operations research at continental airlines judy pastor, senior manager, or continental airlines...

33
Operations Research at Continental Airlines Judy Pastor, Senior Manager, OR Continental Airlines INFORMS - Houston November 16, 2000

Upload: jemimah-watkins

Post on 24-Dec-2015

219 views

Category:

Documents


0 download

TRANSCRIPT

Operations Research at Continental Airlines

Judy Pastor, Senior Manager, OR

Continental Airlines

INFORMS - Houston

November 16, 2000

Continental Airlines OR History

• No centralized OR Group

• OR”-ish” functions throughout company

• Many models purchased - some black box - most unrelated to each other

• New Management, 1994-95

• Analytical skills sought

Continental OR Group

• Late 1994, Revenue Management saw a need for a simulator and Operations Research group to use it

• Rev Mgt simulator specified by Continental and built by Aeronomics (Talus) early 1995

• OR Manager and Analyst hired April, 1995

• First mission of OR group to use simulator to experiment with OD heuristics

Airline OR

• Many departments in an airline can benefit from OR expertise

• Historically, airlines have been some of the largest users of OR models

• Five major users (others exist)– Planning (Long and Medium Term)– Scheduling (Short Term Planning)– Pricing– Revenue Management– Operations

Planning

• Fleet Planning• Market Planning• Real Estate (gates, terminals)• Finance• All areas need

– forecasting– optimization– network analysis

Scheduling

• Schedule Generation– uses forecasting, discrete choice (logit)

modeling, simulation, some optimization• Fleet Assignment

– uses forecasting, integer optimization• Crew Scheduling

– integer optimization (column generation)• Airport Services Scheduling

– integer optimization, simulation

Pricing

• Usually broken up into Domestic and International components

• Much less scientific than other areas• Deregulation in US has led to two actions

– lead a sale or a price increase– match the competition

• 30,000 origin-destination pairs to price domestically

• Fares filed twice a day

Revenue Management

• Most “micro” of all areas• Decides the number of “discount” seats to

sell• Saves seats for last-minute, high yielding

passengers• Determines how much to overbook a full

flight• Chooses whether to sell to one connecting

passenger or to two locals (OD problem)

Revenue Management (RM)

• Uses many OR techniques– forecasting– optimization (deterministic and stochastic)– simulation

• Constraints to RM problem received from other departments– Prices from Pricing– Number of Seats from Scheduling– Available Connections from Scheduling

Revenue Management (RM)

• Airlines sell their inventory (seats) in a variety of ways– Airline reservations service– Travel Agents– Consolidators/”Bucket Shops”– Internet (own website/Travelocity/Priceline)

• Seat sales controlled by Computerized Reservations Systems (CRS)– antiquated, developed before hub and spoke system– RM must work in this environment

Back to 1995 - Initial OR Group

• Both Manager and Analyst from outside company

• Manager, Judy Pastor– BS, Computer Science, w/ 9 yrs in oil industry– MS, Operations Research, w/ 2 years in

transportation OR at UPS and 4 years in LP modeling for oil refineries

– Was interested in returning to transportation applications

Continental OR Group

• Analyst– Several years of programming in oil industry– Returned to school for MS, Mathematical

Sciences, Rice University– First Operations Research position.

Initial Conditions

• No formal job descriptions

• Mission was to understand RM practices and look for process improvements

• Limited analytical software

– no LP optimizer or modeling language

– no statistical package

– no generalized simulation package.

– Lotus 123, student version of LINDO, and a C compiler

Initial Conditions (continued)

• OR Group under Revenue Management

• At same time, RM was in transition– installation of latest version of PROS (Pax

Revenue Optimization System) as DSS– change from market analysts with reservations

and/or operations backgrounds to MBAs from highly quantitative and analytical programs

– formal training program was in its infancy

First Projects

• Initial projects centered around using RM simulator to examine OD strategies

• RM simulator was good example of a black box model– output showed a total revenue and total pax boarded but

gave no clue as to what caused changes from run to run.– a “pretty” user interface was being built for Phase II but was

useless since this information was still not available– we wrote C programs to read “debugging” output to create

multiple run comparison reports showing differences in pax acceptance with parameter changes, booking curves, EMSR curves

– answering question of “WHY??” because all important

First Projects (continued)

• To aid in defining OR projects, the “Questions Group” was formed– met monthly to identify things we would like to

know but do not currently such as• what drives denied boarding costs?

• what is the true cost of a posted (full) flight?

