OPERATION ANALYTICS AND

INVESTIGATING METRIC SPIKE

PROJECT DESCRIPTION

 The given project consists of 2 case studies:-

 First is regarding Operation Analytics where job data is provided and number of
jobs reviewed , 7day rolling average of throughput, percentage share of language
used and duplicates are found out.
 Second is Investigating Metric Spike where user engagement, user growth,
weekly retention, weekly engagement and email engagement is determined.
 The following information is found with the help of SQL queries.
APPROACH

The required information was determined via SQL queries where the data base was
created first in SQL and moreover for the second case study due to the size of the
data excel was used to make charts for better visualisation.
TECH STACK USED

 MySQL was used to run the queries.

 The language was selected because of comfort and experience in the same.
 MS Excel was used in the second case study for better visualisation.
 As I am currently learning this tool, it was utilised to get more hands on
experience.
CASE-I: OPERATION
ANALYTICS
Insights
 INSIGHTS-NUMBER OF JOBS REVIEWED

select
avg(t) as 'avg jobs reviewed per day per hour’,
avg(p) as 'avg jobs reviewed per day per second’
avg jobs avg jobs reviewed
From
reviewed per day per day per
(select ds,((count(job_id)*3600)/sum(time_spent)) as t, per hour second
((count(job_id))/sum(time_spent)) as p
from job_data 126.1804833 0.03505
where month(ds)=11
group by ds) a;
 INSIGHT-THROUGHPUT AND 7-DAY
ROLLING AVERAGE OF THROUGHPUT
Select throuput_ throuput_7_d
ds, ds per_day ay_rolling
c/t as throuput_per_day,
c7/s7 as throuput_7_day_rolling 25-11-2020 0.0222 0.0222
From
(select 26-11-2020 0.0179 0.0198
ds, 27-11-2020 0.0096 0.0146
count(job_id) as c,
sum(time_spent) as t, 28-11-2020 0.0606 0.0176
count(job_id) over(order by ds rows between 6
preceding and current row) as c7, 29-11-2020 0.05 0.0202
sum(time_spent) over(order by ds rows between 6
preceding and current row) as s7
30-11-2020 0.05 0.0229
from job_data
7 day rolling average is better because it can offset the
where month(ds)=11
throughput fluctuations of one day and create a more accurate
group by ds) a;
picture
 INSIGHT-PERCENTAGE SHARE OF
LANGUAGE USED IN LAST 30 DAYS

with a as language percentage

(select max(ds) as m from job_data)
select distinct
Italian 12.5
language, Persian 37.5
(count(event) over(partition by language rows between unbounded French 12.5
preceding and unbounded following) /count(*) over(order by ds rows
between unbounded preceding and unbounded following) ) * 100 as
Hindi 12.5
percentage Arabic 12.5
from English 12.5
(select *
From
job_data cross join a
Where
datediff(m,date(ds)) between 0 and 30)a1;
 INSIGHT-FINDING DUPLICATES

select *
When no
from(
duplicate
select *,
data
row_number() over(partition by ds,actor_id,job_id) as row_num
from
job_data) a
where row_num>1;

When duplicates(inserted the same

data twice for the example)
CASE-II: INVESTIGATING
METRIC SPIKE
Insights
 INSIGHT WEEKLY USER ENGAGEMENT
week of the engagem weekly engagement
year ent growth Chart Title
17 8019 NULL
18 17341 9322 25000
19 17224 -117
20000
20 17911 687
21 17151 -760 15000
23 18280 1129
22 18413 133 10000
24 19052 639 5000
25 18642 -410
29 20067 1425 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
26 19061 -1006 -5000
30 21533 2472
28 20776 -757 -10000
27 19881 -895 -15000
31 18556 -1325
32 16612 -1944 -20000
33 16145 -467
week of the year engagement weekly engagement growth
34 16127 -18
35 784 -15343
Overall Reduction in the engagement is seen
(*note:-the data of the 35th should not be considered as it is only for the first
day of the week)
(**query on next slide)
INSIGHT WEEKLY USER ENGAGEMENT

select *,
engagement-lag(engagement) over(partition by'week of the year') as 'weekly
engagement growth’
From
(select
week(occurred_at) as 'week of the year’,
count(event_name) as 'engagement’
from events
where event_type!='signup_flow’
group by week(occurred_at))a;
 INSIGHT-USER GROWTH

year_ quarter_ new_user_activated user_growth

2013 1 470 NULL
2013 2 608 138 Overall increase in quarterly seen
2013 3 930 322 (*date of 2014 quarter 3 is not of full
quarter)
2013 4 1275 345
2014 1 1692 417
2014 2 2378 686
2014 3 2028 -350
select *,
new_user_activated-lag(new_user_activated) over( order by year_,quarter_ ) as user_growth
from(select year(created_at) as year_,quarter(created_at) as quarter_,count(user_id) as new_user_activated
from users
where
activated_at is not null and state='active’
group by 1,2)a ;
INSIGHT-WEEKLY RETENTION COHORT
ANALYSIS
cohort weekly retention
4000

