Forecasting & Capacity Planning - IWSM Mensura · Single Exp Smoothing =PF*W7D+DF*W7PREDICTION...
Transcript of Forecasting & Capacity Planning - IWSM Mensura · Single Exp Smoothing =PF*W7D+DF*W7PREDICTION...
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
Niteen KumarCapgemini
Demand Forecasting & Capacity Planning Using TimeSeries Data
We believe that technology
is made by people, for people
Peopleare at the
centerof everything
we do
At Capgemini we see things differently
2© Capgemini 2019. All rights reserved |
Learn more about us atwww.capgemini.com
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
“ The goal of
forecasting is not to
predict the future but
to tell you what you
need to know to take
meaningful action in
the present ” - Paul Saffo
Forecast is the estimate of the future demand
WHAT IS TIME SERIES?
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
TIME ORDER
TREND
SEASIONALITY
CYCLICAL
RANDOMNESS
COMPONENT OF TIME SEREIS
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
WEEK DATA TYPE TECHNIQUE
W1 30 L
SIMPLE AVERAGE
MOVING AVERAGE
EXPONENTIAL SMOOTHING
W2 32 L
W3 31 L
W4 30 L t
d
WEEK DATA TYPE TECHNIQUE
W1 30 L+T
LINEAR REGRESSION
HOLT’S METHOD
W2 34 L+T
W3 36 L+T
W4 39 L+T t
d
WEEK DATA TYPE TECHNIQUE
W1 30 L+T+S
BASIC MODEL
WINTERS MODEL
W2 34 L+T+S
W3 38 L+T+S
W4 32 L+T+S
W1 32 L+T+S
W2 37 L+T+S
W3 40 L+T+S
W4 35....+N L+T+S
t
d
WEEK DATA TYPE TECHNIQUE
W1 30 L+T+S+C
SINUSOIDAL MODEL
W2 34 L+T+S+C
W3 38 L+T+S+C
W4 32 L+T+S+C
W1 32 L+T+S+C
W2 37 L+T+S+C
W3 40 L+T+S+C
W4 35...+N L+T+S+C
t
d
CAN YOU FORECAST?
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
WEEK DATA
W1 33
W2 34
W3 35
W4 37
W5 38
W6 34
W7 39
W8 44
+39+(39-34)
WEEK DATA
W1 33
W2 30
W3 33
W4 23
W5 30
W6 53
W7 47
W8 36
Simple Average
AVERAGE(W1:W7)
WEEK DATA FACTOR
W1 33
W2 30
W3 33 10%
W4 23 15%
W5 30 20%
W6 53 25%
W7 47 30%
W8 40
Weighted Moving Average
SUM(D*F) FOR ALL WEEKS
WEEK DATA PREDICTION
W1 33 -
W2 30 33
W3 33 31
W4 23 32
W5 30 27
W6 53 29
W7 47 43
W8 46
Single Exp Smoothing
=PF*W7D+DF*W7PREDICTION
Primary Factor (PF): 60%Damping Factor (DF): 40%
WEEK DATA PREDICTION
W1 33 26,5
W2 30 30,3
W3 33 34,1
W4 23 34,7
W5 30 36,5
W6 53 43,1
W7 47 47,2
W8 47,2
Double Exp Smoothing
Primary Factor: 0,20Secondary Factor: 0,03
CAN YOU FORECAST?
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
WEEK DATA
W1 33
W2 34
W3 35
W4 37
W5 38
W6 34
W7 39
W8 44
+39+(39-34)
WEEK DATA
W1 33
W2 30
W3 33
W4 23
W5 30
W6 53
W7 47
W8 36
Simple Average
AVERAGE(W1:W7)
WEEK DATA FACTOR
W1 33
W2 30
W3 33 10%
W4 23 15%
W5 30 20%
W6 53 25%
W7 47 30%
W8 40
Weighted Moving Average
SUM(D*F) FOR ALL WEEKS
WEEK DATA PREDICTION
W1 33 -
W2 30 33
W3 33 31
W4 23 32
W5 30 27
W6 53 29
W7 47 43
W8 46
Single Exp Smoothing
=PF*W7D+DF*W7PREDICTION
Primary Factor (PF): 60%Damping Factor (DF): 40%
WEEK DATA PREDICTION
W1 33 26,5
W2 30 30,3
W3 33 34,1
W4 23 34,7
W5 30 36,5
W6 53 43,1
W7 47 47,2
W8 47,2
Double Exp Smoothing
Primary Factor: 0,20Secondary Factor: 0,03
MEASURING FORECAST ACCURACY
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
WEEK DATA PREDICTION
W1 33
W2 30 26,5
W3 33 30,3
W4 23 34,1
W5 30 34,7
W6 53 36,5
W7 47 43,1
W8 47,2
Double Exp Smoothing
Primary Factor: 0,20Secondary Factor: 0,03
ERROR MAD MSD MAPE
3,54 3,54 12,50 0,12
2,69 2,69 7,22 0,08
-11,07 -11,07 122,61 0,48
- 4,75 - 4,75 22,55 0,16
16,45 16,45 270,71 0,31
3,92 3,92 15,38 0,08
7,07 75,16 0,21
BASIC STEPS IN FORECASTING
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
Problem definition
Gathering Data
Exploratory Analysis
Choosing & Fitting Model
Evaluating Forecast Model
PROBELM DEFINITION
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
Forecasting monthly service demand and required capacity to service those demand with higher accuracy
YEAR MONTH QUARTER (X) Q-INCIDENTS ACTUALS REQUIRED FTE
2017 11 3002017 2
2017 32017 4
2 5402017 52017 62017 7
3 8852017 82017 92017 10
4 5802017 11
2017 12
2018 1
5 ?? ??2018 2
2018 3
DEMANDFORECAST
CAPACITYFORECAST
What are those magic numbers?
