards presented a comprehensive picture of customers, with detailed insights about customer demographics and engagement, they were largely static. Shoppar had yet to find a way to effective leverage their rapidly growing data store.

I worked with Shoppar alongside three other Pivigo Data Science Fellows. Our role as full-stack data scientists was to develop machine learning framework that would enable Shoppar to leverage their customer engagement data. They were also keen to explore ideas around intelligent content delivery and the potential for clients to auto pilot their marketing campaigns, resulting in more effective marketing strategies and increasing value for clients. Our work would also validate the quality of their data store, and inform future product development.

WE focussed on three key areas:

  • Validating test data and live client data, developing metrics to assess quality, and identifying any systematics or biases in the data;

  • Developing a machine learning framework to advance existing capability around customer demographics and engagement; and

  • Prototyping ideas around intelligent content delivery (optimised to customer engagement) and to inform future development of a more responsive product.


Key Outcomes



1.

undraw_data_xmfy.jpg

A comprehensive data cleaning and analysis pipeline that;

  • Enables Shoppar to better explore client data, before data insights are deployed to their client facing dashboard application.

  • Helps Shoppar identify any biases resulting from their facial recognition and quickly act on technical problems associated with individual devices.

  • Enables Shoppar to bring in external weather data specific to each device location. Such data has previously been shown to influence spending behaviour and affect customer footfall.

  • Automates data updating and loading from BigQuery.

2.

undraw_predictive_analytics_kf9n.jpg

A modelling framework that leverages sophisticated machine learning algorithms;

  • Our prototype workflow comprises numerous python scripts and detailed notebooks, and includes diagnostic visualisations highlighting useful data & features that better predict attention span.

  • This enable’s Shoppar to immediately leverage their rapidly growing data store, and provides a useful workflow for future development and more robust predictions.

3.

undraw_developer_activity_bv83.jpg

Prototyped solutions for intelligent content delivery based on predicted attention span

  • Explored open source and proprietary solutions – APIs, scene detection code & object identification and labelling.

  • Existing marketing content was modified and optimised to match customer attention span and demographics.


Outputs



Process & Timeline


 

The Team


Data Scientists:  Arna Karick, Jack Roberts, Alex Dunhill & Antonello Macchia
Clients – Shoppar: Peter Ward (CEO & Founder), Kris Milne (CAO), Akshay Kumar (CTO)
S2DS MentorsAdi Andrei (Technical Mentor), Ben Turner (Business Relationships Mentor)

Project Management & Client Communication | Shared responsibility
Commercial + Technical Research | Shared responsibility
General Programming | Shared responsibility
Integrating External Data | Alex Dunhill, Jack Roberts
Workflow Design & General Prototyping | Shared responsibility
Data Visualisation | Shared responsibility
Machine Learning Algorithms | Shared responsibility
Object Recognition & Scene Detection | Jack Roberts, Alex Dunhill, Antonello Macchia
Commercial + Open Source APIs | Jack Roberts, Alex Dunhill

Mid-session Pitch | Alex Dunhill
Final Presentation + Business Case Study | Shared responsibility