3500

3000

2500

2000
users

1500

1000

500

0
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69 72 75 78 81 84
weeks

cohort_retained

Major drop in the first 10 weeks at the end of 85 weeks only 2 users remain
INSIGHT-WEEKLY RETENTION COHORT
ANALYSIS
Select
week_period,
first_value(cohort_retained) over (order by week_period) as cohort_size,
cohort_retained,
cohort_retained / first_value(cohort_retained) over (order by week_period) as pct_retained
From
(select
timestampdiff(week,a.activated_at,b.occurred_at) as week_period,
count(distinct a.user_id) as cohort_retained
From
(select user_id, activated_at
from users where state='active'group by 1) a
inner join
(select user_id,occurred_at from events )b
on a.user_id=b.user_id
INSIGHT-WEEKLY ENGAGEMENT PER
DEVICE device_name
acer aspire desktop
avg_weekly_users
26
avg_times_used_weekly
32.9474
acer aspire notebook 43.1579 56.8421
amazon fire phone 10.5556 13.7778
asus chromebook 43.5263 58.8947
Given is average weekly engagement dell inspiron desktop 46.6316 62.7368
per device dell inspiron notebook 91.1053 123.4737
hp pavilion desktop 42.1053 55.8421
The weekly data per device was htc one
ipad air
21.8421
51.4444
27.6842
61.7222
very large (960 rows) hence ipad mini 30 34.7368
calculated the weekly data iphone 4s 46.6316 60.5789
iphone 5 123.1579 161.2105

Macbook pro is used the most iphone 5s

kindle fire
73.3158
21.1579
96.7895
25.5263
lenovo thinkpad 172.9474 232.5789
Samsung galaxy table is used least mac mini 20.4737 27.3684
macbook air 123.1579 164.8947
macbook pro 260.1579 358.1579
nexus 10 27.0526 31.8421
nexus 5 76.3684 99.6316
nexus 7 36.3684 43.2632
nokia lumia 635 28.1579 36.2632
samsumg galaxy tablet 10.2778 12.1111
samsung galaxy note 13.4737 17.5789
samsung galaxy s4 91.5789 118.7368
windows surface 18.2105 21.5263
INSIGHT-WEEKLY ENGAGEMENT PER
DEVICE
Select
device_name,
avg(num_users_using_device) as avg_weekly_users,
avg(times_device_use_current_week) as avg_times_used_weekly
From
(select week(occurred_at) as week,
device as device_name ,
count(distinct user_id) as num_users_using_device,
count(device) as times_device_use_current_week
from events
where event_name='login’
group by 1,2
order by 1) a
 INSIGHT-E-MAIL ENGAGEMENT METRIC
Overall increase in the engagement metric seen
num_user time_weekly_digest_s time_weekly_digest_sent_ time_email_open_ time_email_clickthrou time_email_clickthrough_gro
week s ent growth time_email_open growth gh wth
17 981 908NULL 310NULL 166NULL
18 2714 2602 1694 912 602 430 264
19 2787 2665 63 972 60 477 47
20 2874 2733 68 1004 32 507 30
21 2926 2822 89 1014 10 443 -64
22 3029 2911 89 987 -27 488 45
23 3134 3003 92 1075 88 538 50
24 3254 3105 102 1155 80 554 16
25 3343 3207 102 1096 -59 530 -24
26 3439 3302 95 1165 69 556 26
27 3543 3399 97 1228 63 621 65
28 3641 3499 100 1250 22 599 -22
29 3734 3592 93 1219 -31 590 -9
30 3866 3706 114 1383 164 630 40
31 3950 3793 87 1351 -32 445 -185
32 4023 3897 104 1337 -14 418 -27
33 4200 4012 115 1432 95 490 72
34 4294 4111 99 1528 96 490 0
35 48 0 -4111 41 -1487 38 -452
 INSIGHT-E-MAIL ENGAGEMENT METRIC
Select
week,
num_users,
time_weekly_digest_sent,
time_weekly_digest_sent-lag(time_weekly_digest_sent) over(order by week) as time_weekly_digest_sent_growth,
time_email_open,time_email_open-lag(time_email_open) over(order by week) as time_email_open_growth,
time_email_clickthrough,time_email_clickthrough-lag(time_email_clickthrough) over(order by week) as time_email_clickthrough_growth
From
(select week(occurred_at)as week,
count(distinct user_id) as num_users,
sum(if(action='sent_weekly_digest',1,0)) as time_weekly_digest_sent,
sum(if(action='email_open',1,0)) as time_email_open,
sum(if(action='email_clickthrough',1,0)) as time_email_clickthrough
from email
group by 1
RESULT

 Really engaging project, difficulty of the project makes it more fulfilling to

execute.
 Learnt a lot of new things like rolling average, cohort retention analysis.
 Tired to insert excel charts wherever I could, hopefully would be able to use excel
more efficiently next time.
 Became better in using windows function.