Contents

PART ONE Bare analytics 1

1 Business experiments/experimental design/AB testing 3
2 Visual analytics 8
3 Correlation analysis 12
4 Scenario analysis 16
5 Forecasting/time series analysis 21
6 Data mining 24
7 Regression analysis 28
8 Text analytics 31
9 Sentiment analysis 35
10 Image analytics 38
11 Video analytics 41
12 Voice analytics 45
13 Monte Carlo simulation 49
14 Linear programming 53
15 Cohort analysis 58
16 Factor analysis 61
17 Neural network analysis 64
18 Meta-analytics – literature analysis 67

PART TWO Analytics input tools or data collection methods

69

19 Quantitative surveys 71
20 Qualitative surveys 75
21 Focus groups 79
22 Interviews 83
23 Ethnography 87
24 Text capture 91 1
25 Image capture 95
26 Sensor data 98
27 Machine data capture 102
PART THREE financial analytics 107

28 Predictive sales analytics 109

29 Customer profitability analytics 113
30 Product profitability analytics 117
31 Cash flow analytics 121
32 Value driver analytics 124
33 Shareholder value analytics 127

PART FOUR Market analytics 131

34 Unmet need analytics 133

35 Market size analytics 136
36 Demand forecasting 139
37 Market trend analytics 142
38 Non-customer analytics 145
39 Competitor analytics 148
40 Pricing analytics 152
41 Marketing channel analytics 155
42 Brand analytics 158

PART FIVE customer analytics 161

43 Customer satisfaction analysis 163

44 Customer lifetime value analytics 168
45 Customer segmentation analytics 171
46 Sales channel analytics 175
47 Web analytics 178
48 Social media analytics 182
49 Customer engagement analytics 186 2
50 Customer churn analytics 190
51 Customer acquisition analytics 193
PART SIX employee analytics 197

52 Capability analytics 199

53 Capacity analytics 202
54 Employee churn analytics 205
55 Recruitment channel analytics 208

56 Competency acquisition analytics 211

57 Employee performance analytics 214
58 Corporate culture analytics 217
59 Leadership analytics 220

PART SEVEN operational analytics 223

60 Fraud detection analytics 225

61 Core competency analytics 229
62 Supply chain analytics 232
63 Lean Six Sigma analytics 236
64 Capacity utilisation analytics 240
65 Project and programme analytics 242
66 Environmental impact analytics 245
67 Corporate social responsibility (CSR) analytics 248

Introduction
Analytics is something every business needs to stay competitive in today’s
data- filled world. Every manager needs to at least understand the basics of
analytics and when and where to apply it. This is where this book comes in:
it provides a complete roadmap of the key areas where analytics can be
used in business as well as an overview of key analytics techniques. the
book will help you to understand some of the most important analytics
techniques, which areas in business to apply them and how to get the data
to run the analytics.
it is impossible to open a leadership or management journal without
reading something on the explosion of ‘big data’, ‘analytics’, ‘business
intelligence (Bi)’, ‘knowledge management’, ‘data mining’, ‘data discovery’
or ‘decision support’.
There is often a great deal of confusion around these terms and often they
are used synonymously and interchangeably, which can often amplify the
confusion. this book is designed to eliminate some of that confusion and
help you understand the crux of analytics so you can ignore the buzz words
and hype and appreciate what it is and why it’s a vital component of
modern business. And perhaps most importantly you will become familiar
with the various key analytic tools available to you and when and why you
might use them.
there is a great deal of interest in this area because it promises to unlock
commercially relevant insights that can potentially be used to uncover new
markets, new niche audiences within markets and areas for future research
and development. highly publicized stories and business case studies
from data gods like target, Wal-Mart, Amazon, Facebook and Google
can leave normal business leaders feeling vulnerable and overwhelmed –
unsure of where to start or what to do in order to ‘catch up’. The simple fact
is that for most businesses it’s impossible to reach those lofty data analytic
heights, but that doesn’t mean analytics is only for the big guns. Nothing
could be further from the truth. Analytics can improve performance in every
business regardless of size but in order for it to deliver its promise we first
need to understand it and dispel some of the fear around it – and that’s
where this book comes in. In essence, analytics is about data and how we
can use it to improve businesssuccess and performance. 4 Clearly this
concept is not new, business leaders and senior executives have been
using past performance and business data for decades to help decide
strategy and alter course when necessary. But what is new is our ever-
expanding definition of what data is and the technological advances that
allow us to store,analyze and extract value from data that was previously
impossible.