• how much does poor forecasting hurt us?

• lots of ideas and which to tackle first?

Getting Started

• Much latitude was given to define work processes and projects

• Assignments were unstructured and exploratory. Do whatever needs to be done

• Sad fact of matter about OR jobs - the data is never the way you want it, needs to be cleaned up, etc. “Grunt” work often required, especially in the absence of tools

• Modeling not always major part of job• Entry level OR Analysts can be disappointed with this

OR Analysts

• should be able to accept frustration

• require extreme flexibility and ability to change in mid-stream

• able to find the value in coming up against a dead-end (sometimes an opening comes up later)

• must never be satisfied until they understand why a system works as it does

Home Alone

• In the first two years, two OR analysts came and went

• Ads in the paper generated hundreds of resumes, most responding to either “Operations” or “Research” but not “Operations Research”

• Time alone gave me the time to acquire software, build usable tools, document processes, learn more about Continental and the airline industry, etc.

• Created a vision for the group of one that could provide OR techniques to RM and other parts of company, tying together the many black boxes for a common purpose

Building the Group

• Was able to build group with a variety of people– from inside and outside the company– with strong analytical skills– with strong communication skills

• Strong communication skills very important to get the message out

OR Responsibilities

• Understand all vendor supplied systems in place now– forecasting in PROS

• seasonality calculations

• clustering of market groups

• Learn about new features and be able to explain them. Make recommendations as to their use.

OR Responsibilities

• Develop new techniques/systems to improve processes and enhance revenue– used statistical analysis to aid in identifying

causes of frequently late arriving flights– participated in design and delivery of

Enterprise Wide Data Warehouse• eliminating the data “silos” built by each department

OR Responsibilities

• Stay abreast of latest techniques/research done in Operations Research, especially in area of Revenue Management– participation in INFORMS– membership in AGIFORS (Airline Group of

the International Forum of OR Scientists)– software user groups

Current Major Projects

• Demand Driven Dispatch– “Flagship” Project of OR– Combines aspects of Scheduling (Fleet

Assignment) and RM– Algorithm and System developed by OR

• Overbooking Improvement using DW data

• O and D Forecasting

Demand Driven Dispatch

• Aircraft types (and caps) are assigned to routes by a Fleet Assignment Model (FAM) – Input to FAM is based on an estimate of average

demand– Objective is to maximize revenue, minimize

costs, and normalize operations– A consistent fleet assignment throughout the

week to the same flight M-F is seen as advantageous to the operation

Demand Driven Dispatch

• RM knows that demand varies from one DOW (day of week) to another

• Fleet assignment is “pretty” optimal overall, but suboptimal on a flight by flight basis

• D3 (Demand Driven Dispatch) group – examines the schedule– finds sets of flights that are easy to swap– queries the RM forecasts for those flights– prescribes swaps to maximize revenue

Overbooking Improvement

• Determined by “no show” factor– normal no show is 10-15%– Latin flights can have up to 50% no show

• Empty seats are perishable inventory– after plane takes off, those selling opportunities

are gone

• Can more detailed data about pax help?– Data Mining techniques, forecasting, DSS

O and D Forecasting

• CRSs are “leg based”– connecting pax book onto two legs– revenue of 1 cnx pax < 2 local pax (in general)

• 2200 flights a day/10 fare classes = 22,000 leg based forecasts per day * 330 days in a booking cycle = 660,000 for departure day

• 30,000 OD itineraries * 3 paths/itinerary * 10 fare classes * 330 = 297,000,000

O and D Forecasting

• We currently do leg forecasting and optimization with an O and D heuristic to handle connecting itinerary requests

• Theoretically, a network based solution would give us substantially better revenue

• But, network solutions are based on many forecasts all with some type of error associated with them

O and D Forecasting

• Other challenges– using O and D optimization within constraints

of leg based CRS– small numbers problem– constantly changing

network/schedule/environment– is there a compromise that can get us most of

the way to the “optimal”?

Other Issues

• OR must assist in the management decision – develop the DSS in-house or– purchase from vendor?– how will it be integrated into business process?– how will the technology be transferred?

• Currently OR is under RM, but integrated departmental solutions are the holy grail

• Career path for OR Professionals

Who You Gonna Call?

• “Ghost Busters!” - Operations Research group has the skills to understand the ramifications of different optimizations and build bridges between them, if possible.

Ultimate Goal

• We want Continental Airlines to have

THE BEST LITTLE OR HOUSE

IN TEXAS !!