This beautiful future depends on data and AI

A Danish festival raises the bar by combining sales data, music schedules and weather to reduce wasted stock

By | 5 minute read | November 19, 2019

With its electro-light tulip garden, disco ball-adorned trees and no stone-left-unturned music lineup, “Denmark’s Most Beautiful Festival” aims to surpass guests’ expectations on safety, comfort and entertainment, from its uncannily clean bathrooms down to its whimsical camp-in-a-beer-can glamping options.

The Skanderborg Music Festival (aka “Smukfest”), located in the northern European country of Denmark, is no stranger to nature’s mayhem and its impact on tens of thousands of battle-hardened party warriors. Though it takes place during the second weekend of August in a bedazzled eco-village deep in a beechwood forest, the weather doesn’t always comply. After 2018’s sunbaked soiree, a veritable jester court of cloud-bursts in 2019 left lesser warriors stomping out early in their rain boots – leaving waiting oranges longing to fulfill their destiny as a vodka’s partner in crime.

Could those unpreventable forces be mitigated by data and AI? More importantly, could the festival squeeze more insights out of more data types to keep the party lit? The answer, increasingly, is yes – but so goes the journey to AI – ample data needs collecting, organizing and analyzing to get it ready for the epic voyage.

Get to know Smukfest, the world’s first live demo of the AI ladder

Festival fare isn’t designed to do much more than stave off a hangover. Foodies can rejoice in the likes of fresh oysters and champagne, tapas and sangria down to gourmet burgers and beers—all purchasable from the flick of a wristband at various pop up eateries, 65 bars and 600 points of sale.

The flip side of such abundance and variety is the very real potential of food waste. In 2019, nearly 40 percent of 325,000 unused items were returned after the festival, costing time and money carting leftovers away from the site.  But thanks to the festival’s long-standing relationship with IBM Nordics and a loyal crew of volunteers hungry to get their hands on the increasingly tasty data, Smukfest can now address this problem.

And luckily, the riches of various data collected and organized from all parts of the festival have grown over the past three years. This means IBM can put the real muscle behind it, punting over some advanced sales forecasting that can now provide organizers with the right insights to make evidence-driven-decisions around how much and when to stock the bar.

Not only is the data getting richer, but IBM data and AI solutions are raising the (no pun intended) bar. At the outset of 2019’s festival season, the IBM team was able to advance its estimates of expected sales of products in each bar through a sales dashboard created in IBM Cognos Analytics.

Now collected and organized, Smukfest data is available for use with any application available on Cloud Pak for Data, from the machine learning capabilities inside Watson Studio to model testing, bias detection and outcome explainability in Watson OpenScale.

The team can take that sales data, port it into Watson Studio as an add-on within Cloud Pak for Data, and build models to look at correlations between say, a bar’s location, the weather, perhaps also the musical performance – then and test for model drift with Watson OpenScale.

Behind this AI: Gaining crystal clarity on more complex data

Are Smukfest’s bar sales correlated to the music schedule or even genre? Does the bar location have any impact on sales? Does the performer? Do those who purchase Tanqueray gin stick to a certain brand of tonic water? And most importantly, how might the weather impact sales of Smukfest’s beloved vodka and orange juice?

To start surfacing these queries, IBM volunteers grouped data by data and time, then unified genre data to understand correlation between genre and sales, then grouped data by venue to get the number of concerts by venue, then by time stamp to get number of concerts by time range. Finally, they joined sales line data and sales transactions data and group data by product category.

Then, by joining to music data, they could conclude how much money bars made during specific concerts. Then they could obtain a bar’s location, and merge sales data and bar location data to create visualizations using dashboards.

To forecast 2019 product sales, the IBM team had to use eight different data sources and conducted 34 data preparation steps to transform raw and granular 2018 transactional data into clean and aggregated dataset. These recorded sales of products for each bar on both hourly and daily level and merged that data with weather data and the music schedule.

But while doing data prep using Watson Studio, the team discovered data gaps due to transactional system outages and a change in product naming conventions to name a few—all of which had an impact to ability to produce forecasts.

The team had its hopes set on building a time series model to forecast sales. But since only one complete sales cycle was available from 2018 to train the model – machine learning algorithms require three years – the team had to come up with an alternative approach to forecasting that would allow the festival to make the most out of their existing data.

The solution? Provide as much relevant information as possible to Smukfest bar managers to augment their decision-making process.What resulted was a forecasting model consisting of an SPSS Flow providing a “forecasting base” which was cleaned and aggregated with sales data for each product and each bar, then Jupyter Notebooks to provide weather forecasts for the next 48 hours.

With the model now including weather data for the previous year, the weather forecast, plus the data on sales during the same hour and day the year before, IBM can help the festival reduce large errors that lead to either overstocking and shortages of certain products—and place limits on how much the bar managers could order for the duration of the festival.

For example, at a future festival, she might very well have the ability to ask a Watson Assistant-created chatbot how many oranges they’ll need to keep the crowds happy – and she’ll be able to confidentially advise them to stock up for the sunny day ahead, or if it’s colder and rainier than last year, avoid making an inventory run.

Explore the Journey to AI, a prescriptive approach that helps you modernize, collect, organize, analyze and infuse all your data. Find out where you are in your journey to AI.

Take the assessment.