This talk is about data-driven transformation and its contribution to Digital transformation. The first part shows the necessity to adopt the "software revolution" to adapt constantly to the customer’s environment. I then speak about " Exponential Information Systems" that the the foundation for the data-driven ambitions : Enterprise-wide flows, Customer-time data freshness, Future-proof unified semantics, etc.
The last part talks about Exponential Technologies, such as Artificial intelligence and machine learning, to drive more value from data
1. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 1/26
Exponential Information Systems
to support a Data-Driven
Digital Transformation
Yves Caseau
Group CDIO, Michelin
NATF (National Academy of Technologies of France)
http://informationsystemsbiology.blogspot.com/
https://twitter.com/ycaseau DATAQUITAINE
February 10th, 2022 – v0.2
2. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 2/26
Part 1: Digital Transformation
Driving the software revolution to adapt constantly to the customer’s environment
Part 2: Exponential Information Systems
Software excellence matters – Build your foundations
Part 3: Data-Driven Ambition
Enterprise-wide flows, Customer-time freshness, Future-proof unified semantics
Part 4: Exponential Technologies Ambition
Artificial intelligence and machine learning to drive more value from data
Outline
3. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 3/26
Information Systems as Core Digital Capabilities
Digital transformation is a
business transformation, across
the value chain
IS as a backbone (Part 2)
Shared “digital core” … but each
digital world has its own
ecosystems: (IT ≠ Digital)
Digital continuity creates value
Alibaba & Amazon example:
Digital Supply Chain Meets
Demand Management
AI to grow new knowledge from
end-to-end processes
Digital Employee
Support Systems:
Infrastructure, Data, Identity & security, orchestration, API, ….
Product
Development
Supply
Chain
Manu-
facturing
Services
&
Solutions
Sales &
CRM
Digital continuity
4. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 4/26
4
Product Development and Knowledge Engineering
AI as a tool to capture, share and scale process and product
knowledge
Hybrid AI, from DeepMind (cf. Part 4) –
to ML-augmented finite element simulation
AI and generative ML techniques to re-invent product expertise
5. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 5/26
Industry 4.0 : Digital Manufacturing, Digital Twin & Digital Workspaces
AI in Manufacturing to
absorb complexity
cope with variability
cope with manufacturing
process complexity
Augmented humans and
augmented environments
machine vision &
sensors for enhanced
perception
End to End
process
optimization
Merck Example
Middleware / HA/ Containers
Shared
Datalake
Shared
Expert
Services
Infrastructucture / Security
Middleware / HA/ Containers
Middleware / HA/ Containers
Middleware / HA/ Containers
Digital Core (App Server / CICD)
Digital Twin
Digital Twin
Digital Twin
Information
Systems
PLM
ERP
MPM
6. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 6/26
Digital Customer Journeys
Events, Insights and Context
Co-operation Agents/ Digital is built on a virtuous loop
Insights are fed by events, events fed by conversations, conversations fed by content
Daily
Yearly
Life
Time
Customer
Journey
(Sales / Service ) Agent
CDP
3rd Party
Data
Process
Data
Engagement
DMP
Advice /
Recommendation
Service /
Assistance
Content
Insights
Mining
Reactive
care
Conversations
feedback
events
requests
Web/
Mobile/
Social
02.04.2021 Retain for: 90d
v. Croc / t. fraudet / a. LemblÉ / t. signarbieux
Steering committee api transformation < N° >
d3
Group Martech Stack at Michelin
Apostrophe
MediaMath
BlueConic
inRiver
Wedia
????
Salesforce
Didomi
?????
?????
WordPress
Salesforce
Marketing
Cloud
Rul.ai
Sprinklr
Pixel
MediaMath??
Excel Sheet
Qualtrics
Pixlee??
CloudImage??
