QLLM For Airlines

TensorQ Air Powering New Generation RM For Airlines

Demand Forecasting Is Underestimating Demand – Not Quantified
Dynamic Product / Pricing – Not Personalized
Local Optimal Revenue – Not Quantified
Every Function Has Own Model For Generating Demand / Revenue
Revenue Effectiveness Is Unknown – Not Quantified
AI Agents Are Not Accurate …

With These Gaps Its Demonstrable That about 10-15% of Revenue is Left on the Table. Uncollected

Main Themes of TensorQ Air

Dramatic Improvement in Revenue Management, addressing these gaps
Autonomous Agentic AI For Airline Commercial

What is TensorQ Air

What is TensorQ Air ?

↙

Revenue Finder AI

↘

Metric Mover AI

Fully Layered Commercial Engine

Based on a Foundation Model for Quant – QLLM

Forecasts AND Optimizes True Demand

Potentially drives 100% Effectiveness

Provides APIs To Connect To Any App / System or Data Repo

TensorQ Stack Architecture

C-Suite
Strategic Analytics

Commercial Alignment for Enterprise RM

NDC
Dynamic Pricing

Network
Linear Optimization

Microsegment
Numerical Optimization

Network
Dynamic Optimization

Microsegment Demand Forecast Model

Demand Enrichment, ETL

Streaming Data

Batch Data

Data Model of Microsegments

HR, IT, Finance,
Cargo, Ops

Transactional Data
Orders, PNRs, Schedules, Promotions,
Loyalty, Distribution, Marketing

External Data – Pricing, Commissions,
Schedules, Events, GDP, Weather

Use APIs to link up any app to the Microsegment data model

Execute Local Optimization for each Microsegment in parallel. Network is globally optimized due to elegant MS design

Create Multi Variable Demand Model For Each Microsegment

Vectorize Inputs, Outputs in Microsegments

Create 1 training and 1 scoring pipeline

Automatically create Microsegment Model

Data imported from Source. No ETL Needed

Core RM / NP Strategic IT / AI Functions

Why Today’s RM Cannot Find Revenues, And Avoid Losses ?

Virtually all RM systems work on the past-predicts-future Time Series theory. But past sales alone do not predict future sales.

This theory fell apart a long time ago, and Covid made it obsolete.

Why Today’s RM Cannot Find Revenues, And Avoid Losses ?

Only TensorQ uses a combination of:

Microsegments +

Bayesian regression +

Monte Carlo method.

There are numerous inputs, from past sales to social media signals, to virus spread data to news announcements to discounts to weather patterns…

This delivers predictions that fit actual outcomes better.

Microsegment Revenue Opportunity Heat Map

Siloed vs TensorQ Integrated Modeling

Example of Siloed vs TensorQ’s Integrated Modeling

Today	Siloed Model	Siloed Data	Siloed Output
RM System	RM Modeling	Historical Bookings, Price	Bookings as a function of Price
Sales Force	Sales Modeling	Agency Sales, Incentives	Sales as a function of Incentives
NP System	NP Modeling	Revenues, Schedule	Revenue as a function of Capacity, Slot
Excel	Marketing Modeling	Responses, Offers	Response as a function of Offers
Future	Unified Model	Unified Data	Unified Output
TensorQ	Full Commercial Modeling	Historical Bookings, Price, Agency Sales, Incentives, Revenues, Schedule, Responses, Offers	Bookings as a function of Price, Incentives, Schedule, Offers Sales as a function of Price, Incentives, Schedule, Offers Revenue as a function of Price, Incentives, Schedule, Offers Response as a function of Price, Incentives, Schedule, Offers

Siloed vs TensorQ Integrated Modeling

Example of Siloed vs TensorQ’s Integrated Modeling

Today	Siloed Model	Siloed Data	Siloed Output
RM System	RM Modeling	Historical Bookings, Price	Bookings as a function of Price
Sales Force	Sales Modeling	Agency Sales, Incentives	Sales as a function of Incentives
NP System	NP Modeling	Revenues, Schedule	Revenue as a function of Capacity, Slot
Excel	Marketing Modeling	Responses, Offers	Response as a function of Offers
Future	Unified Model	Unified Data	Unified Output
TensorQ	Full Commercial Modeling	Historical Bookings, Price, Agency Sales, Incentives, Revenues, Schedule, Responses, Offers	• Bookings as a function of Price, Incentives, Schedule, Offers • Sales as a function of Price, Incentives, Schedule, Offers • Revenue as a function of Price, Incentives, Schedule, Offers • Response as a function of Price, Incentives, Schedule, Offers

Fully Autonomous Optimization Agent

Agent is switched on
Agent instructed to Optimize Metrics
Itcontinuously optimizes the entire commercial function
Pricing, Promotions, Marketing decisions autonomously taken
Management informed of all the key levers optimized