GATHERING DATA
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
0
200
400
600
800
1000
1200
1-j
un
6-j
un
8-j
un
13
-ju
n
15
-ju
n
20
-ju
n
22
-ju
n
27
-ju
n
29
-ju
n
4-j
ul
6-j
ul
11
-ju
l
13
-ju
l
18
-ju
l
20
-ju
l
25
-ju
l
27
-ju
l
1-a
ug
3-a
ug
8-a
ug
10
-au
g
15
-au
g
17
-au
g
22
-au
g
24
-au
g
29
-au
g
31
-au
g
5-s
ep
7-s
ep
12
-se
p
14
-se
p
19
-se
p
21
-se
p
26
-se
p
28
-se
p
3-o
kt
5-o
kt
10
-okt
12
-okt
17
-okt
19
-okt
24
-okt
26
-okt
31
-okt
2-n
ov
7-n
ov
9-n
ov
14
-no
v
16
-no
v
21
-no
v
23
-no
v
28
-no
v
30
-no
v
5-d
ec
7-d
ec
12
-de
c
14
-de
c
19
-de
c
21
-de
c
Incident Count By Date
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
Total
juni
juli
augustus
september
oktober
november
december
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
53
0
2000
4000
6000
8000
10000
12000
14000
jun
i
juli
augu
stu
s
sep
tem
ber
okt
ob
er
no
vem
ber
de
cem
ber
jan
uar
i
feb
ruar
i
maa
rt
apri
l
mei
jun
i
juli
augu
stu
s
sep
tem
ber
2016 2017
Critical
High
Low
Medium
EXPLORATORY ANALYSIS
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
REAL TIME DATA IS NOT SO CLEAR AND VISUAL INSPECTION IS NOT SO HELPFUL
0
200
400
600
800
1000
1200
1-j
un
6-j
un
8-j
un
13
-ju
n
15
-ju
n
20
-ju
n
22
-ju
n
27
-ju
n
29
-ju
n
4-j
ul
6-j
ul
11
-ju
l
13
-ju
l
18
-ju
l
20
-ju
l
25
-ju
l
27
-ju
l
1-a
ug
3-a
ug
8-a
ug
10
-au
g
15
-au
g
17
-au
g
22
-au
g
24
-au
g
29
-au
g
31
-au
g
5-s
ep
7-s
ep
12
-se
p
14
-se
p
19
-se
p
21
-se
p
26
-se
p
28
-se
p
3-o
kt
5-o
kt
10
-okt
12
-okt
17
-okt
19
-okt
24
-okt
26
-okt
31
-okt
2-n
ov
7-n
ov
9-n
ov
14
-no
v
16
-no
v
21
-no
v
23
-no
v
28
-no
v
30
-no
v
5-d
ec
7-d
ec
12
-de
c
14
-de
c
19
-de
c
21
-de
c
Incident Count By Date
CHOOSING A FITTING MODEL
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
YEAR MONTH QUARTER (X) Q-ACTUALS INCIDENTS
2017 1
1 3002017 2
2017 3
2017 4
2 5402017 5
2017 6
2017 7
3 8852017 8
2017 9
2017 10
4 5802017 11
2017 12
2018 1
5 4162018 2
2018 3
2018 4
6 7602018 5
2018 6
2018 7
7 11912018 8
2018 9
2018 10
8 7602018 11
2018 12
Does the data follow a normal distribution?
