BigData Challenge | Efes

ROCADA team

Table of context

Issue description

01

02

03

04

Solution

Competitive advantages

Revenue model

05

06

Functional diagram

Challenges approach

07

Model result

08

Dashboard

2

Issue description

01

Issue description

Efes representatives are bleeding both time and money in a futile attempt to decipher store needs

​

Manually-cranking data analysis? It's a magnet for costly mistakes

Stores are constantly starved of the very products customers crave the most

Efes representatives are so buried in data, they're missing golden opportunities to bond with store owners

4

Solution

02

Solution

Imagine a future where your Efes representatives know exactly what they need, before even the retail shops do. That's the magic of our Predictive Order Intelligence

фото

6

Challenges approach

03

Challenges approach

Data wrangling challenge

• Python libraries were employed to aggregate and clean the data. Numpy, Matplotlib, Pandas,

​

• Normalization of the data was performed to ensure the model's accuracy

​

Usability challenge

• A user-friendly dashboard was tailored for Efes representatives. ReactJS for frontend and Kotlin (Spring framework) for backend

​

• The dashboard offers granularity, allowing representatives to select any date range, from a single day to an entire month

​

Algorithm challenge

• A Prophet model has been applied for precise predictions of store requirements

​

• The algorithm demonstrates rapid processing capabilities and has the potential for scalability, making it capable of forecasting years ahead with an increased data volume

​

8

Challenges approach

Data wrangling

9

• Merged sales with the table parent-child to get the parent

​

• We didn’t merge Material with the customer because it doesn’t have any relevant data

​

• We dropped the child because we needed to group by parent

​

• We grouped all the data by shop, parent, and date and aggregated the sales to see each day in total for parents

​

• Verified if no null values and got rid of duplicates

9

Challenges approach

Data wrangling challenge

• Python libraries were employed to aggregate and clean the data. Numpy, Matplotlib, Pandas

​

• Normalization of the data was performed to ensure the model's accuracy

​

Usability challenge

• A user-friendly dashboard was tailored for Efes representatives. ReactJS for frontend and Kotlin (Spring framework) for backend

​

• The dashboard offers granularity, allowing representatives to select any date range, from a single day to an entire month

​

Algorithm challenge

• A Prophet model has been applied for precise predictions of store requirements

​

• The algorithm demonstrates rapid processing capabilities and has the potential for scalability, making it capable of forecasting years ahead with an increased data volume

​

10

Challenges approach

Usability

11

• Made a user-friendly dashboard that shows graphs of forecasted values for the next period

​

• Search engine that allows you to find the required store by either city or store ID

​

• Streamlined final list of next day retail orders

11

Competitive advantages

04

Competitive advantages

Tool tailored precisely for Moldovan nuances that can be expanded to the Eastern European market and beyond

Moldova Focused

The solution adeptly recognizes and captures enduring trends and seasonality that could impact daily ordering patterns

The solution we offer is scalable, can handle large datasets and deliver real-time predictions

​

The solution enhances order precision, boosts customer satisfaction, and increases sales efficiency

The information and forecasts are presented in a user-friendly and straightforward manner for the Efes representatives

Accuracy

Trends considering

​

Usability

​

Value added

Accuracy

Trends

Trends considering

​

Usability

Usability

​

Trends considering

​

Scalability

13

01

05

03

02

04

Revenue model

05

Revenue model

Application subscription

Source code sale

15

• Use a monthly subscription model for immediate access to a user-friendly application

​

• Benefit from versatile search and filtering options to locate suitable stores and forecast consumption

​

• Access user-friendly dashboards featuring graphical representations and charts for easy data interpretation

• An algorithm enabling FMCG companies to seamlessly integrate the model into their existing IT infrastructure

​

• Source code prepared for easy modification and updates

​

• Comprehensive documentation and technical support to assist buyers in efficiently implementing and customizing the source code to meet their specific needs

15

Functional diagram

06

Functional diagram

17

Model results

07

• MAPE: 0.46
• Accuracy: 0.48

Dashboard

08

ROCADA team

Dominic Flocea

Data Scientist

Full-Stack Developer

Cristofor Fistic

Data Scientist

ML Engineer

Costel Voica

Product Marketing Developer

Nichita Catrecico

Product Manager

22

Contacts

You can find us here

Dominic Flocea: dominicflocea.dev@gmail.com

Cristofor Fistic: cristofor.fistic@ati.utm.md

Costel Voica: contact@costelvoica.com

Nichita Catrecico: t.me/katrechko_ni

Контакты