How QLLM Supports Network Revenue Management

Today’s Network RM

Non-Linear Demand
High-Dimensionality (100s of Legs, 1000s of ODs)
Dynamic Pricing Next Big Challenge
LP and DP are difficult to solve

Full Network Effect and Linearization of Revenue, Demand Can Be Expressed with Independent Microsegments

MS1

MS2

MS3

MS100

MS1004

MS10005

Delivers Full Demand Forecasting

Supports Any Optimization Architecture

Leg RM Optimization
Network LP Optimization
Dynamic Programming
Numerical Optimization

Bid Prices For Dynamic Pricing

Option I : QLLM – Supports Dynamic Optimization

Max

∑_i=1,j=1^I,J p_i max { f + v(x − 1), v(x) } + p₀ v(x)

I – Itineraries
f – Itinerary Fare Class fare
x – demand allocated to Itinerary i on Leg j
J – Leg j

Max

∑_i=1,j=1^M,J p_i max { f + v(x − 1), v(x) } + p₀ v(x)

M – Microsegment Itineraries
f – Microsegment fare
x – demand allocated to Microsegment having Itinerary i on Leg j
J – Leg j

QLLM Microsegments replace Fare Class Segmentation

Option 2 : QLLM Supports Network RM LP Optimization

Fare-class based LP

Max ∑_i=1,j=1^I,J f y

S t
∑ y_i ≤ c_j

I – Itinerary Fare Class
f – Itinerary Fare Class fare
y – demand allocated to Itinerary i on Leg j
J – Leg j

Microsegments based LP

Max ∑_i=1,j=1^M,J f y

S t
∑ y_i ≤ c_j

M – Microsegment Itinerary
f – Microsegment fare
y – demand allocated to Microsegment M
with itinerary i on Leg j
J – Leg j

QLLM Microsegments replace Fare Class Segmentation

Option 3: QLLM’s Own Demand Optimization

Max_i=1,j=1^M,J { f y − ( COS + COM + COT ) }

S t
∑y_i ≤ c_j

M – Microsegment Itinerary
J – Leg j
f – Microsegment fare
Y – demand allocated to itinerary i on Leg j
COS – Direct or Allocated Cost of Sale
COM – Direct or Allocated Cost of Marketing
COT – Direct or Allocated Cost of Technology

Forecasting in Microsegments
Demand = Linear function (company levers, domains, customer choice, Competitor Levers)

Optimization
Monte Carlo Numerical Optimization using Demand Model and Demand Drivers for Local Revenue Maximization of Microsegment

Local and Global Network Revenue Optimization
Primary effects of Demand is Local to microsegment
Demand Drivers including Fare optimized for each microsegment subject to Leg capacity constraints
Aggregating Local optimal demand and revenue results in Global optimal of demand and revenue
Supply optimization is implemented through Bid Price control

Microsegment Optimization Ensures Local and Network Wide Optimal

Benefits of using QLLM

Revenue Optimization Becomes Futuristic, Fully Blending AI and OR

Very Accurate Leg-Segment AND Origin and Destination Forecasting
You can continue your leg forecasting, use QLLM’s Leg and OD Forecasting, making development unnecessary but test it out thoroughly.
QLLM forecasting also gives you the same total forecast revenue by any view : Sales, RM, NP, Distribution, Loyalty, Corporate vs Retail, GDS vs Online, etc
Provides an accurate and easy path to migrate RM towards NDC requirements, dynamic pricing
You can internally develop this LLM using our design. Total ownership and control based on license
Very economical in the long run and a positive step function change in Revenue

Economics of QLLM

Your entire AI roadmap is solved in a single Model

Takes 6 months vs 5 years for normal AI roadmap

Has potential to drive up Incremental Revenue 3-5%

This is mainly due to better demand forecasting and local optimization

Very simple hardware needs. Very small storage needed. Very few GPUs needed (10–20 max)

ROI can be very big as you would not need to spend on multiple AI projects all over

Yet the Revenue impact can be very high

Year 1

Year 2

Year 3

Year 4

Year 5

5 Years Normal AI Path

6 Months

6 Months to Full Optimization with QLLM Path

License Model

License the Product

Full QLLM Software Product License

Product is configured by Quad Optima
Product on Prem or Private Cloud
No ETL or significant work done by Indigo team
Work with IT to get access to data
Configuration , Training of Models 2-3 months
Model outputs available for Testing in 3-4 Months from Start

Pricing : Upfront plus annual

License the Design

Full Algorithm Design License

Fully Functional Design provided by Quad Advisory
License agreement for Indigo to USE the DESIGN
Indigo can develop platform to support internal use
Quad Advisory team will help Indigo Development team
Configuration and Training may take upto 6 months due to internal development
Model outputs available for Testing from 6 months from Start

Pricing : One time

For 3 Billion Airline expected annual incremental revenues – 100-150 million USD

About Us

Our vision is delivered through an elegant Digital Twin AI architecture featuring a single enterprise catalog, single deep learning AI platform anchored by a single data fabric of your entire organization.