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
YEAR MONTH QUARTER (X) Q-INCIDENTS ACTUALS
2017 1
1 3002017 2
2017 3
2017 4
2 5402017 5
2017 6
2017 7
3 8852017 8
2017 9
2017 10
4 5802017 11
2017 12
2018 1
5 4162018 2
2018 3
2018 4
6 7602018 5
2018 6
2018 7
7 11912018 8
2018 9
2018 10
8 7602018 11
2018 12
CHOOSING A FITTING MODEL
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
CHOOSING A FITTING MODEL
YEAR MONTH QUARTER (X) Q-ACTUALS INCIDENTS2017 1
1 3002017 2
2017 3
2017 4
2 5402017 5
2017 6
2017 7
3 8852017 8
2017 9
2017 10
4 5802017 11
2017 12
2018 1
5 4162018 2
2018 3
2018 4
6 7602018 5
2018 6
2018 7
7 11912018 8
2018 9
2018 10
8 7602018 11
2018 12
Is there a seasonality in the data?
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
QUARTER (X)Q-ACTUALS INCIDENTS
PERIOD AVERAGE
SEASONAL FACTOR
1 300 358 0,53
2 540 650 0,96
3 885 1038 1,53
4 580 670 0,99
5 416 0,53
6 760 0,96
7 1191 1,53
8 760 0,99
Can we deseasonalize the data?
CHOOSING A FITTING MODEL
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
QUARTER (X)
Q-ACTUALS INCIDENTS
PERIOD AVERAGE
SEASONAL FACTOR
DESESAONLIZEDINCIDENTS (Y)
1 300 358 0,53 569
2 540 650 0,96 564
3 885 1038 1,53 579
4 580 670 0,99 588
5 416 0,53 789
6 760 0,96 794
7 1191 1,53 779
8 760 0,99 770
Is there a trend in the data?
CHOOSING A FITTING MODEL
DATA UNDER ANALYSIS
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
QUARTER (X)
Q-ACTUALS INCIDENTS
PERIOD AVERAGE
SEASONAL FACTOR
DESESAONLIZEDINCIDENTS (Y)
1 300 358 0,53 569
2 540 650 0,96 564
3 885 1038 1,53 579
4 580 670 0,99 588
5 416 0,53 789
6 760 0,96 794
7 1191 1,53 779
8 760 0,99 770
Is there a trend in the data?
Model Summary
S R-sq R-sq(adj) R-sq(pred)
59,5622 75,59% 71,53% 60,93%
Regression EquationDE-SEASONLAIZED INCIDENTS COUNT = 500,7 + 39,62 QUARTER (X)
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
Can we forecastusing trend
Model Summary
S R-sq R-sq(adj) R-sq(pred)
59,5622 75,59% 71,53% 60,93%
Regression Equation
DE-SEASONLAIZED INCIDENTS COUNT = 500,7 + 39,62 QUARTER (X)
QUARTER (X)TREND FORECAST
500,7 + 39,62 QUARTER (X)FINAL FORECAST
9 857 452
10 897 859
11 937 1432
12 976 963
Can we also inclue the seasonality factor?
CHOOSING A FITTING MODEL
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
Can we forecastusing trend
Model Summary
S R-sq R-sq(adj) R-sq(pred)
59,5622 75,59% 71,53% 60,93%
Regression Equation
DE-SEASONLAIZED INCIDENTS COUNT = 500,7 + 39,62 QUARTER (X)
QUARTER (X)TREND FORECAST
500,7 + 39,62 QUARTER (X)
FINAL FORECAST(TREND + SEASON)
9 857 452
10 897 859
11 937 1432
12 976 963
Can we also include the seasonality factor?
CHOOSING A FITTING MODEL
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
HmmmmThe forecast is better compared to moving avg
EVALUATING THE FORECASTING MODEL
QUARTER (X)
Q-ACTUALS INCIDENTS
TREND FORECAST
FINAL FORECAST
FORECASTING ERROR
9 514 857 452 12%
10 907 897 859 5%
11 1376 937 1432 -4%
12 976 963
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
CAPACITY PLANNING
QUARTER (X)Q-ACTUALS INCIDENTS
TREND FORECAST
FINAL FORECAST
FORECASTING ERROR
PRODUCTIVITYPLANNED
FTEACTUAL REQUIRED
FTE
9 514 857 452 12% 6 5 6
10 907 897 859 5% 6 10 11
11 1376 937 1432 -4% 6 17 16
CAPACITY UTILIZATION IS BETTER W.R.T TO PLAN AVAILABITY
Niteen Kumar | Demand Forecasting Techniques | IWSM 2019
Will they have questions at the end?