Flutter
Apostrophe
Apostrophe
7. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 7/26
7
Support Hybrid Way of Working : remote collaboration
DIGITAL WORKPLACE
Augmented collaboration : “Kolmogorov compression”
The smarter the AI, the more succinct the context synchronization
Cognitive Agents: From ontologies to “GPT-3 + semantics”
knowledge management and augmentation
8. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 8/26
Part II
Exponential
Information
Systems
8
9. Yves Caseau - Lean Software Factories and Digital Transformation – 2021 9/17
Entreprise 3.0
“Holomorphism”
Open
Platforms
Antifragile
Customer
Homeostasis
Recognition &
Response
Network
of Teams
Massively
Transformational
Purpose (MTP)
Orientation
Client
Scalability
On-demand
Algorithms
& Automation
Continuous
Learning
Experimentation
Interfaces
Autonomy
Agility
Short steps
Communities
& Crowds
E3.0
Customer
Orientation
10. Yves Caseau - Lean Software Factories and Digital Transformation – 2021 10/17
Exponential Information Systems Principles
Digital Homeostasis
Outside-in
Reactive
Open frontiers
Accelerate Takt Time
Automation
Short release cycles
Living System
In/out breathing
Continuous architecture
(emergence)
Outside-In
Customer focus
EDA
Data-Driven
Multimodal
CICD
Elastic
Resources
Algorithms
Continuous
Refresh / Refactor
Sustainable
Growth
SRE
AI4Ops
API
11. Yves Caseau - Lean Software Factories and Digital Transformation – 2021 11/17
Architect for Change : Multimodal Architecture
Four zones illustration:
different rates of change,
different software ecosystems
Extension of bi-modal IT
pattern (but everything must
change)
Edge is the software domain
that is not controlled by IT but
where it must project its
services
The supporting integration
capabilities is a key enabler
for digital transformation
Enterprise Integration Capabilities
Business capabilities
• Records
• Transactions
• Business
Intelligence
CORE
Renewal
every 10 years
Engagement capabilities
• Mash-up
• Contextual
• Conversations
• Personalization
MATRIX
Renewal
every 5 years
Edge capabilities
• Smartphones
• Social Platforms
• Connected
Objects
• Etc.
EDGE
Fast & imposed
Renewal
Imported
Capabilities
• AI/ ML
• NLP/Semantics
• IOT
Exponential Cloud
Renewal
every 3 years
API
API
API
12. Yves Caseau - Lean Software Factories and Digital Transformation – 2021 12/17
Lean Software Factories
Lean & Agile: short-term delivery
of small value increments, long-
term iterative learning
Lean thinking : continuous
management of technical debt to
develop « situation potential »
(tomorrow’s agility)
Lean practices : right on the first
time, continuous learning through
kaizen
Product mode : short-term
delivery of small value increments,
long-term iterative learning
Agile Manifesto
• Sprints
• Focus on user
• Cross-functional
teams
• Coevolution of
code/design
SCRUM
• Rites
• Retrospectives
Lean
• Kanban
• Kaizen
• 5S and
waste
removal
CICD
Continuous
Integration
Extreme Prog.
Continuous
Delivery
DEVOPS
Continuous Testing
Infrastructure
as Code
Dev & Ops
cross-functional
teams
Product
Lifecycle
• Test-driven
• Code is valuable
13. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 13/26
Software Craftmanship – Embedded Agility
“Show & Share” : Develop and value software excellence
through peer reviews
Make Code Reviews more pleasurable and more efficient:
Coding standards and Pair-programming
“Love your code” : code elegance as a support for business
agility (code that one likes to modify)
13
14. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 14/26
Part III
Data-Driven
Company
14
15. What we need to build this ambition:
• Digital continuity & Unified data models, to break siloes and
create enterprise-wide flows
• Data lakes & APIs, to foster open innovation and data
democratization
• AI-ready software stacks, to leverage the outside innovation
flow, and access to large and elastic computing resources.
Data -Driven Innovation
Define &
Produce Data
Store &
Forward
Data
Analytics
Data
Services
Data Products
Creating
Value
from
Data
Data strategy
Internal data
External data
Datalakes
Data Fabric
See
Understand
Predict
Service
exposure
Virtual Goods
Data API
Exchange
Platform
Empower our customers
Data
Intelligence
Adapt
Automation
Data
Collection
Architecture
Infrastructure
Create value for Michelin processes Create value for our customers
Turn data into assets that enable us to make better decisions, to deliver better operations and
to offer better solutions to customers and partners
Mindset: distributed and emergent innovation
Data collection/ training sets
AI-friendly software environments
Lab Culture (Data Science)
Perseverance
Constant
flow of
software
It takes
time to
build
skills
Data-Driven Innovation
16. Yves Caseau - Event-Driven Architecture for Smart Systems – 2020 16/19
Data Infrastructure
Collection Flow Infrastructure
Storage
Infrastructure
Referentials
Datalake(s)
« Hot »
Analytics
Platforms
« Cold »
Analytics
Platforms
Events
External
Sources
IOT / Video
Integration
Service
Platforms
AI & ML
Services
Classification
Forecast
NLP / Semantics
Planning
Optimization
Distributed
Data Ledgers
Sharing
Distribution
Synchronisation
17. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 17/26
Data Infrastructure Principles
CAP Theorem : it is a new world
Eventual consistency
synchronized with Business Processes
Right-time architecture (events)
High Availabilty & Scale
Data Quality emerges from QoS &
Synchronized Process Design
Quality of User Experience matters
Excellence requires focus and
perseverance
Break data siloes with federated
models and pivot objects
Shared semantics (AI ready)
Rosetta stone for standards and
platform strategy
Pivot Business Objects Data Quality & Processes
18. Yves Caseau - Event-Driven Architecture for Smart Systems – 2020 18/19
Event-Driven Enterprise Architecture
Hot & Cold Interplay
Cf. LSTM architecture
Bio-mimicry : combine cortex with
reflexes
Smart routing of events as
distributed system control
Where we plug the AI toolbox
Reactive (reflexes) and Reflective
(learning from event patterns)
Hierarchical event model
Events and Business Process
duality
Outside-In thinking to design
companies as platforms
Event Model Complex Event Processing Beyond Lambda
19. Yves Caseau - Event-Driven Architecture for Smart Systems – 2020 19/19
Data Meshes
Distributed Agility
Scalable Event-Driven
Think of data as evolving flows
“Lambda architecture” is built-in
(data lakes as nodes – temporal decoupling)
Modularity / Federation
“Architectural Quanta” based on “distributed
domain driven architecture”
More an art (experience) than a science –
CAP vs transactions.
