Show simple item record

dc.contributor.authorD'Hooghe, Ruben
dc.contributor.authorOosterlinck, Thomas
dc.date.accessioned2021-04-27T19:02:00Z
dc.date.available2021-04-27T19:02:00Z
dc.date.issued2018
dc.identifier.urihttp://hdl.handle.net/20.500.12127/6836
dc.description.abstractProblem statement:Tomorrowland, globally known and often described as the most loved and best-known music festival on the planet. Each year, the festival hosts a total of 65.000 attendees per day. Starting from the edition in 2017, the organization created a two-weekend concept which sums up the attendees to a total of almost half a million. Coordinating such large numbers of people translates to the need of an operational and logistical wonder. This work focusses on the Food & Beverage department, a strong nine headed team that serves food and beverages to half a million people over a period of 10 days (start and end date of Dreamville included). In order to get a sense of the numbers behind this department, please find the infra listed overview of activities this team fulfils. 1. Coordinating logistical streams towards the more than 100 bar locations 2. Coordinating the crew that operates both bars and warehouses 3. Creating and implementing a total of four unique restaurant experiences on the Tomorrowland festival site (Aperto, B-Eat, Tomorrowland Restaurant & Brasa) 4. Producing a large number of organization operated food stands 5. Managing stock levels for both the Food and Drinks warehouse It is clear to see that structured process flows are needed in order to look over the operational and logistical challenges this team oversees. To get a hold of these challenges, a system was put in place during the edition of 2017. A full stock ordering and management platform was created and a dashboarding platform was added as an extra layer. This enabled the team to keep track of all stock movements as well as view current stock levels. This upgrade in operational excellence then led to the following challenge, that we tried tackling with this project. As there are huge people flows on the terrain, a lot of stock needs to be foreseen within the bars that deliver drinks to the public. Furthermore, it is a daring piece to estimate the stock levels that need to be present in the bars at what times and thus what the total stock to order should be. Apart from stock levels, a huge part in keeping this department operational during the festival has to do with the bar personnel. It is therefore of utmost important to get the people planning right in order to catch incoming waves of traffic at the bars. Last but not least, there is also the crisis management hat has to be addressed. It is a difficult task to keep track of all activities that are going on during such operation. If something goes wrong, it takes a lot of time to either see what exactly is off-track or to get notified on what went wrong exactly. This project aimed to tackle all stated challenges by creating and implementing a forecasting platform that attempts to predict sales volumes on site during operating hours. This will present a solution to the volume challenge as these volumes can be predicted long before the festival (in an offline system that does not take changing factors into account) as well as with more accuracy during the festival (in an online system that takes weather and other factors into account). The platform can also cope with crisis management as a deficit in sales can be seen every hour. Solution Our solution consists of two distinct parts: an offline prediction models that can make a prediction of the sales volumes before the festival starts, and an online prediction model that adjust these predictions throughout the festival by taking into account real-time sales data. The offline prediction was done by using a linear regression model. We started from transaction data from 2016 and 2017 and aggregated those transactions per product, per bar and per quarter. To those aggregated sales figures, we added additional data sources: - Bar data (bar types, bar sizes, bar locations, …) - Line-up data - Weather data - Product data This data sources were used to train our regression model to deduct relationships between those parameters and the sales volumes. This regression model was then applied to the new parameters of 2018 to make predictions of the sales volumes of 2018 The online model has three functions. - It gets the most recent weather forecasts and uses this to reevaluate the regression model - It rescales its prediction for a certain product at a certain bar by comparing the predicted sales vs the actual sales of that product at the bar since the start of the festival - It notices discrepancies between the forecast of the last quarter and the actual sales, and uses this information to adjust the prediction for the upcoming five quarters. Results The model is relatively accurate when it comes to predicting aggregate sales volumes of the products. The MAPE is around 12% for all drinks, while it reduces to 9% when taking into account only the most commonly sold products. Making predictions on a bar level is more difficult. There we see that we have a MAPE of around 50% during peak times, and an even higher MAPE during the first and last hours of the festival. We believe that this is due to the limited amount of data available to train the model and to detect quantify relations between parameters. However, this is not necessarily an issue, since the offline prediction will mainly be used to assess order quantities and to get an idea of the busiest periods at a certain bar. We are thus interested in the shape of sales curve, rather than the absolute volumes. Conclusion Our advice is to not base any important decision on this prediction tool yet during this year’s edition. It should be rather used as a proof-of-concept test case, and its performance should be thoroughly assessed after the festival. The interesting thing about this tool is that the more data that will be available, the more accurate the predictions will become. Therefore, the tool will become more useful throughout the years. Throughout this project, we identified a couple of issues in the data gathering process. Data wasnot stored in a consistent way. Besides, the exchange data and the access to data was not as straightforward as it could and should be. We therefore advice to invest into projects related to data streamlining in order to better allow similar project to this one.
dc.language.isoen
dc.titleData engineering in an entertainment driven business: Sales forecasting within a real-time platform
dc.source.numberofpages124
vlerick.knowledgedomainOperations & Supply Chain Management
vlerick.supervisorCardoen, Brecht
dc.identifier.vperid120992
vlerick.companynameWeAreOne.world
vlerick.companysupervisorZwart, Sjoerd
vlerick.programmeMGM Gent
vlerick.typebusresprojectIn-Company Project


This item appears in the following Collection(s)

Show simple item record