�ݺ�ߣ

© 2015 GrayMatter All Rights Reserved
Advanced Analytics for
Airports Commercial Function
from
GrayMatter

© 2015 GrayMatter All Rights Reserved 2
• Business Challenges
• Problem Statement
• GrayMatter’s Airport Analytics (AA+) Solution Explained
 Retail Revenue
 Car Parking Revenue
• Data Modelling Process
• GrayMatter Centre of Excellence for Data Science
Agenda

Business Challenges

Decreasing
aeronautical
revenues
Pressure on
Profitability
Need for increase
in non-aeronautical
revenues
Pressure on Profitability

Problem Statement

How to maximize Non-Aero Revenue?
Ads &
Others
Parking
Retail
(Revenue
Comm. &
Rentals)
 How to increase per customer
retail revenue?
 How to determine optimal
rentals that can be obtained
from stores?
 How to maximize car parking
utilization?
Key Sources of Non-Aero Revenue

GrayMatter’s Airport Analytics Solution (AA+) Explained

Retail Revenue

Key Business Drivers
Passenger Profile:
• Income
• Nationality
• Purpose of flight
• Demographics
Commercial Offer:
• Shop Profiles
• Variety
• Price
Architecture:
• Layout
• Store Location
• Traffic flow
Passenger
Spending Rate
Passenger
Footfall
Passenger
Conversion Rate
Revenue

• Retail surface density to maximize performance
 Store right sizing
 Store location
• Increase footfall and time spent in shopping
 Aligning stores with the gate
 Gate reallocation to direct traffic flow
• Increase conversions
 Product mix and display adapted to passenger profiles and
liking
Airport Retail Revenue Optimization Approaches

• Store location is one of the most important factor
influencing performance of the store
• Depending on the class of store and segment of
customers it attracts, it is important to locate the
store at the right traffic flow
• Trend and correlation analysis of store size and
revenue per departing passenger allows resizing
store or commercial contracts
Store Size and Location

• Another important factor to influence revenue is to
make the stores attractive to the customers
• Understanding customers and their likes enables
 Correct store mix at the airport
 Product mix and display in the store
• This enables to understand whether customer is
price sensitive, their buying preferences, based on
their demographics, purchase history
Customer Segment Profiling

• Airports have unique advantage of controlling the
footfall
• Based on the profile of passengers, reallocation of
gates can result in increased footfall to a store, which
are likely to convert
• Correlating profiles of passengers (both departing
and arriving) with land side store performance can
help in making land side stores more attractive for
passengers
Increasing footfall

Detailed Use Case: Store Size and Location

Use Case: Store Size and Location - Attributes
• Store Attributes
– Store Name
– Store Classification
• Fast Food, Beverages, F&B, Fashion, Large Format,…..
– Store Size / Area
– Location
• Closest Gate, Security, Air-Side, Land-Side
– Days in Operation
• Departure Gate Attributes
– Gate
– Flight Details
• Airline, Destination, #Passengers (Total Aircraft Capacity),
Allocated Date Time, Departure Date Time (Scheduled and
Revised), Passenger break by nationality would be ideal
– Neighboring gates

• Arrival Flight Attributes
– Arrived from, Airline, #Passengers (total capacity), Arrival date and
time, Late minutes
• Product Attributes
– Purchase Date/Time
– Store
– Product (and Product Classification)
– Price / Qty.
– Nationality (DFS if available)
• Performance Measures
– Revenue
– #Transactions
Use Case: Store Size and Location - Attributes

• Analysis
– Study correlation between various attributes with Revenue and
#Transactions
– Correlation between total passengers near store and revenue and
#Transactions
• Study map of departing passenger traffic with heat map of store sales
• Total arriving/departing passengers and land side store performance
– Store Model
• Additional revenue with change in size and/or location
• Recommendation
– Under performing Stores
• Actions required to improve their performance
– Terminate, Resize, Relocate
– Other Stores which will benefit resizing and relocation
Use Case: Store Size and Location

