Presentation on "AI beyond the hype: Local (Belgian) Machine Learning success stories" by Peter Depypere (element61), at the BI & Data Analytics Summit on June 13th, 2019 in Diegem (Belgium)
2. My name is Peter Depypere
Iâm a Data Strategist in Data Science & Strategy team
Part of element61, 53 FTEs focusing on purely data &
analytics consulting
I want to inspire you in
âhow to get started with AI in your line of workâ
3. What we are about ...
Corporate
Performance
Management
Business
Intelligence
& Data Warehousing
Data Science &
Strategy
4. Whatâs happening in the world?
Increase in data generation1
⢠Mobile Phones, Apps and IoT have boomed
unstructured data-streams
(e.g., logs, clicks, pictures, video, tweets, etc.)
⢠Through the increased ability to track behavior,
there is a lot more data-to-capture
â e.g. online cookies, smart meters
⢠Trend to âstore now, use laterâ (Data Lake)
5. Whatâs happening in the world?
Boost in connectivity2
Connectivity allows us to connect to any
device and manage them
⢠As source of insight:
e.g., a phone or sensor (IoT)
⢠As tool of insight:
i.e., spread the load of one computation
across >1 machine
⢠As solution provider:
e.g., apps become the user-interface for
providing solutions
6. Whatâs happening in the world?
Access to analytics3
⢠Complex algorithms have been translated
into simple to-use functions
⢠Expensive software is now available for
free (= Open Source)
(e.g. SAS vs. Spark, Tensorflow, R, Python)
7. Whatâs happening in the world?
Enhanced computational power4
Through the evolutions in computational technology
we can workâŚ
⢠On larger data sets - e.g., Query 50GB (in <1sec)
⢠In a fraction of the time (GPUs â TPUs)
⢠At a lower cost
â i.e., major drop in price of chips (Mooreâs Law)
⢠Abstracted
â i.e. without even the need to touch hardware
Any connected device offers computational power
(Fog computing)
8. Whatâs happening in the world?
More Cognitive tools5
⢠More & more technologies can
directly connect w. humans
⢠Speech and gesture recognition
⢠Assistant devices
⢠These allows to use robots to replace
human labor
⢠Manual tasks âŚ
e.g., manufacturing
⢠⌠and knowledge tasks
e.g., Customer-service bots
9. âAm I tapping into all
information me, my
customers and my processes
are generating?â
âAm I understanding
all the available information
and using it in the
smartest way possible?â
Reflections to makeâŚ
These technology trends urge organizations
to make 2 reflections
Big data; defined as âthe
process of collecting, organizing
and using large sets of dataâ
Data Science or âthe use of
new mathematical approaches
aimed to turn data into new
insightsâ
⌠resulting business activities
13. âDeep Learningâ is a hot topic
but also the more advanced area of AI
COMPLEXITY
BUSINESS OUTCOME
Typical Belgian
organization
⢠Deep & Reinforcement
Learning
⢠Robotics
⢠Tactical Machine Learning
⢠Data Driven Process
improvements
âWOWâ
17. We shouldnât forget about the low hanging fruit
COMPLEXITY
BUSINESS OUTCOME
Typical Belgian
organization
⢠Robotics
⢠Deep & Reinforcement
Learning
⢠Tactical Machine Learning
⢠Data Driven Process
improvements
Less âWOWâ
but huge Business Value
18. Open new
business
opportunitie
s
Build
competitive
edge
Use Real-time & historic
alarm data
Sensors trigger alarms when values
exceed threshold
Leverage
5 000 000+ sensor entries
50 clients > 5-25 cooling systems
> 6 sensors > every 2-10 min
3. Predictive
Maintenance
1. Proactive
Notifications
2. Interactive
Reporting
We help Sabcobel move from
a Break-Fix model to Prevent-Optimize
Tactical machine learning
19. Contracts specify
a certain volume
commitment
⌠but the actual order
quantities might differ
influenced by end-product
customer demand or
alternative supplies
By having automated alerts
towards account managers to
(1) Have the customer buy the
goods anyway
(2) Sell the goods to another
customer
The sales department wants a higher
contract fulfilment
We help an international B2B-er to manage their
sales contracts data-driven & automated
Tactical machine learning Build
competitive
edge
Gain insight
and agility
20. We are helping an HR office to leverage data to be
3x more efficient
Our Customer:
Recruiting
organization
3xmore efficient
We profile all incoming applicants &
give each applicant an âemployabilityâ
score.
Tactical machine learning Build
competitive
edge
21. We help detect & predict Sales trends
Tactical machine learning
âSales of ribs next weekend will be X
in shop X and Y in shop Yâ
⢠We have historical data with
various products on various locations
⢠We can detect the impact of
trends, macro-events, weather,
seasonality, promotions, etc.
⢠We run an automated demand
forecast
How?
⢠Belgian Retailer with butchers all
over the country
Who?
Redefine
customer
service
Build
competitive
edge
Gain insight
and agility
22. Our Customer:
A bank
Impact: 80% of Churners could be identified 90
days prior to churn
CRM
We help to dedicate your efforts to those customers
churning
Tactical machine learning Redefine
customer
service
Gain insight
and agility
Each customer is â as of
now â tagged with a
likelihood to churn & this
information can be used in
marketing and customer
communication
- Specific offers
- Special attention by
account managers
- Embedding in call center
23. Smart & directed communication through
a data-driven segmentation of your customers
Tactical machine learning
Customers differ based on
Socio-demographics
Geographical coverage
Historical sales
Interactivity & proximity with
your companyâŚ
What we knowâŚ
⌠We can use this segmentation for
tailored customer interactions and upsell
opportunities
⪠Who should we send leaflets & when
⪠Who should we invite for specific
events
⪠Who is interested in buying which types
of products and at what time of the
year
Redefine
customer
service
Gain insight
and agility
24. We help HR to give a Next-Best-Training
Recommendation
Need in HR
Employees want to get the opportunity to grow themselves and expand their skillset
There is a vast amount of training opportunities defined (soft skills, technology, âŚ)
Our objective is to identify growth opportunities in your organization & define which
training an employee is most likely to benefit from given his profile and organizational needs
We gather employee
details
- age, gender, education,
job description, job
history, performance
reviews, #trainings,
We train a statistical
model aimed to
determine which
training a person would
benefit the most from
Personalized
recommendation on
training opportunities
improving efficiency of time
and resources.
Approach
Tactical machine learning Gain
insight and
agility
26. What does it take to be successful?
Strong use-case with strategic importance
The right team with the right skills
Technology to enable your ideas
27. Find a strong use-case
with strategic importance
What is a real pain-point in
the organization today?
Where are we losing money
today and what insight would
avoid it?
What do our customers expect
us to know (but we donât)?
On what insight-aspect
are we blind today?
29. Evaluate technology to enable your ideas
Some misconceptions:
⌠No, you donât (always) need a Big Data platform
⌠No, you donât (always) need to build a Hadoop cluster
⌠No, you donât (always) need a whole data science team
What you do needâŚ
⌠start simple (eg use the cloud)
⌠enable good collaboration within your team
⌠work in open source