Artificial Intelligence program
AI Paris June 12th 2018
LES CHIFFRES DU GROUPE AIR FRANCE KLM
milliards d’euros de chiffre d’affaires
552 avions en exploitation
314 destinations dans plus de 116 pays
80 595 collaborateurs
25,8
98,7 millions de passagers transportés en 2017
200 compagnies aériennes clientes
2 000 avions traités
3
AI program : an IT initiativeLead by CIO Office and OR/DS dept
Ambition
Reinforce AFKL value proposition by offering cognitive services to customers and employees
Impact AFKL profitability substantially by optimizing processes and transform organizations
Objectives
Create awareness on AI through use cases
Coordinated different organization around similar initiatives
Reinforce internal capabilities
4
Program organization : 3 waves to deliver AI value
2017 2018
October March September
Wave 1
• Achieve quick wins with highly feasible initiatives leveraging off-the-shelf solutions around conversational• Launch internal initiative building on existing momentum with business teams: cargo rebooking automation
Build credibility
Wave 2
• Finalize the completion of all quick win initiatives• Launch new initiatives with significant company value : Augmented Reality, Smart
Automation (fraud detection)
Deliver significant value
Wave 3
• Implementation of skills & sourcing strategy in AI• Work on organization• Propose a full catalogue of AI services
Sustainability
2019
5 months
Communicate on achievements & raise awareness on AI capabilities
December
5 months 4 months
• Long-term success : Set organization for AI initiatives within the Group
Post program
TODAY
5
Cognitive Councils
3 sessions July 7th November 29th
Participants VP – SVP representative from
all businesses
VP - SVP representatives
BlueLink / Cygnific
Location Paris CDG HQ AMS Digital Studio
Content • AI in the industry
• Watson capabilities
• Conversational tools
• HR re-imagined by AI
Workshops • Digital Assistant
• Customer facing staff
• E&M / Cargo / Commercial
/ Ground Ops
• HR journey
May 17th
Representative from all
businesses, BL CY
Paris Airport HQ
• Smart automation
• RPA using AI
AI Game :
Ethical impact of AI
6
Introduction to Repair
Remaining capacity after passengers is allocated to cargo
Sometimes, shipments cannot go in their associated flight: Repaired bookings
Multiple causes:
• Late shipment
• Cancelled flight, strike, …
• Wrong overbooking
• Priority bookings or previous repairs
Repairs must then be reallocated to new flights : Time consuming task, no previously existing process
How to reallocate the repairs?
CargoPax
7
Introduction to Repair
Today :
• Analysts are doing it manually
• Time consuming (10-15% of their time)
• Not efficient (multiple application to dig into)
• Solution not optimal
Opportunities :
• Let analysts focus on added value tasks
• Time saving
• Good quality of solution
• Better quality of service
CargoPax
8
Reinforcement Learning
Reinforcement Learning allows machines and software agents to automatically determine the ideal behavior within a specific context, in order to maximize its performance
9
Cargo Smart Repair
First idea was to look at historical data to apply Machine Learning algorithms; but it was not usable regarding the disparity in the process• We needed to explore a new domain : simulations
Automatize cargo repairprocess
Historical data
• Create fictive flights • Create fictive bookings/events• Environment representation :
• State: Booking configurations and available capacities of flights• Actions: Remove booking of the category volume and put it in a backlog• Rewards: Penalty corresponding to the removed booking category
Simulations
10
Timeline & Results
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
50
00
02
50
00
04
50
00
06
50
00
08
50
00
01
05
00
00
12
50
00
01
45
00
00
16
50
00
01
85
00
00
20
50
00
02
25
00
00
24
50
00
02
65
00
00
28
50
00
03
05
00
00
32
50
00
03
45
00
00
36
50
00
03
85
00
00
40
50
00
04
25
00
00
44
50
00
04
65
00
00
48
50
00
0
Perf
orm
ance
/Op
tim
al
Iteration
DQN Dueling DDQN Greedy
• More training, tuning of the model, modelisation
• Run a pilot this summer on selected flights
• Implement the solution to give an advise to analyst before the end of the year : real time data + integration
Nexts steps
• First discussions in oct 2017 to define the use case
• Historical data exploration in nov-dec 2017
• Modelisation and simulations 3 months jan to march 2018
• Proposal in april 2018
Timeline
11
Organizational support from the ProgramFocus on conversational tools
AI program support for chatbot initiatives
• Participation in the definition of bot architecture
• Referencing all conversational tools initiatives within the group (13)
• Technical support and provide guidance for NLP / training
• Ideation and finding new uses cases
• Investigating vocal tools Google Home & Amazon Alexa
We created the Chatbot Board with businesses
• To make sure that we are becoming more and more efficient & performant on bot’s
deployment
• Ensure a good coordination between different bots’ initiatives.
One monthly meeting
1
2
12
Text
Ramp-up of AI capabilities
End-to-end AI pipeline
Skills
Ecosystem of AI partners
Increase learning from data & simulations
Text, Speech & Image processing
Technology
13
Concrete achievement :
• Scientific role in the industry : Part of France IA• Set up a modern and stimulating work environment : cooperation with Digital
Studio/Digital Factory• We provided internal trainings : Text Mining, Reinforcement Learning, Computer
Vision• We offer exposure to AI conferences
Point to address and be share with IT :
• Work on data organization with IT• Build critical mass in the AI community by collocating teams if necessary• Adapt people policies to the competitiveness of the AI market
Text
Where do we stand regarding key enablers to attract and develop AI talents
14
Return on analysis on AI partnership ecosystem
• Raise company AI awareness via presentations• Leverage best in-class APIs to save development
time and improve performance
• Conduct workshop by business with specialized vendors to identify solution fit with use case
• Implement the off-the-shelf solutions for a given use case to achieve quick results
Internal capabilities
AI as a service Rapid external implementation
Robust internal implementation
• Support the implementation of strategic initiatives and help ramp-up internal capabilities
• Identify new opportunities and foster change in the organization
Exploration
• Explore cutting-edge technologies• Lower cost and longer term partnerships
Smallvendors
Largevendors
AI specialists
AI labs
03/10/2017
Artificial Intelligence program