Car Parking Revenue

• Industry Analysis
 Industry growth/de-growth (airport passenger volumes)
 Competitor pricing
• Demand Management
 Fluctuating demand with seasonality, unplanned events further skewing
the predictability
 Optimal pricing to reduce vacancy and maximize per slot revenue
• Customer Segmentation
 Identification of different customer segments and adopting differential
pricing
• Hedging Risks
 Mitigating revenue losses due to “force majeure” incidents
 Promotional schemes to ensure committed revenues in advance
Key Business Drivers

• Airport Analytics (AA+) CPRM Implements a demand driven dynamic
pricing system for all the car parks in Airport
 Car park prices will be updated every day for next X Days in advance
• Dynamic pricing will result in
 Increase in the car parking revenue
 Increase in occupancy levels in all the car parks managed by Airport
• Pricing decisions should adhere to business rules and constraints
• System should monitor demand and decision accuracies and recalibrate
when demand predications deviate significantly from actuals
 Recalibration can be a mix of automatic and manual tuning of models and
algorithms
• User interface to control and override recommendation as needed;
ability to do what-if analysis
Solution Outline

• Historical data processing
 Uses past 2+ years of historical transactions and reservations
 Data cleaning – outlier detection, identifying constrained periods & special
events
 Customize demand prediction model according to specific business needs
• Decision Optimization
 Factors affecting demand analysed
 Historical data as well as recent trends included in analysis
 Demand models generated for identified demand groups
 Using business constraints, predictions generates pricing decisions
• Monitoring Process
 Monitors patterns and predictions in a predefined frequency
 Significant deviations results in alerts and re-modelling activities
3 Staged Solution Approach

• Data cleaning and imputation of missing data
• Identify outliers and associate reasons with outliers
 Low occupancy with maintenance of car park
 Low occupancy with airport closure due to weather
 High occupancy with holidays
• Extract of useful insights/ patterns from the Car Park data, like
 Special Events, Price Elasticity, Seasonality, Cancellation rates
Historical Data Analysis

• Car park demand is a function of several attributes like,
– Seasonality (Week of Year (WOY), Day of Week (DOW), Special Events- Holidays, Extended Weekends,
Conference, ...
– Length Of Stay (LOS)
– Type of carpark, Price
– PAX volume , Competition
• Using statistical testing techniques, extent of the impact of each of the attributes on the
demand is inferred
– Time Series
– Pattern based
– Regression
• Historical price changes are used to extract price elasticity and cross elasticity functions
• Recent trends in transactions and bookings factored to include recency effect and enable
daily optimization
• In order to reduce uncertainty in the demand predictions, demand groups having similar
demand functions are made
– Clustering algorithms are used to find these demand groups
• Demand models are build for each demand group
Decision Optimization

• Dynamic pricing comes in two scenarios
– Demand exceeds the carpark capacity
• Optimizer should not allow lower rates/discounts to be offered
– Demands are very low compared to capacity
• Optimizer should attract more demand by reducing the rates/offering more
discount
• Optimizer will scan price and demand space and select a price for
each carpark to achieve the objectives within the business
constraints
Decision Optimization

• Quality of decisions depends on the forecasts which in turn is a
function of various patterns and models
• System will monitor Occupancy Levels, Revenue per slot and
predictions v/s actual arrivals by LOS on a weekly basis
• Alerts are raised when monitored metrics crosses a threshold
• Pattern monitoring and model re-validations are carried out once
a quarter
Monitoring

DATA MODELLING PROCESS

Typical Approach
Data Preparations
and understand
Business process
Data
Analysis
R
E
P
O
R
T
S
Predictive
Models
Prediction
Validation
Predict /
Optimize/
Simulate/
What-if
Data
Cleaning
Patterns
and
Clusters
Descriptive Predictive Decisions
Historical Data
Current Data
Granularity Decisions
Outliers Missing Integrity