The raw material – data

The raw material of this insight extraction process is data – whether that is
traditional data or ‘big data’. Currently the term ‘big data’ is used to
describe the fact that

everything we do, say, write, visit or buy leaves a digital trace, or it soon
will, and the resulting data can then be used by us and others to gain new
insights and improve results. Although the term ‘big data’ will probably
disappear as ‘big data’ becomes plain old data, it is currently considered
‘Big’ because of 4 Vs:

 Volume – relating to the vast amounts of data that are being

generated every second not least because of our love affair with
smart technology and constant connectivity.

 Velocity – relating to the speed at which new data is generated and

moves around the world. for example, fraud detection analytics tracks
millions of credit card transactions for unusual patterns in almost real
time.

 Variety – relating to the increasingly different types of data that are

being generated from financial data to social media feeds, to photos
to sensor data, to video footage to voice recordings.

 Veracity – relating to the messiness of the data being generated –

just think of twitter posts with hash tags, abbreviations, typos, text
language and colloquial speech.

Used effectively the 4 Vs can also deliver the 5th V – Value. And that’s
what ana- lytics is really all about – the use of data5to deliver value. And
analytics allows us to derive value by answering four key questions:

1. What happened?
2. Why did it happen?
 3. What’s happening now?
4. What might happen in the future?

Clearly these are important questions to know the answers to and analytics
makes it possible. the easiest way to think about business analytics is that
it is the process by which you take the raw material (data) and convert it
into commercially relevant insights (analytics) that can inform business,
improve performance and guide strategy (business intelligence).

Of course the validity and accuracy of that process depends on how clear
you are about the key strategic questions you are seeking to answer and
the quality of the data you use to answer those questions. So before we
dive into the various key analytics let’s step back and get really clear about
the types and formats of data that can now be analyzed.

Data types and format

when it comes to data there are a few key distinctions that are important to
under- stand. Data is structured, semi-structured or unstructured, and it is
sourced from either inside your business or outside your business.
Structured data is data that is highly organized and located in a fixed field
within a defined record or file. This includes datacontained in relational
databases or spreadsheets. Structured data is easy to input, easy to store
and easy to analyze.

Because it follows rules and is often accessed using Structured Query

language (SQl).WhileSQl represented a huge improvement over paper-
based data storage and analysis not everything in business fits neatly into
a predefined field and that’s where semi-structured and unstructured data
comes in. It is estimated that 80 per cent of business-relevant information
originates in unstructured or semi-structured data. And essentially it’s
everything else that can’t be easily inserted into fields, rows or columns. It
is often text heavy but may also contain dates, numbers and different types
of data such as images or audio files.
6
Semi-structured data is a hybrid of unstructured and structured. Thisis data
that may have some structure that can be used for analysis but large
chunks are unstructured. For instance, a LinkedIn post can be categorized
by author, date or length but the content is generally unstructured.
Likewise, word processing software includes metadata detailing the
author’s name, when it was created and amended but the content of the
document is still unstructured.

Of course the source of data is also important because most businesses

are already data rich. the problem is they are insight poor and don’t often
know how to use the data they have, never mind utilize the treasure trove
of external data that also exists. As a rule of thumb, internal data is usually
easier and cheaper to access because it is owned and controlled by the
business. This might include financial records, customer feedback,
transaction history, employee surveys, hR data, etc.