“Data as product”
Align people (governance) and systems
(flows)
Change management as #1 priority
Discoverable, self-describing, inter-operable,
trustworthy
20. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 20/26
Part IV
Artificial Intelligence
and
Machine Learning
20
21. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 21/26
Exponential Tech: Software Engineering Matters
Data Engineering (flows) and
platforms
Lessons from Google, Criteo,
Amadeus, etc.
Ecosystems / rate of change /
integration
Software engineering
because of integration and
speed of change
Future data is better than past
Continuous learning /
enrichment cycle /
Speed of cycle is critical
Iterative
Developement of
AI Practice
Speed of learning
depends on
computing power
Smart
Algorithms
Smart
Engineering Smart
Services
Service
Usage
Growing
Large
Datasets
Distributed Software
Engineering Practices
Management Vision
& Grit
Ease to collect Trust &
Acceptability
21
22. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 22/26
Leveraging the Diversity of the AI Toolbox
Todai Robot Example
Systems of systems brings
Resilience (biomimicry)
Multi-scale
Explainability
From AlphaGo to AlphaFolds
Large-scale Intelligent Agents
communities
Game theory to reason about
competition and cooperation
Reinforcement learning
Transfer learning, GAN,
recurrent networks
Generative Approaches,
Randomization (MCTS)
Meta-Heuristics Hybrid Machine Learning Systems of Systems
23. Yves Caseau - Event-Driven Architecture for Smart Systems – 2020 23/19
Designing Systems of (Smart) Systems
Individual and collective
learning
Hybrid AI required
Time horizons (reflexes to LTP)
Explainable / certifiable / black-box
“Prediction is the essence of
intelligence”. Yann Le Cun
Requires a “model of the world”
Deep Learning for perception
Anticipation requires system
engineering and adaptive
learning
Event-Driven Architecture
Vision
Perception
Communication
Neighbors
Robots
Information
System
Autonomous
Robot
React
Reflexes
Plan
Execute
Think
Decisions
Sensors
Goals
Individual
Memory
Forecast
Learn
adapt
Collective
Memory
Behaviors
Rules
Valuation
Patterns
Analysis
Machine
Learning
Behaviors
Rules
Valuation
Patterns
Human
worker
Cloud
hosting
24. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 24/26
TRAINING SETS
“Data is the new code ?” (P. Haren & H. Verdier @ NATF)
… with training protocols
Importance of benchmarking ( Netflix, Allstate, UPS, …)
Learning curve / communities /
nested guild structure at Michelin
Training sets as a KPI for advanced data capabilities
25. Yves Caseau - Exponential Information Systems for Digital Transformation – 2019 25/23
AI @ Michelin : Some Examples
Quantitative Identification
of aggressiveness of mine
roads using computer vision technology
Stone density and size identification
Plotting over the mine area map with heat
signatures
MVP
Productionized
Predict the volumes of the
replacement market for the
next 5-year horizon
Productionized
Product to scan tyres using
tyre image, extract
dimensions and other details (SI, LI,
etc.) in order to recommend the right
Michelin tyre for online consumers to
increase sales
MVP
Product to use AI to recommend
tyres to e-retail customers
Blackcircles POC: Model performance is promising, and
areas of improvement identified.
