Business Data Analysis w/Excel

March 8th, 2017

1
Get these Books!
This one
for your
manager!

This deck
uses many
examples
from this
This one book!
for you!

2
Who Am I?
▪ Dave Langer, VP of Data Science – Data Science Dojo

▪ 20+ years in technology:

• Roles in development, architecture, & BI/DW/analytics.
• Last job – Sr. Director, BI & Analytics @ Microsoft.

▪ Hooked on Data Science 5 years ago:

• Extensive background in data and analytics.
• Learned Machine Learning from 2nd place Netflix Prize winner.
• #1 Data Scientist on YouTube.

▪ Joined Data Science Dojo to democratize Data Science.

3
Motivation
▪ To deal with complexity and to compete
effectively, business is increasingly data-driven:
• Key Performance Indicators (KPIs)
• Balanced Scorecards
• Executive Dashboards

▪ However, skills to rigorously interpret, analyze, and

understand this data is rare.
4
Can We Celebrate?
We implement new
cost controls here.

Look at all this

goodness!

5
Not So Fast!
Statistically, this is
expected variation!

It is too soon to determine if controls

are working – we need more data!

6
The Challenge
▪ Unfortunately, business data presents
challenges:
• Business data is often aggregated.
• Business data is usually autocorrelated.
• Small amounts of data is the norm.

7
Intuition
▪ Aggregated data:
• Totaled by division, geography, time, etc.

▪ Autocorrelation – current values are related to previous

values:
• This quarter’s revenue is related to (i.e., a function of) the
previous quarter’s revenue.

▪ If you have 2 years of quarterly data, you only have 8

values!
8
What We Need
▪ Tools that can deal with summarized business
data.

▪ Techniques that accommodate small amounts of

business data over time.

▪ Recipes, patterns and rules to use the tools and

techniques to rigorously interpret, analyze, and
understand business data.
9
Our Toolkit
▪ Histograms to understand the distributions of
business data.

▪ Running records to identify trends over time.

▪ Process behavior charts to apply statistical

rigor to analyze the changes and differences
in our business.
10
Expectation Setting
▪ This presentation is very much the art of the possible:
• Not enough time to teach Excel, resources in Appendix.

▪ You will not be an expert in these techniques:

• We cannot cover all aspects of data analysis.
• There are many gotchas and prerequisites for rigorous
analysis of business data.
• Good news – you don’t need a PhD. in Statistics!

▪ Buy and study “Making Sense of Data”!

11
EXCEL WORKBOOK

12
THE HISTOGRAM

13
Example Questions & Scenarios
▪ Questions:
• “How is my business process actually executing?”
• “Is my business process behaving as expected?”

▪ Scenarios:
• You suspect that the nature of a business process is
more complicated in practice.
• Productivity isn’t what I expected, is something going
on?
14
The Histogram
Peak #1
Peak #2

Are there actually 2

processes at work?

Source – “Making Sense of Data”, Page 52. 15

THE RUNNNING RECORD

16
Example Questions & Scenarios
▪ Questions:
• ”What has been happening over time?”

▪ Scenarios:
• You’ve implemented a new marketing
campaign, what has been happening?
• Cost seem to be on the rise, how can we tell?
17
The Running Record
See trends
over time

Any Associated
Trends over time?

Can’t be used to answer

rigorous questions!

18
Source – “Making Sense of Data”, Page 174-175.
PROCESS BEHAVIOR CHARTS

19
Understanding the Chart
Upper Natural
Process Limit

50+% of
values

Lower Natural
Process Limit

20
Understanding the Chart
Danger Zone!

96+% of
all values!

Danger Zone!

21
Chart Requirements
▪ All the data being used was collected under
similar conditions.

▪ The data can be logically compared.