External data as the name would suggest is any data that exists outside
your business which is held either publicly or privately by another
organization. if data is public then you can collect it for free, pay a third
party for it or hire a third party to collect it for you. Private data is usually
something you would need to source and pay for from another business or
third party data supplier. External data might include weather data, social
media profile data, trend data or government-held data such as census
information.
Structured Semi-Structured Unstructured

Internal ● Point of sale ● Photos or graphics with ● website

● financialdata tags or categories ● textfiles
● customerdata ● Videos withtags orcategories ● Photos
● hRrecords ● email ● Audio
● Social media

External ● Marketresearch ● taggedphotos ● website

● gPS position data ● organised graphics ● textfiles
● Sensordata ● taggedvideos ● Photos
● weatherdata ● categorisedtext ● Audio
● Social media

7
Business experiments/
experimental design/
aB testing
What is it?
Business experiments, experimental design and aB testing are all techniques for testing
the validity of something – be that a strategic hypothesis, new productpack- aging or a
marketingapproach.
Business experiments tend to be the blanket term for this type of testing in busi- ness;
experimental design is testing that occurs in product development; while
aBtestingisthetermappliedtoteststhatoccurin marketingactivity.regardless of the term the
principle goal in this type of testing is to extract the maximum amount of unbiased
information regarding the various factors being tested so that the best course of action
can be determined before implementation.
tVshowsrunthistypeoftestingwhentheycreatea‘pilot’showtogaugeaudi-
encereactionandinterestbeforespendingahugeamountoftime,moneyandeffort
creatingthewholeseries.theseanalyticstoolsarethesameforbusiness.

When do I use it?

In order to grow and develop as a business you need to innovate and take a few risks
now and then. But innovation, product or service development or strategic changes can
backfire because of faulty assumptions or information or your cus-
tomersmaysimplynotreactthewayyouexpect.Somesortofbusinessexperiment can help to
reduce that riskconsiderably.
It should be used if you have two or more options to decide between andgetting
itwrongwouldposeaseriousproblemforyourbusiness.runningatestbetween
thevariousoptionsonasmaller,moremanageablescalecanallowyoutoworkout which one is
likely to yield the bestresults.
Plus the feedback from the experiments can help you to further refine and improve the
winning option making it even more effective.

What business questions is it helping me to answer?

In essence, business experimentation can help you to decide which option to get
behind when you are faced with one or more choices. For instance,
8 it can help you
toanswer:

● Which of these options will increasesales?

● Which of these products should we focus on releasingfirst?
● Which of these products do our customersprefer?
● Which marketing campaign produces the highest responserate?
 ● Which recruitment channels are mosteffective?

How do I use it?

thedetailwilldepend on whatmethodologyyouuse and whatyouaretryingto achievebutthe
basic process, outlined by analyticsexpertthomas H. Davenport, is essentially the same:

1 Create ahypothesis.
2 Design theexperiment.
3 runtheexperiment.
4 analyse results and followup.

Create a hypothesis
Consider what it is you are testing and create a hypothesis around that outcome. For
example, you might be keen to test changes to your product packaging to see how or if it
affects sales. Chances are you already have an idea about how the test will work out – that’s
your hypothesis. So in this case your hypothesis might be, ‘I believe that less packaging on
our products will be appealing to our customers and increasesales’; or, ‘I don’t think changes
in the packaging will influence sales atall.’
Perhaps you want to refresh your website but you’re not exactly sure what you should
change and what you should leave the same. So you decide to test it and your
hypothesis is, ‘Moving the “buy now” button from the bottom left to the top right will
increasesales.’
When you are creating your hypothesis make sure:

● Whateveryouaretestingcanbeaccuratelymeasuredfor‘better’/‘worse’or ‘pass’/‘fail’
otherwise it’spointless.
● thetestfitstheteam’sorbusiness’soverallstrategyandvalues?neverrun
atestthatcoulddamageyourreputationevenwithasmallgroupofpeople.
● thetestwilladdvaluetoyourbusiness.

