﻿1
00:00:00,300 --> 00:00:01,000
Okay, off we go.

2
00:00:01,000 --> 00:00:03,633
So this data set, it's got a few columns.

3
00:00:03,633 --> 00:00:06,900
It's got row number customer ID surname.

4
00:00:06,900 --> 00:00:10,833
So what we're looking at is a data
set of a bank.

5
00:00:11,466 --> 00:00:13,366
of course it's all fictional.

6
00:00:13,366 --> 00:00:16,366
It's not a real bank,
but it's very realistic.

7
00:00:16,366 --> 00:00:19,166
And here we've got a snapshot of

8
00:00:19,166 --> 00:00:22,833
if you scroll down to the bottom
is going to be 10,000 customers.

9
00:00:22,833 --> 00:00:24,933
So 10,000 rows in this data set.

10
00:00:25,900 --> 00:00:28,166
And what the bank

11
00:00:28,166 --> 00:00:31,800
did is they measured some things
about these customers.

12
00:00:32,033 --> 00:00:32,333
Right.

13
00:00:32,333 --> 00:00:34,366
Why do
why did they undergo this whole thing.

14
00:00:34,366 --> 00:00:35,333
And what's the challenge here.

15
00:00:35,333 --> 00:00:39,300
Well the bank has been seeing unusual
churn rates.

16
00:00:39,300 --> 00:00:41,700
So churn is when people leave the company

17
00:00:41,700 --> 00:00:45,633
and they've seen customers leaving,
at unusually high rates.

18
00:00:45,633 --> 00:00:47,600
And they want to understand
what the problem is,

19
00:00:47,600 --> 00:00:50,266
and they want to assess and address
that problem.

20
00:00:50,266 --> 00:00:54,766
And that's why they've hired you
to look into this data set for them.

21
00:00:54,900 --> 00:00:57,000
And give them some insights.

22
00:00:57,000 --> 00:00:59,000
And how did this dataset come to be?

23
00:00:59,000 --> 00:01:03,033
Well, six months ago, the bank said,
all right, there's a big problem.

24
00:01:03,433 --> 00:01:06,200
We got to take a sample of our customers.

25
00:01:06,200 --> 00:01:07,433
By the way, this is a sample.

26
00:01:07,433 --> 00:01:10,433
That 10,000 is a tiny number for this
bank.

27
00:01:10,433 --> 00:01:12,733
This bank has millions of customers.

28
00:01:12,733 --> 00:01:16,066
It operates this
this fictional bank operates in Europe

29
00:01:16,533 --> 00:01:19,666
in three countries
France, Spain and Germany.

30
00:01:20,233 --> 00:01:22,233
And they have lots and lots of customers.

31
00:01:22,233 --> 00:01:25,833
So what they did is
they took this sample of 10,000 customers

32
00:01:26,333 --> 00:01:29,600
and they measured six months ago
everything they knew

33
00:01:29,600 --> 00:01:33,233
about them, their customer ID,
their surname, credit score,

34
00:01:33,400 --> 00:01:37,900
their geography,
their gender, their age, their tenure.

35
00:01:37,900 --> 00:01:42,066
So how long they've been with the bank,
the balance of the customers at that point

36
00:01:42,066 --> 00:01:45,466
in time, the number of products
they had at that point in time.

37
00:01:45,466 --> 00:01:46,466
So a number of products says

38
00:01:46,466 --> 00:01:49,500
things like how many products do
they have to have a savings account?

39
00:01:49,500 --> 00:01:51,100
They have a credit card.

40
00:01:51,100 --> 00:01:52,533
do they have a loan?

41
00:01:52,533 --> 00:01:55,400
did the customer
have a credit card or not?

42
00:01:55,400 --> 00:01:57,033
So there's a yes no flag.

43
00:01:57,033 --> 00:01:59,566
Is the customer
an active member? Another yes.

44
00:01:59,566 --> 00:02:00,233
No flag?

45
00:02:00,233 --> 00:02:03,266
Active member can be measured differently
by different organizations.

46
00:02:03,500 --> 00:02:04,600
It could be whether or not

47
00:02:04,600 --> 00:02:07,600
the customer logged into their online
banking in the past month,

48
00:02:07,800 --> 00:02:11,666
whether they did a transaction
in the past two months or some other.

49
00:02:12,033 --> 00:02:15,966
A measure like that,
and estimated salary.

50
00:02:15,966 --> 00:02:18,300
So the bank doesn't know
the salary of the customers,

51
00:02:18,300 --> 00:02:21,566
but based on the other things they know,
they could estimate a salary

52
00:02:21,566 --> 00:02:24,000
for that customer.
And they also gave you this information.

