This document presents a framework for implementing IoT data analytics in retail to understand the relationship between environmental conditions and customer behavior. The framework was tested in a case study of a large retail chain using sensor data on lighting, temperature, humidity and air quality correlated with sales performance metrics. The analysis found air quality and humidity significantly impacted sales while temperature had a non-linear effect. An IoT platform was developed to enact decisions from the analytical models by dynamically adjusting environmental conditions and integrating real-time point-of-sale data.

1. 1
J. GRABIS , K. JEGOROVA, K. PINKA
IoT DATA ANALYTICS IN
RETAIL: FRAMEWORK
AND IMPLEMENTATION
Institute of Information Technology, Riga Technical University, Latvia

2. 22
 Customer experience has many dimensions
– Sensorial, affective, physical, social and cognitive
 Modern computing provide opportunities for measuring
and improving customer experience
– IoT
– Data analytics
 Limited understanding of relations between customer
experience and environmental conditions
 Deployment of IoT devices and supporting data
analytical solutions requires sophisticated
technological platforms
– Continuous operations and enactment of analytical results
Background

3. 33
 To empirically test relations between
environmental conditions in a retail store
and customer behaviour
 To outline a technological solution for
deploying IoT data analytics
Objective

4. 44
Customer Experience
Dimenssions

5. 55
Sales & IoT
A0
SalesSensor
measurements
BMS
IoT
platform
Sales
performance

6. 66
 Data are provided by a large retail chain with more than 2000 stores
and 30.000 employees
– 60 000 purchase lines or registered transactions
– Over 150 000 measurements are available for each sensor
 The company aims to understand the effect of lighting conditions,
temperature and humidity on the customer basket size
 All customer transactions are recorded and the following sales
performance measurements are considered in this investigation:
– Number of items (N) – number of different products purchased by a
customer in one store visit (i.e., number of items in shopping basket);
– Weight of purchases (W) – weight of all products purchased by a
customer in one store visit;
– Quantity of items (Q) – quantity of items all products (summed across all
types of products) purchased by a customer in one store visit.
Case Study

7. 77
Impact of the Hour of the Day
Sales
by hour
Air
quality
by hour
Lighting
by hour

8. 88
Sensor DF Sum
Sq
Mean
Sq
F value P
Air 1 60 60.1 59.1 0.000
Light 1 4 4.2 4.1 0.042
Humidity 1 85 84.8 83.4 0.000
Temp 1 0 0.3 0.3 0.583
Hour 1 435 434.9 427.9 0.000
Residuals 7182 7299 1 1
ANOVA Analysis
𝑁𝑠𝑗
∗
= 𝜇 + 𝑎𝑖𝑟 + 𝑙𝑖𝑔ℎ𝑡 + ℎ𝑢𝑚𝑚𝑖𝑑𝑖𝑡𝑦 + 𝑡𝑒𝑚𝑝 + 𝜀 𝑠𝑗

9. 99
Average number of items N
according to the quintile of
sensory measurements

10. 1010
If the air quality
deteriorates
beyond the lower
boundary of the
air quality 5th
quintile
Then improve by
powering AC
To avoid
decreasing sales
Implementation
Stream processing (SP)
Evaluation of environmental conditions (EEC)
Persistent
storage
Building management system (BMS)
Other
data
sources
K1 KM...
P1 PL...
Adaptation engine (AE)
R1 RN...
Archiving
jobs
Evaluation
jobs
Triggering
jobs
8
2
1
3,5
4 6 7
910
IoT
devices
POS data

11. 1111
Air quality changes and the
number of items according to
time
0
20
40
60
80
100
120
140
0
50
100
150
200
250
300
350
400
7500 7550 7600 7650 7700 7750
Numberofitems
airquality
Time,minutes
air (MA)
air treshold
Items

12. 1212
 Sales performance is significantly affected by
the air quality and humidity
 Temperature has a non-linear impact on the
customer behaviour
 IoT platform supports enactment of the
decisions, dynamic adjustment of data
analytical models as well as integration of
real-time point-of-sales data for dynamic
pricing and personalized recommendations
Conclusion

13. 13
grabis@rtu.lv
http://iti.rtu.lv/vitk/lv/katedra/darbinieki/janis-grabis
Thank you!