Part of this stage is figuring out how you will measure your hypothesis so you
needtoknowwhatsuccesslookslike.Soinourearlierexamples,thetestaround
packaging will have been proven successful if the product with less packaging yields
more sales than the standard packaging, and the website redesign test will be
considered successful if moving the ‘buy now’ button generates more sales.

Design the experiment

next you need to consider how to conduct the experiment as cost-effectively as possible
and decide how long the experiment will take.
Some experiments are easier to design than others. If you are making a change to a
product or service then you simply assess results prior to the change, make the change
and then assess results after the change, comparing the two results. If, however, you are
testing a modification to an existing product or existing marketing 9 campaign then you
need to ensure you can accurately measure the impact of the change and compare like
with like. For example, if you were testing modifications to a product, you couldn’t test the
standard product against the modified product with five modifications because you
wouldn’t know which modifications were
makingtheimpact.thisiswhereaBtestingcomesinbecauseitallowsyoutotestversion a against
version B where version B has only one modification; that way you know with a high
degree of certainty that any difference in result is due to the singlemodification.
 Only test one aspect at a time and where possible keep your experiments as
simpleaspossible.thisisnotonlytomaintainaccuracyovertheresults,butgener- ally speaking
the more complex the test the more expensive thetest.

Run the experiment

Make sure you tell people about the experiment – especially those that will be affected by
it. Make sure people understand why you are running the experiment and that they have
plenty of advance warning.
Once the experiment is live simply monitor what’s happening to make sure
nothinghashappenedduringthetestperiodthatcoulddistortresults.Forexample,
ifyouaretestingaproductwithlesspackagingandthatproductrunsoutforfour
daysinthemiddleofthetest,ifyoudon’tknowaboutthatfromdispatchthenyou
couldassumethattheproductdidnotsellaswellasthealternative.thetruth, however, is that
customers couldn’t actually buy it for fourdays!

Analyse results and follow up

theonlypurposeof doingtheexperimentwas totestyourhypothesisand use
theresultstoguideyourdecisionmaking,so,assoonastheexperimentisover, analyse the
resultsthoroughly.
Compareactualperformanceandoutcomesagainstyourhypothesistoestablish whether or
not your assumptions and expectations were proven accurate or not. Whether you were
proven right or wrong, consider the reason for the result. Was the experiment a success?
and remember you can be proven wrong and still have
asuccessfulexperiment.Ifyouthinkmovingthe‘buynow’buttononyourwebsite
will increase sales and it doesn’t, well at least you know something you didn’t know
before and you don’t have to consider that option again. Success in anyexperiment is a
conclusive result one way or theother.
When you are assessing results look out for any unintended consequences that
occurred as a result of the experiment that you did not expect. Consider how you could
better manage them if you roll out the results or how you could take advan- tage of them
for even greater reward.
Plusconsiderwhatyou’velearnedabouttheprocessofexperimentationthatyou can apply
for better results in thefuture.
Some experiments will be clear-cut and the result will point to one course of action
over another. In those cases you can move forward with the implementation of the best
idea as soon as possible. In other instances you may need to conduct additional follow-
up tests before knowing for sure which option to pursue. For example, if the first product
modification doesn’t yield any difference to sales then you would need to test the control
product against the second product modification
andsoonacrossallintendedmodificationssoastoidentifytheonethatmakesthe biggest
positive impact onsales.

Practical example
1
Say you are the fundraising manager for a large environmental charity. You know
directmarketingisagreatwaytoraisefundsandyouhavea‘control’campaignthat works
extremely well, but you want to see if you can lift response because direct mail can be
very expensive and you want to ensure you get the most bang for your buck. So you
decide to test a few different approaches to see if any alteration can cost-effectively
increaseresults.
You have three hypotheses that you want to test:
 1 ‘Handwritten sticky notes with a personal “ask” attached to the letter
increaseresponse’
2 ‘Pre-paid return envelopes increaseresponse’
3 ‘Changing the order of the “ask” so that a high donation is asked for first will increase the
gift size on the responseform’.

toruntheexperimentfourrandomlyselected filesof 2,000customers are selected from the

customer universe of 500,000 active customers. One group of 2,000 customers
received the control pack which is currently working well. One group of 2,000
customers gets exactly the same pack except there is a handwritten post-it note on the
letter; another group of 2,000 customers receives the control pack with a pre-paid return
envelope included; and the last group of 2,000 customers receives the control pack
with a change to the size of initial donation request on the donation response form.
Whereas the control pack pro- vides four tick-box options starting with £20, going down
to £5 with a box for ‘Other’, the test pack starts at£50.
all the test mailers are sent their pack on the same day so the only difference between
each test is the specific issue you want to test for.