53
00:02:24,000 --> 00:02:27,000
So six months ago
they measured all of these things

54
00:02:27,233 --> 00:02:30,500
and said, all right,
so for these 10,000 randomly selected

55
00:02:30,500 --> 00:02:33,933
customers, we're going to just, 
watch them.

56
00:02:33,933 --> 00:02:36,700
So we're just going to wait
six months and six months down the track.

57
00:02:36,700 --> 00:02:38,300
We're going to check

58
00:02:38,300 --> 00:02:41,300
who all of those customers left
and who those customers stayed.

59
00:02:41,533 --> 00:02:43,933
And that's what this column exited
represents here.

60
00:02:43,933 --> 00:02:48,200
It tells you whether or not the person
left the bank within those six months.

61
00:02:48,200 --> 00:02:53,166
So this person over here, sometime
within the six months he left the bank.

62
00:02:53,166 --> 00:02:56,166
And as of a couple of days ago, he's
no longer with the bank.

63
00:02:56,533 --> 00:02:59,400
This person over here, on the other
hand, is still off the bank.

64
00:02:59,400 --> 00:03:00,400
So there's a zero here.

65
00:03:00,400 --> 00:03:04,266
This person left the bank,
this person stayed and so on.

66
00:03:04,266 --> 00:03:06,966
So if you see a one, that means
the person is no longer for the bank.

67
00:03:06,966 --> 00:03:08,766
Is there a person still for the bank?

68
00:03:08,766 --> 00:03:12,600
And your goal is to create
a geographic segmentation model

69
00:03:12,600 --> 00:03:17,500
to tell the bank which of their customers
are at highest risk of leaving.

70
00:03:17,700 --> 00:03:21,300
What I wanted to mention today
is that for a lot of customer

71
00:03:21,300 --> 00:03:23,333
centric organization,
this is going to be valuable.

72
00:03:23,333 --> 00:03:24,766
So I've done this personally.

73
00:03:24,766 --> 00:03:26,433
I've done this so many times.

74
00:03:26,433 --> 00:03:31,366
It is such an a value add to any customer
centric organization.

75
00:03:31,366 --> 00:03:36,200
So when whenever an organization deals
with customers this is a lot of value add.

76
00:03:36,233 --> 00:03:40,200
And then the other thing
is that this skill that you're going to

77
00:03:40,200 --> 00:03:42,300
learn is very transferable.

78
00:03:42,300 --> 00:03:44,200
It doesn't have to be for a bank.

79
00:03:44,200 --> 00:03:46,900
It doesn't have to be for churn rates.

80
00:03:46,900 --> 00:03:49,500
Geo demographic segmentation
models can be applied

81
00:03:49,500 --> 00:03:52,533
to millions of scenarios.

82
00:03:52,533 --> 00:03:56,700
So here, for instance, even in a bank,
the same scenario could work.

83
00:03:57,433 --> 00:04:00,833
should the person get a loan or not?

84
00:04:00,833 --> 00:04:03,700
Should
the person be approved for credit or not?

85
00:04:03,700 --> 00:04:05,466
And once again,
you'd have a binary outcome.

86
00:04:05,466 --> 00:04:09,666
So based on prior experience
you would know, whether or not

87
00:04:09,666 --> 00:04:12,033
a person is reliable
and you would build a model and say

88
00:04:12,033 --> 00:04:13,066
which people are

89
00:04:13,066 --> 00:04:16,800
more likely to be reliable and
which people are more likely to default.

90
00:04:17,000 --> 00:04:20,000
And that could govern the bank's decision
on whether or not to give loans.

91
00:04:20,200 --> 00:04:23,100
This could be in terms
of fraudulent transactions,

92
00:04:23,100 --> 00:04:25,966
not even in a bank,
in a different financial institutions.

93
00:04:25,966 --> 00:04:28,233
You could, find out which transactions

94
00:04:28,233 --> 00:04:31,233
are more likely to be fraudulent,
which are less likely.

95
00:04:31,800 --> 00:04:35,400
there's lots of, scenarios
where you could apply

96
00:04:35,400 --> 00:04:37,466
a geo demographic segmentation model,

97
00:04:37,466 --> 00:04:39,633
and it doesn't
even have to be geo demographic.

98
00:04:39,633 --> 00:04:43,366
So whenever you have a scenario,
when you have a binary outcome

99
00:04:43,366 --> 00:04:47,400
and you have lots of,
independent variables, you can build a

100
00:04:47,600 --> 00:04:51,133
proper robust model and that will tell you
which factors influence the outcome.

101
00:04:51,300 --> 00:04:54,000
So you can apply this knowledge.

102
00:04:54,000 --> 00:04:56,133
You're going to learn in
any kind of scenario

103
00:04:56,133 --> 00:04:59,100
where you have a binary outcome
and lots of independent variable.
