Charlene Grajales, Dina Derla and Maria Regina Corazon Sibal of Team IntensiPy took on the challenge of identifying sales drivers from an appliance distributing and sales company. The challenge was to look into a year’s data with 347 thousand sales records and 14 features, as they tried to solve, “What are the sales drivers for an appliance? How do these drivers affect sales?”
Weekly company sales data as well as client validate competitor data was explored for data visualization, analytics and model development. The team was able to produce an interactive dashboard highlighting major sales drivers for the company, share kite key insights and observations from the data and looked into possible external factors that can validate the results. They went further to create a prototype sales predictor web application that can generate weekly projections for sales for user selected features.
3. DATA
CHALLENGE
What are the sales drivers for an appliance?
How do these drivers affect sales?
WHAT ARE THE MAJOR
DRIVERS THAT AFFECT
APPLIANCE UNIT SALES?
5. 1 YEAR's worth of data
821k estimated appliances
sold
DATA DIMENSIONS EXCLUDING COMPETITOR* DATA
6. Prepare dashboards for visualization
Analyze trends
Predict sales
From weekly sales data of one year:DATA
SCIENCE
MISSION
7. 2
Data Exploration
3
Data Cleaning and
Validation
4
Data Analysis
5
ML Model
Development
6
Hyperparameter
Tuning
7
Web Application
Design
8
Web Application
Deployment
COMPANY SALES PREDICTOR MODEL DEVELOPMENT
Machine Learning Model Training and Web Application Deployment
1
Business
Understanding
10. 24k
MEDIAN
NUMBER OF
UNITS SOLD IN
THE PEAK
SEASON 1
5
9
13
17
21
25
29
33
37
41
45
49
40,000
30,000
20,000
10,000
0
COMPANY UNITS SOLD
11. DESIGN 1 DESIGN 2 DESIGN 3
600,000
400,000
200,000
0
65%
COMPANY
SELLS MORE
OF DESIGN
TYPE 1
COMPANY UNITS SOLD
DESIGN 1 DESIGN 2 DESIGN 3
600,000
400,000
200,000
0
12. 65%
DESIGN 1 DESIGN 2 DESIGN 3
500,000
400,000
300,000
200,000
100,000
0
DESIGN 1 DESIGN 2 DESIGN 3
500,000
400,000
300,000
200,000
100,000
0
COMPETITOR UNITS SOLD
COMPETITORS
SELL MORE OF
DESIGN TYPE 2
13. 0 100,000 200,000 300,000
DEALER 1
DEALER 2
DEALER 3
DEALER 4
DEALER 5
DEALER 6
DEALER 7
DEALER 8
DEALER 9
DEALER 10
31%
DEALER 1 SELLS
THE MOST
NUMBER UNITS
COMPARED TO
OTHER DEALERS
(NATIONAL,
LOCAL,
REGIONAL)
COMPANY UNITS SOLD
0 100,000 200,000 300,000
DEALER 1
DEALER 2
DEALER 3
DEALER 4
DEALER 5
DEALER 6
DEALER 7
DEALER 8
DEALER 9
DEALER 10
15. 27%
COMPETITOR UNITS SOLD
COMPETITORS
HAVE A
STRONG
DISTRIBUTION
IN VISAYAS
M
odern
Trade
N
C
L
G
M
A
/SL
V
IS
M
IN
250,000
200,000
150,000
100,000
50,000
0
M
odern
Trade
N
C
L
G
M
A
/SL
V
IS
M
IN
250,000
200,000
150,000
100,000
50,000
0
21. 2
Data Exploration
3
Data Cleaning and
Validation
4
Data Analysis
5
ML Model
Development
6
Hyperparameter
Tuning
7
Web Application
Design
8
Web Application
Deployment
COMPANY SALES PREDICTOR MODEL DEVELOPMENT
Machine Learning Model Training and Web Application Deployment
1
Business
Understanding
22. AUTO ML
Random forest Regressor*
- more than 400 models tested
*with lowest RMSE (Most Accurate)
Techniques ExploredML Model
Development
23. WEB APPLICATION
SALES PREDICTOR
Machine learning backend
Estimate the number of units
sold based on sales drivers
Interactive: Easy for business to
validate
Portable: Can be deployed on
any desktop
Locally-hosted for security
PREDICTION IS ACCURATE WITH
1.45
UNITS
Margin of Error
THE PREDICTOR IS
VIABLE...
TO SOME EXTENT
5.6% PERCENTAGEE
ERROR
25. NEXT STEPS
DATA DRIVES MORE ACCURATE
RESULTS
one or two more years sales reports
Profitability per unit sales
price/ price range of sold models
purpose (for home, for business)
customer data (sex, age range, income) and
behavior (recommended by a friend,
recommended by salesperson, quality, price, etc.)
More accurate competitor data
Collect more data:
26. NEXT STEPS
DATA SCIENCE
Update the data weekly
Use of dashboards for
visualization of sales
Loading more data for a
more reliable predictive
model