1
 after three weeks you check on the results and find that the post-it note increased response rate the most; adding the pre-
paid return envelope did not increase response at all; and the average gift size increased dramatically whenyou
changedtheorderofthe‘ask’ontheresponseform.thisinformationisofcourse veryuseful.
Whilethehandwrittenpersonaliseaskonapost-itnoteincreasedresponsesig-
nificantlyitalsoincreasedthecostofthemailersignificantly,resultinginaslightnet
loss.Sothiswasshelvedasausefulideatobeusedforhighdonorsonly,where theadditionalexpensewouldbe worthit. thelackof
differenceinresponsebyexcludingareturnenvelopemeanttherewasnoneedtoincludeit,therebymaking
themailerlessexpensivetoproduce.andfinallytheincreasedinitialaskwasimple-
mentedacrosstherolloutwhichresultedinsignificantlymoremoneyraised.
Using business experimentation allows you to test things without the expense orrisk.

Tips and traps

For business experiments to be useful you can only test one thing at a time and you must compare like with like. Seek to
ensure that everything about your test is the same bar the one element you are seeking to test. For example, in thescenario
above if the fundraising packs were sent on different days then that could influence the result and you might assume it was
the test element and so roll out the wrong change.
Marketingtestsaremucheasiertofacilitatethanproductdevelopmentordesign tests. If you want to test prototypes then they
must be of the highest quality other- wiseitwillskewtheresults.andthistakestimeandmoney.

Further reading and references

See for example:

● anderson, e.t. and Simester, D. (2011) ‘Step-by-Step Guide to Smart Business experiments’, Harvard Business Review,
March(http://hbr
.org/2011/03/a-step-by-step-guide-to-smart-business-experiments/ar/1)
● Davenport,t.H.(2009)‘HowtoDesignSmartBusinessexperiments’,
Harvard Business Review, February (http://hbr.org/2009/02/how-to-design-smart-business-experiments/ar/1)
http://www.mindtools.com/pages/article/business-experiments.htm

2: Visual analytics 8
 Visual analytics

What is it?
Data can be analysed in different ways and the most simple method is to create a visual or graph and look at it to spot
patterns. This is called visual analytics and is an integrated approach that combines data analysis with data visualisation and
humaninteraction.
Data is produced at an alarming rate. In 1981 futurist and inventor Buckminster Fuller proposed the ‘knowledge doubling
curve’ to explain the fact that the more knowledgeweaccumulatethefasterwecreatemoreknowledge.Upuntiltheendof the
nineteenth century human knowledge doubled every one hundred years or so. By the end of the Second World War the total
knowledge of mankind was doubling every 25 years. Today it is thought to be every 13 months and IBM have already
predicted a point where our knowledge will double every 11hours.
Now that’s a lot of data! Unfortunately our ability to collect and store that data is increasing faster than our ability to
analyse it. And while there have been a number of tools developed to automatically analyse some of it, the complexity of the
data and the questions being asked means that human beings still need to be involved to bring their creativity, flexibility and
background knowledge of the situation to the process. Visual analytics therefore allows decision makers to combine
humaninput with the enormous storage and processing capacities of modern technology to gain insight into complex
problems using advanced visual interfaces to help them to make betterdecisions.

When do I use it?

