yield management as a process governed by data mining in the auto industry

26
Analytics, Big Data and the Cloud Edmonton, April 23, 2012

Upload: lara-solara

Post on 22-Mar-2016

214 views

Category:

Documents


1 download

DESCRIPTION

by Ayman Ammoura

TRANSCRIPT

Analytics, Big Data and the Cloud

Edmonton, April 23, 2012

» Introducing main concepts

» Applying our science and technology to a Canadian small business

» Mining on The Revenue Side - Rates

» Mining on The Expense Side – Insurance

» Sharing success stories

» Yield management is the process of understanding, anticipating and influencingconsumer behavior in order to maximize yield or profits (Wikipedia)

» Understanding Observation and analysis

» Anticipate Forecasting

» Influencing Management actions

» Data Mining is a step in the knowledge discovery process. (Osmar Z.)

» Data mining is a process of extracting previously unknown, valid, and actionable information from large databases then using the information to make crucial business decisions (Cabena, et al, 1998)

» Data repository built to facilitate OLAP (OnLineAnalytic Processing) not OLTP (Transaction).

» Warehouse Multidimensional, Subject-Oriented, data model Data Cube

» To support OLAP, a data warehouse is often implemented as a hierarchical N-Dimensional data cube.

Rental Days

Location

Time

Vehicle Class

Usually you need SIC, Source, Sold Extras .. N-Dimesions

Fact Table

Dimension TableTime

Class

Location

Each slice it an n x m 2D Table

» There are 2 items that define the financial well being of an organization.

» Revenue (our example Rental Days)

» Expense (our example Insurance)

» In our case, we need to create a data repository with Fact tables “Rental Days” and “Insured Units”

This fires @ 4:00 AM Everyday

Daily @ 0600

Canada Winter Games

» How and when to adjust.

» Utilization Based rate adjustment˃ Not Competitive

˃ Big missed opportunities (explained next)

» To answer the When question we needed to get more insight into the data

» Understanding the Cycle

City Sold-out

» Create a system that would issue new booking rates based on utilization.˃ 0%- 50% +0%˃ 51% - 65% + 10%˃ 66% - 75% + 15 % etc …

» This will be transparent to the agent and has been widely used for over a decade.

Build Availability

Cube

Every 10 Minutes

Branch Rates

Publish Intranet

Walk-in Rates

System Wide

» Using this model, we were able to increase revenue by 30% in the first cycle (May-September)

» During busy season, booking are received 90 days in advance» Shoulder Season as low as 6 days average

90 days

Sold Out

» Using the utilization tiered rate adjustment process alone 50% of the business can be improved by at lease 20% Because 50% booking is required to achieve the next tier

» On Average, most bookings during busy cycle were entered 3 months in advance

Build Availability

Cube

Every 10 Minutes

Branch Rates

Publish Intranet

Walk-in Rates

System Wide

Insert Cyclical Adjustments

Known Dates

Up $1.3Million

Utilization based Tiers

Up $2.2Million

Utilization + Cyclical and Localized Adjustments

» Phase I and Phase II were constructed one cycle apart

» Complete project spanned 14 months

» So far we talked about an example of how we applied simple Data Mining tools to achieve great results on the revenue side, helping a small business.

» Next we will examine how we have effectively used analytics to impact profitability by reducing a major expense.

» Next to depreciation, this is usually the second biggest expense in the auto industry.

» Existing Scenario is that the business had to pay the insurance premium per unit ($m) on all used units in a calendar month.

» Existing solution was: Identify units that were rented (n), and pay monthly ($mxn)

» How to reduce this cost?

Visualization of the number of active days of every insured unit for a typical month

» Examining the number of insured units against the number of units on rent

Insured Vehicles

Rented Vehicles

Co

un

t

» As there are more units in the fleet than was required, the company insured way more than was required Information that was implicit data

» Time to renegotiate the insurance model! –Preferably without sharing your results with the broker

Insurance cost decreased by $120,000 per year

» Instead of paying on all units, we negotiated a policy that allows us to pay higher prorated premiums but on a daily basis.

» Without the ability to transform the data into information, this effort was “unnecessary” and probably have not happened!

» Recall our definition (Data mining is a process of extracting previously unknown, valid, and actionable information)

» Love to answer any questions ….