Taking Home the Hackathon Prize

How the Analytic Partners team achieved an inspirational victory at I-COM

If you are a marketing measurement/data science/analytics enthusiast like we are, there is no better place to get your fill than the I-COM Global Summit. I-COM hosted its annual global summit in Porto, Portugal last month to explore marketing data measurement, among other topics. Analytic Partners participated throughout the event, including CEO Nancy Smith and Director Katie deGorter presenting an innovative case study on the use of weather data for client Scotts Miracle-Gro. But the most exciting part of the event for the team on the ground was the Hackathon. The annual I-COM Global Data Science Hackathon competition focuses on solving real analytics problems, accelerating outcomes that normally take months to achieve.  Analytic Partners sent an experienced team to Porto to compete in the Tier 1 Master Category for the coveted Golden Rooster trophy.

The Analytics Partners team “Red is the New Hack” were:

Before the event in Porto started, our team assembled to prepare for the intense 24-hour competition.  “We all got to Porto a day early,” said Joshua Bayne. “We had a chance to talk about what we wanted to get out of the experience and learned a little about each other’s work styles.  It helped that we all had really positive attitudes.”

Katie deGorter agreed, “Also we had a good game plan going in.  We divided the work efficiently.  We’re all multi-faceted, and that worked to our advantage.  Everybody had a role at every stage of the analysis.”

Even though the team had never worked together before, as multi-faceted  “full-stack analytic marketers” our team was able to do what Analytic Partners does best: Be Adaptive. And as experts in marketing analytics and modeling with wide-ranging experiences across industries, methodologies and tactics, each team member was able to pitch in at every step along the way and to be flexible when necessary.

The Challenge Begins

Each team competing in the Intel challenge was given 24 hours, a USB drive containing detailed PC performance and marketing data, and an actual marketing challenge to meet: “What is the impact of discussions in social media and brand health indicators on advertising effectiveness for high consideration purchases such as consumer PC sales in the US?” In addition, the teams were tasked with the challenge of accurately predicting 2 months of sales revenue for 22 different PC products.

Ioana Badea, Director of Global Insights & Analytics for Intel describes the challenge on the I-COM site as such: “For an ingredient brand like Intel, it is challenging to directly link marketing activities with business outcomes.  Being able to detect early performance indicators would help optimize our marketing investment and maximize our impact.”

Our team was ready to go. Marcus Del Rio brought the USB drive back to the team, anxious to initiate the game plan.  But something else happened.  “We didn’t want to rush in frantically,” said Caitlin Hill.  “We said: let’s be calm.  Let’s enjoy this.  So the first thing we did was yoga – the Sun Salutation.  Well, all of us except Marcus…”

Del Rio would have done yoga too, but he held the “data warrior” pose and downloaded data while his teammates limbered up. Everyone had their role to play.

The Data

The team began the intensive data landscaping process. This process of understanding and determining the accuracy of data spanned many hours. They ran into a snag when the data only went through December when they thought they would have media spend data through February.  But when you are adaptive, snags like this can’t throw you off your game. Adapting on the fly, they refocused their attention to the brand tracking data, which did exist through February. Combined with a multi-stage modelling approach, this data would allow them to tackle the prediction challenge without sacrificing insights or findings on the business challenge.

The second snag happened while they continued to process the data and undertake data landscaping. The team uncovered inexplicable inconsistencies in one of the data sets. Again, relying on their Adaptive training, they explored possibilities to understand where the issue was but soon realized there was an issue with the data itself. When they went to the judges, the organizers realized our team was right. And in fact, the AP team was the only team to discover that the data was wrong. With corrected data they returned to work.

Next, they encountered the metrics: literally hundreds of them.  “I was worried about that,” said deGorter.  “But Marcus had adapted an algorithmic approach that we ran to identify which metrics were most relevant.  We got the list down considerably and were able to spend more time exploring the validity and explanatory power of this subset of metrics further within the models.”

The Model

Ready to begin the modelling, the team provisioned for the long night ahead.  Raiding a local bodega, they stocked up on energy drinks and Belvita breakfast cakes.  The proximity of an espresso machine didn’t hurt.  “It was an effortless coffee-drinking environment,” Del Rio enthused.  “Tiny little cups.  You drink it in two seconds.”

The team split the main forecast model in two, so they could multi-task.  Iterating and testing numerous variables and transformations of inputs, they discovered three inputs that did the trick.  “We tested many iterations and tried adding more into the model,” said deGorter. “But parsimony is key – the rest was just noise. A lot of data for a little explanatory power doesn’t yield a powerful predictive model.”

This insight emerged around 3am, and the team immediately began building their presentation.  Whenever exhaustion appeared, it was countered by a tiny little cup of Portuguese espresso, as the team forged ahead.  They delivered the presentation to the judges just under the buzzer.

The Rooster

When it came time to present, deGorter and Bayne delivered the insights and early performance indicators that the team had identified.  Not only was this beyond the scope achieved by the other teams, it was exactly what Intel was looking for. In fact, the Intel team was so excited by the approach and the insights uncovered, that the full 8 minutes of questioning by the judges was spent at their request to share more findings.

Thrilled to be among only two teams in their tier to be chosen as finalists, the team worked on packaging and positioning the insights in the presentation.

When they climbed the stage to make their final presentation to the entire I-COM Summit audience, the team knew they had solid insights to offer Intel. In the end not only did they have very strong Model Fit, they also had the most accurate forecast of any team by a large margin.

By being adaptive, approaching the challenge with an amazing attitude and keeping the “so what” question for the business and real-life impacts in mind, the Analytic Partners team “Red is the New Hack” took home the top prize: the Golden Rooster. This combined with the strong showing in the Data Creativity Award competiton, Analytic Partners also walked away with an award for Smart Data Agency of the Year.

Caitlin Hill summarized best: “This experience made it clear to me that being multi-faceted, adaptive and cooperative is very effective for us.  The other teams all had very interesting approaches.  But I think this win is a potent validation of AP’s business model.  The way we do things at Analytic Partners makes all the difference.”

Yoga, Belvita and coffee helped too.


To learn more about…

The Hackathon Overview from I-COM Global

The Hackathon Video Facebook post from I-COM Global

Analytic Partners Named Smart Data Agency of the Year at I-COM Global Summit