▪ The charts used in this talk don’t work with

well with outliers.
22
TREND ANALYSIS

23
Example Questions & Scenarios
▪ Questions:
• “Is it working?”
• “Is there something going on here?”
▪ Scenarios:
• You’ve implemented a new marketing
campaign, is it actually moving the needle?
• Cost seem to be on the rise, are they really?
24
The Rules of Trend Analysis
▪ Rule # 1 – Points Outside the Limits:
• Single point outside the limits is an indication
of a dominant effect that needs investigating.

25
Trend Rule #1
What’s going
on here!?!?

Danger Zone!

96+% of
all values!

26
Source – “Making Sense of Data”, Page 102.
The Rules of Trend Analysis
▪ Rule # 2 – Runs About the Central Line:
• Eight successive values on the same side of
the central line is an indication of a weak
sustained effect.
• Might want to investigate.

27
 What’s changed?

Trend Rule #2 PSAs?

After school specials?

50+% of
values

28
Source – “Making Sense of Data”, Page 113.
The Rules of Trend Analysis
▪ Rule # 3 – Runs Near the Limits:
• Three out of four successive values within the
upper 25% of the region between the limits or
within the lower 25% region between the
limits.
• Could be indicative of a moderate sustained
effect.
29
Trend Rule #3
Why the spike?
Marketing campaign?
Good economy?

50+% of
values

30
COMPARING GROUPS

31
Example Questions & Scenarios
▪ Questions:
• “Are things different between these two groups?”
• “Is the West division doing better than the East
division?”
• “Is org A doing worse than org B?”
Let’s use this as a
hypothetical scenario.
▪ Scenarios:
• Comparing on the job accident rates.
• Comparing employee attrition.
• Comparing sales, expenses, profit, etc.
32
Hypothetical Scenario
▪ We run a trucking company and want boost
productivity with a bonus plan.

▪ We conduct a year-long trial with select sites

having the bonus.

▪ We are interested in the possible effects of

the bonus on injury rates.
33
 Comparing Groups
Average injury rate for bonus sites is 34.75. Average injury
rate for no bonus sites is 28.67. Is the difference real?

Are there statistically significant

differences between the groups?

34
Source – “Making Sense of Data”, Page 121.
Comparing Groups If Group A and Group B were
the same, these values
would be very unlikely!

96+% of
all values!

Danger Zone!

35
Source – “Making Sense of Data”, Page 121.
Comparing Groups
Danger Zone!

96+% of
all values!

If Group A and Group B were

the same, these values
would be very unlikely!

36
Source – “Making Sense of Data”, Page 121.
Results
▪ By using process behavior chart limits, we can detect
statistically significant differences between groups.

▪ While only one “Danger Zone” situation was sufficient

to determine the groups are different, we saw two
“Danger Zone” situations.

▪ We are safe in interpreting the differences in average

injury rates are real – sites with the bonus average
more injuries!
37
Summary
▪ Business data presents unique analytical challenges.

▪ However, histograms, running records, and process

behavior charts work well with business data.

▪ Using process behavior charts we can apply rigorous

analytics to our business data – including detecting
statistically significant differences between groups!

38
QUESTIONS

39
APPENDIX

40
Get the Files!
▪ GitHub Repo:
• https://github.com/datasciencedojo/meetup/tree/master/b
usiness_data_analysis_with_excel

▪ Direct links to files:

• Excel
▪ https://github.com/datasciencedojo/meetup/blob/master/busin
ess_data_analysis_with_excel/BusinessDataAnalysis.xlsx
• PDF:
▪ https://github.com/datasciencedojo/meetup/blob/master/busin
ess_data_analysis_with_excel/BusinessDataAnalysis.pdf
41
Want more goodness?

These books
provide more
in-depth
mathematical
details.

42
Excel Resources
▪ Histograms:
• Windows Excel 2010 and Mac:
▪ https://www.youtube.com/watch?v=ujqgyrDUX1o
• Windows Excel 2016:
▪ https://www.youtube.com/watch?v=53DOu_vstvI
▪ Running Records:
• Windows and Mac:
▪ https://www.youtube.com/watch?v=mTnsxNfTFKo
43