Consumer segments generated from the model along with
the recommendations were new and useful insights
POC results
validated
Generative Adversarial
Network
''Application of GAN for Reducing Data
Imbalance under Limited Dataset'' accepted
for VISAPP 2022 conference authored by
Gaurav Adke
Technology
Enablers
Predict raw material prices
for next 1 year horizon
Deployed the product for 7 raw materials
(6PPD, TMQ, CBS, TBBS, ZNO, Insoluble
Sulphur, COCL2 )
Productionized
26. Yves Caseau - Exponential Information Systems towards Data-Driven Digital Transformation – 2022 26/26
Digital transformation is homeostasis :
Perpetual change to leverage best the possibilities of “exponential tech” in order
to match the expectations of a fast-changing world
Data Science & Systems Engineering have never been so exciting …
Exponential revolution is happening now
The main risk is not to create too little value with data,
It is to leave the field of disruption to some (possibly unknown) competitors
Conclusion
Editor's Notes
CRITICAL : print the version with Notes !
The target
The means to reach the target
How to build the mean
Part II : we will now consider the impact on IS
Intro:
SW is eating the world does not mean that IT is running the company
It means that each part of the value chain will undergo a digital transformation that requires support from IT capabilities
Digital Fabric : core of IT services that are required = cf Digotal core of first Digital Manuf illustration
Digital Twin and Digital Environment => world of IoT & connected objects => need support for IOT management, data and security
Recall the ecosystem argument : one size does not fit all !
Each digital transformation story has its own opportunities and constraints
IT is a key enabler, but digital transfo is lead by the business
Introduction : Reinventing the cookie repicipe with Tensor Flow
Many similar story in manufacturing – they are not public
Digital transformation implies a complete reinvention of processes and servicesLook at Human+Machine Nike example with rurring shoes
ML works much better with meta-data human expertise : training a ML is a form of knowledge captureonce capture this may be shared and distributed
ML training as a knowledge collaborative platform : what has happended for machine vision (with the ImageNet data set)
(3) Digital supply chains take the order management data in real time => from forecasting to reactive scheduling
Demand management uses the digital supply chain to provide a better experience (real time update) – the B2C standard (not for tires yet)
Title : Three component of DT in manufacturing
Digital to automate & optimize the manuf process (continuity)
Digital Twin: more advanced optimization based on simulation (anticipation / forecast / …)the Age of IOT
Digital workspace : human augmentation => the complete environt is helping (cobots, smart visualization)The age of smart objects (your world is the user interface to comuter assistance)
DT is about complexity management
DT is augmented humans & augmented machines
Digital Twin : end to end optimization and reingineering – true transformation versus 30 years of siloed planning.
Innovation in the digital world is more difficult => Lean Startup is born from analyzing failures and successes
Need the customer cooperation to understand the pain point and to build a value proposition
MVP : most famous term from LS Up, tool to collect feedback and accumulate knowledge (only way not to dispair)
Kevin Kelly : complex smart systems are grown not designedThese three loops are the summay of my 3 years as AXA head of digital = listen, do, learn
The target
The means to reach the target
How to build the mean
Part II : we will now consider the impact on IS
Intro:
SW is eating the world does not mean that IT is running the company
It means that each part of the value chain will undergo a digital transformation that requires support from IT capabilities
Digital Fabric : core of IT services that are required = cf Digotal core of first Digital Manuf illustration
Digital Twin and Digital Environment => world of IoT & connected objects => need support for IOT management, data and security
Recall the ecosystem argument : one size does not fit all !
Each digital transformation story has its own opportunities and constraints
IT is a key enabler, but digital transfo is lead by the business
(1) La première zone « Core » regroupe les capacités « métiers » du système d’information classique en tant que support des processus métiers. Cette zone est bien sûr elle-même multimodale, de façon fractale, ce qui permet d’implémenter une transformation continue, par exemple avec l’introduction de micro-services et d’APIs internes. On retrouve ici le concept du « Operational Backbone » dont l’exposition de services recomposables au moyen d’API est considérée comme une des pièces angulaires de la transformation digitale.
Une architecture de SI, qui décompose un système en sous-systèmes et modalités de composition, est fractale dans le sens ou cette décomposition s’applique de façon récursive aux sous-systèmes.
(2) La zone qualifiée de « matrice » est la zone d’engagement et de composition de services. Cette zone, de type « fast IT », est construite pour un taux de changement plus élevé
(3) La zone « exponentielle » représente l’ensemble des services fournis par des fournisseurs externes pour des fonctionnalités « avancées ». Cette séparation permet de mettre l’accent sur le rythme encore plus élevé de changement et sur le fait que l’entreprise utilise ces services tels qu’ils sont et ne maîtrise pas leur évolution. Penser comme une zone séparée permet de mieux visualiser le besoin d’expérimentation, de test et de protocole d’intégration (il faut expérimenter pour comprendre, et comprendre avant d’intégrer).
