View
4
Download
0
Category
Preview:
Citation preview
Introducing Data Science in big organizationsAdrian Badi
Senior data analyst
Demant
2
Introducing data science can sometimes feel like getting
lost in the wilderness
3
The best way is to imagine theprocess as a food recipe
Ingredients:
• One business case;
• Data supporting the case;
• The “master mind” pot stirring method
• Some basic understanding of ML;
• In a separate pan, make a success criteria;
• Plating the dish (preliminary results);
• Give taste samples (Marketing your findings and
involve management);
• Serve while it’s hot.
5
The case
6
7
Turn this … into this
The objective was therefore to
In other words: make the selection process easier for the salesmen – with data behind it!
8
The data
9
The “Master Mind”
pot stirring method
10
Successful individuals such as Benjamin Franklin, J.R.R. Tolkien, Henry Ford, Andrew Carnegie and Thomas Edison, all met with groups of like-minded people on a regular basis, to help one another achieve common goals and grow. Today, this is called a “mastermind”, first coined by Napoleon Hill in 1925.
11
DATA WRANGLING AND MACHINE LEARNING TEAMDIGITAL & IT CONSULTING TEAM
BUSINESS CASE TEAM
International Sales
Kasper L. Krogager
Business Development
Manager
International Sales
Mikkel Jarner Brevadt
Senior Business
Analyst
International Sales
Alejandra Garcia
Gonzalez, Student
Assistant
Sales & Marketing
Kasper Juul Jensen
Senior Manager
Sales & Marketing
Alessandro Pasta
Analyst
BI
Adrian Badi
Senior Data Analyst, BI
R&D
Anders Vinther Olsen
Audiology and DSP
Developer
Information
Technology
Julie Ingstrup
Digital Consultant
Information
Technology
Troels Christensen
IT Consultant
IT Strategy & Cloud
Bernadeta Jakubczak,
Data Scientist
KAPACITY
Bernafon & Sonic, DE
Danilo Krautz
Controller
Kapacity
Milan Mirkovic
Data Scientist
12
A success criterion
13
Success criterion:
A visit is successful if there is a sale no
more than 5 days after said visit.
14
Success criterion:
A visit is successful if there is a sale no
more than 5 days after said visit.
An average salesman had,
historically, a 12% success rate,
based on this definition
15
Success criterion:
A visit is successful if there is a sale no
more than 5 days after said visit.
An average salesman had,
historically, a 12% success rate,
based on this definitionSpoiler alert: our model reached 40%
16
Tasting your food
while cooking
17
18
Some ML knowledge
19
Data sets
• Historical sales
visits
• Historical sales
• Customer targets
Data model
• (Potential Units /
Revenue)
• Zip code
• Total revenue YTD
• Revenue between
visits
• Days since last
purchase
• Days since first visit
• Days since last visit
• Is first visit
• Consignment Stock
• Current target ratio
• Sales targets units
• Sales Visits with
sales within 5 days*
Decision tree/Random forest
Second time: after feedback from sales people
20
Plating
21
0
500
1000
1500
2000
2500
3000
3500
4000
-45 -40 -35 -25 -20 -15 -10 -5 510
1520
2530
3540
Units b
ought
Days
Visit ”pulse”
22
12,5% of sales visits create
sales within 5 days
About 3% of all transaction
match success criteria of
sales within 5 days
23
Over the course of the experiment (approx. 1 month, 2 salesmen, with the final product), we
reached an increase of approx. 60%* on sales. (A/B testing)
24
Serve while hot
25
Tips
• Always base a business case on data;
• Find a Master Mind that will cover almost all aspects of that business
process;
• Test hypothesis on the fly and involve the users;
• Aim your data science project towards a success criteria from the get-
go;
• Make sure you market your results as much as possible to activate
stakeholders’ attention;
• Keep up the momentum – such projects are “eye openers”, but the
initiative must be kept alive by the master mind with more and more
projects.
26
Hard pills you need to swallow to help with digestion of your meal
Backup
Time
Feedback
DeployBig organizations
have big agendas:
your time allocation
will be limited and
you need to accept
that in the beginning
Pill 1 Allocate
concentrated time
with the master mind
in order to make
sure you make
progress
Pill 2
When doing your
POC, make sure you
have your thick skin
on – you’re going to
need it!
Pill 3
Deployment will be a
sensitive subject,
since you don’t want
to “hand carry” your
solution forever. Talk
with your IT
department to see
what automation
solutions you can
adopt
Pill 4
27
Build your proof of concept
(or MVP) as a “hand
carried” solution, as fast as
possible to get initial
feedback – make sure you
are solving a problem
Build
Show your MVP to the
users. See their reaction,
their usage and ask about
the good and the bad.
Show
Give your users time to
actually “consume” your
solution and see if they
can get any value out of it
– don’t count your chicken
until they hatch.
Use
Find out the pain points
you haven’t solved yet and
start over with the next
release cycle (version x.0)
Feedback
Hypothesis
test
Hypothesis
test
Hypothesis
test
The process
28
The secret ingredient:
29
Excitement
Thank You
Recommended