Organising the Zoopla Hackathon

Zoopla just ran a successful 2 day hackathon and it all started with 3 people and a dream. Wouldn’t it be great if we had the time to work on some of our own ideas and showcase the innovative potential of Zoopla’s biggest asset: the talented people working at Zoopla! This is my personal reflection of the hack-day organisation.

3 people and a dream: from the right Matthew Chubb (Software Engineer), Jyoti Goyal (Senior Product Manager) and me (Data Science).

The first milestone we had to reach was senior leadership buy-in. If you want to take 2 full workdays out of the day-to-day than this is not an insignificant investment for any company to make. Zoopla is working towards challenging targets of becoming Number 1. We couldn’t just go with the idea of a hackathon alone but with a compelling case for it to go ahead.

Over a series of lunch meetings we brainstormed about what motivates people to participate in a hack-day. But as important, what the company would expect us to deliver from 2 days of hackathon. This let to some difficult questions: Should there be a theme? How much should we curate the event? A selection of well curated topics really helps with preparations and setup, but wasn’t the original motivation to allow innovation and creativity away from prioritised backlogs?

In the end, we agreed to focus on company culture rather than the details around themes and topics to get senior leadership buy-in. And so our hackathon had an objective: strengthen the Zoopla culture! Zoopla is a growing business and most of us work with a strong product focus in individual product teams. As an example, being in touch with other teams and for new starters to feel well integrated is not easy in such an environment. But culture is extremely important to Zoopla and culture had the senior leadership recognition to get us a more or less official go ahead. Culture is also a crucial part for a data science team to be successful and deliver.

A focus on culture for a hackathon is a perfect win-win scenario! Because culture eats strategy for breakfast.

However, we knew we only delayed the inevitable decisions around the hack-day theme and topics. But with the go-ahead we had the option to ask people at Zoopla to submit ideas and find a more democratic solution to that problem. Unfortunately, it’s not that simple. This needs engagement and so far no one had heard of our hackathon nor reason to feel excited about it.

Do not underestimate how sceptical we all are in particular when we are busy.

To increase engagement and the chance of success the hackathon needed a professional identity as much as details around the themes and topics. At the same time it dawned the three of us how much work it will be to organise the hackathon. We had to start delegating.

Delegate, delegate, delegate!

Heldiney Pereira (UI Designer) gave us a hand with some desperately needed polish: e-mail templates, logos and backdrops.

As we started to send hack-day communications on all channels — from face to face discussions, to stand-up takeovers, email and slack channels — the ideas for topics and themes started to come in.

As it turned out we were too worried about the topic selection and curation. Common topics quickly emerged from the submissions and most of the topics were highly business relevant. After we spend some time grouping and refining the topics we were back to delegation!

We identified facilitators for each topic to look after driving the engagement for the topic and to determine necessary setup and preparations. Without the help of our highly motivated facilitators we simply wouldn’t have been able to get the hackathon delivered.

We put together posters alongside our communications to get people sign-up for a topic. This was a good success metric to measure our efforts in driving engagement.

Creating lasting excitement and engagement is more difficult than you might expect.

With this out the way, we started to organise the details of a hackathon and gosh was the list of things to do endless:

• Working with the facilities team to find a space for the hackathon
• Budget planning and approval
• The food and drinks for the event had to meet dietary requirements and the no single use plastic policy of Zoopla
• The social events for the hack-days
• Hackathon presentations
• Awards, voting and prices
• Content capture for PR and marketing purposes
• How to engage our remote offices
• Getting buy-in and approval from line managers for people to attend
• Approval so our contractors are able to attend

Based on feedback, we decided that a market-place style presentation of the hackathon projects would be the most dynamic and engaging format. It’s also the best usage of our Exec’s time which we wanted to attend the demonstrations.

No death by powerpoint!

Regardless the amount of planning, there was a lot of things left to the last minute which caused some trouble: the trophies we ordered were not the trophies we received and fixing it did cut it very close. Many more people showed up than planned for the hack-days as well as the presentations and awards ceremony. This caused some last minute setup challenges, cutting out an additional 100 odd voting cards with toy scissors as people dropped in for the hackathon presentations and running out of trophies because of the much bigger team sizes. We learned valuable lessons on the size of margins in event planning! ??

At the end, the hackathon was a great success which made all the work worthwhile!

What did Data Science hack on? Kitchens!

A cross-functional team of data scientists, backend and frontend engineers set out to find Zoopla’s most beautiful kitchens. The aim was to sort property listings on Zoopla by kitchens and recommend properties based on the similarity of their kitchens. It’s been a fun topic to work on with some great data science challenges but also some great business relevance: the kitchen can easily be the deal breaker for people who are looking to buy their new home.

To deliver a fully functional proof of concept the team created:

• A CNN to classify images of kitchens
• A numeric fingerprint of kitchen images
• Extract kitchen features like chrome taps, extractor hood etc.
• Calculate a jaccard similarity between the kitchen features and a cosine similarity between image fingerprints to combine them into a kitchen similarity score
• Batch score Zoopla property listings ahead of time and load labels and scores into a fast caching layer which acts as the presentation layer for the models
• Create an API to query by kitchen and to return property listings and images of similar kitchens
• Change the Zoopla frontend to dynamically show kitchens and property listings sorted by their kitchen similarity score

The ability to deliver this in just 2 days was only possible because the solution built on the computer vision and infrastructure foundations we created in the data science team over the last year. Nevertheless, this was no easy feat and won the project the hackathon award for the cleverest hack.

Massive kudos goes to Natalia Koupanou and Harpal Sahota for making data science at Zoopla award winning! ????