(4) La dernière zone, dénommée « edge », représente l’environnement logiciel et numérique du client, qui est choisi par le client et construit par des acteurs multiples. L’entreprise n’est qu’un acteur parmi d’autres, voire de temps en temps un « parasite » (un petit acteur qui profite de l’effort massif d’acteurs plus gros).
DevOps
CICD
Infrastructure as Code
Mixing Dev & Ops roles and Skills
Cycle : Repeated Loop
Peer review at every possible scale
Make the reviews more pleasurable and more efficient
(3) Key insight : Agility is not only a matter of mindset and post-its, it is a property of the code
Arnaud Lemaire
(4) Elegance : Minimal; Intent readability and virality
Data-Driven at Michelin :
The three contributions of IS
Break siloes
Democratization
Le bon environnement logiciel et matériel => cf rapport de l’ADT
Key idea: not very subtle ; data is like water in the see everywhere Data infrastructire
collection
Data Fabric
Data consumption platforms – everywhere / where AI can be applied
What is expected from IS as far as data is concerned
Ability to share data every where => implies a shared semantics => Digital starts with data, but data starts with shared data model
(2) Data is no longer static => need to cope with massive distribution and rates of change
data quality => data freshness + Qualigy of operations
(3) We live in the world of massive amounts of data that are distributed (place of use and place of creation) Right-time : pseudo-temps réel adapté aux besoins métiers
So, what does it mean to apply EDA at the EA scale ?
Les systèmes réactifs sont définis dans le Reactive Manifesto comme étant responsive (réaction rapide), résilient, élastique et message-driven (assemblés par envoi de messages).
Une des caractéristiques des systèmes digitaux modernes est précisément leur scalabilité, qui s’appuie sur une approche par événement, une distribution massive des traitements et des outils de traitement des flux d’événement.
(2) Ouverture = pub/sub + standardized API
(3) Hot : flow (use cold)
Part II : we will now consider the impact on IS
Intro:
SW is eating the world does not mean that IT is running the company
It means that each part of the value chain will undergo a digital transformation that requires support from IT capabilities
Digital Fabric : core of IT services that are required = cf Digotal core of first Digital Manuf illustration
Digital Twin and Digital Environment => world of IoT & connected objects => need support for IOT management, data and security
Recall the ecosystem argument : one size does not fit all !
Each digital transformation story has its own opportunities and constraints
IT is a key enabler, but digital transfo is lead by the business
Lessons from 3 years at ADT (Big Data & AI) + NAE Conférence
System engineering to handle lots of data + data flows => leverage tech constant moving edge
Past data is not the new oil – contrary to what macron says – data should be a flow
The technology power (CPU & skills) dictate the speed of the reinforcement learning cycle
Intro : la grande révolution de Michelin est de passer du batch au fil de l’eau dans le traitement des intéraction clients.
Les systèmes réactifs sont définis dans le Reactive Manifesto comme étant responsive (réaction rapide), résilient, élastique et message-driven (assemblés par envoi de messages).
Une des caractéristiques des systèmes digitaux modernes est précisément leur scalabilité, qui s’appuie sur une approche par événement, une distribution massive des traitements et des outils de traitement des flux d’événement. Grand projet stratégique : passer du EDA local au EDA global
(2) Ouverture = pub/sub + standardized API
EDA inter entreprise + API = perte de contrôle (de EDI to API)
(3) Reactive = Fast & Smart
Exemple de robot autonome dans une usine – formant une communauté
Comme la maison inteligente, c’est un systeme de systèmes
(1) Double apprentissage de l’individu et de la communauté – cf. Tweet récent de Elon Musk« Everyday your @Tesla gets smarter from all the data feeding into the AI system.(2) Illustration du besoin d’IA hybride
(3) Idée clé de Yann LeCun : un système autonome intelligent dispose d’un modèle de son environnment et il fait des prévisions pour anticiper (pas seulement réactif). Message clé pour EDA :)
(4) Utilisation des meta heuristique (reinforcement learning par ex) et de system engineering (loops) pour construire cette intelliegnce adaptative
Key idea: not very subtle ; data is like water in the see everywhere Data infrastructire
collection
Data Fabric
Data consumption platforms – everywhere / where AI can be applied
Introduction : smart home story
Biomimicry : simpler systems for low level functions
Need memory / multi-time -scale thinking
Need planning / goals to action – smart systems develop intents dynamically
Need foracasting / requires « real world modelling » … more than time series or pattern forecasting
T