GrayMatter Center of Excellence
for
Decision Science

Data Science – Service Offerings
• Horizontal
• Vertical
• Retail & CPG
• Airport
• Insurance
• BPO
• Data Mining Process
• Tools
• R, Weka
• KXEN, SAP PA
• Big Data
• Predictions
• Decisions
• Reports
• Discovery Workshop
• Data Models
• Data Cleaning and
analysis
• Predictive and
decision Modelling
• Team Handholding
Consulting
Analytics
As A
Service
Solutions
Trainings

Vertical Solutions
•Workforce planning
•Agent’s Productivity
•Recruitment
•Persistency / Lapse
•Agent performance
•Fraudulent Claims
•Optimum Store size and
location
•Customer Segment
Profiling
•Gate Allocation
•Car Parking Revenue
Optimization
•Price Elasticity
•Price Optimization
•Allocation and Replenishment
Retail and
CPG
Airport
BPO
Insurance

Horizontal Solutions
• Customer/Product Segmentation
• Promotion Effectiveness
• Social Media Analysis
Marketing
• Sales forecasts
Sales
• Personalized content
• Social Network Analysis
Digital

• Depth in Technology and Data Science capabilities
– Dedicated team of data scientists
– Experienced in Machine Learning
• Data Cleaning, Outlier Detection, Statistical Testing
• Classification, Regression, Attribute Selection, Segmentation/Clustering, Forecasting,
Association Rule Mining
• Pattern Based Modelling
Team Capability
• Deployed over 30+ predictive and decision
models across domains
– Planning (Forecasting)
– Price Optimization (Demand Prediction)
– Allocation and Replenishment (Demand Prediction)
– Churn and Persistency (Likelihood Prediction)
• Strategic Partnership with SAP and Revolution Analytics

Technologies

Case Study
Store Performance at Airports
Business Problem
Airports are interested in getting a good estimate of revenues in next 12-18 months from various stores. This helps
them in their yearly financial planning (a) Identify potential non-performing stores to take proactive actions (b)
Consider prospect of new stores
Solution Approach
A custom growth-seasonality model was build using
weekly sales by store
Company and Store level special events are identified
using outlier detection methods. A separate special
event model is developed and overlaid on the growth-
seasonality model.
Stores with similar seasonality were grouped together
and their seasonal patters were extracted. Similarly
Store with similar growth/decay trends were grouped
together and combined parameters were extracted.
Another model to detect recent inflexion points was
build and data after inflexion point was used for
growth/decay trends.
Solution is deployed as a web application integrated
with the clients planning system.
Benefits
Automated , integrated financial projection system with
what-if analysis. Since it uses both long term and short
term effects, accuracy of predictions are better.
Accuracy Achieved: Weekly incremental: 10% , Cumulative
12 months: 2% - 5%
Inflection Point

Case Study
Demand Forecasting for Retailers
Business Problem
Predicting accurate demand for each item in a store is a challenge for all the retailers for three reasons (1) Large
number of item and Store combinations (2) Low volumes of sales at item/store level (like fashion apparels) and (3)
Short shelf life (perishable items). An automated and reliable demand predictions are needed to assist in planning
and operational decisions.
Solution Approach
Used recent 3 years of historical data (1) POS
Transactions – sales, returns, audit, (2) Price changes, Inventory
movements & PO and (3) Product and Store masters
Developed a demand model as a collection of several
patterns – Store & Product Seasonality, Impact of time
on Sales & Sell-Thru, Effect of Special Events, Impact
of inventory and price elasticity.
These patterns are extracted using segmentation &
clustering, regression, outlier detection and time
series algorithms.
Demand is computed as a mathematical combination
of these patterns with uncertainties.
This model has been deployed in several retail
solutions like Price optimization, Allocation, Order and
Replenishment
Accuracies
Company level weekly item forecast ranges between 10%
to 15% while season level accuracies are in 5% - 8% range

Thank You

�ݺ�ߣ

advancedanalyticsforairportscommercialfunction-.pdf

More Related Content

advancedanalyticsforairportscommercialfunction-.pdf