The most appropriate time to use visual analytics is when you are trying to make sense of a huge volume of data and/or if the
complexity of the problem you face could be assisted by some additional computational horsepower.
Visualanalyticsisthereforeausefultoolwhenyouneedtoattacklarge,complex
andinterrelatedproblemswherethereisalotofdatatoanalyse.Technologyis

2: Visual analytics 9
clearly essential to analytics but technology will probably never replace human beings because it’s not yet possible for
technology to get a ‘big picture’ grasp of the problem and what needs to be done from multiple different angles. Visual ana-
lytics seeks to take the best of human intellect and technology to combine them in a way that allows the technology to do
most of the hard computational work while ensuring that it is solving the right problems and the end result is palatable and
useful for the human being that will have to interpretit.
Technology can therefore amplify human cognitive ability by increasing cogni- tive resources, expanding working memory,
reducing search time and enhancing pattern recognition capabilities across large data sets.
Using visual analytics when turning data into pictures and graphics would help tell a more complete story and help to
reveal the patterns and trends hidden within that data, which could in turn aid decision making at all levels of your business.

What business questions is it helping me toanswer?

Essentially, visual analytics can help you spot patterns in data and allow you to make vast amounts of data accessible and
understandable to anyone regardless of whether they are a data scientist or statistician or not. It can help you toanswer:

● Where are my best customerslocated?

● What is the profile of my bestcustomers?
● Is my market share increasing ordecreasing?
● Is there any connection between factor X and factorY?

Visual analytics also allows you to answer these questions faster and provide the answers in a visual, more engaging way.

How do I use it?

According to the European project VisMaster, there are four separate stages in the visualanalyticsprocess–
data,visualisation,knowledgeandmodels.Thefirststage is the input and transformation of the data. Often the source data exists
in different formats and in different locations so it first needs to be integrated before visual or automatic analysis
methodologies can be applied. Other typical pre-processing tasks include data cleaning, normalisation andgrouping.
The next step is to run the automatic analysis, which will often use data mining methods to generate models of the original
data. Once a model is created you must then evaluate and refine the models. Visualisations then allow you to interact with
the automatic analysis and play around with the data by modifying param- eters or selecting other analysis algorithms. Model
visualisation can then be used to evaluate the findings of the generated models. Alternating between visual and automatic
methods is characteristic for the visual analytics process and leads to
acontinuousrefinementandverificationofpreliminaryfindings.Remember,itis
 the combination of human and technology that makes visual analytics so useful. If something doesn’t look right to the
human eye then it can be checked, refined or new analysisrun.
Ultimatelyitistheuserinteractionwiththevisualisationofthedatathatisneeded
torevealinsightfulinformation,forinstancebyzoominginondifferentdataareasor by considering different visual perspectives.
Essentially, knowledge can be gained fromvisualisation,automaticanalysis,aswellastheprecedinginteractionsbetween
visualisations,modelsandthehumanbeingdoingtheanalysis.Thankfullythereare many commercially available visual analytics
tools on themarket.

Practical example
Swedish medical doctor and academic Hans Rosling is Professor of International Health at Karolinska Institute. He is also a
statistician, data guru and brilliant public speaker. If you want to see the power of visual analytics then I recommend you
watch any of his really interesting, funny and engaging Technology, Entertainment, Design (TED) talks.
In one (http://www.ted.com/talks/hans_rosling_at_state), he talks about how his students often discuss ‘them’ and ‘us’ in
terms of the developed world and the western world or developing world. So he asked them to define exactly what they meant
by these labels. They had all learned about them in college and were con- fident they knew what they meant. DrRosling
pushed for a specific definition, and onestudentsuggestedthatthedevelopedworldwascharacterisedby‘longlifeand small
families’ and the developing world was characterised by ‘short life and large families’. It was a neat and concise definition but
was it true? DrRosling decided to test the hypothesis. Obviously in order to test such a theory an enormous amount of data
was required toprocess.
He needed mortality rates per country, birth rates per country and all that data every year for decades. To look at that data
in its raw form – perhaps in spread- sheetsordatabases–wouldhaveyieldedverylittleintermsofinsights.Thehuman brain would
not have been able to process such massive data sets and come up with any meaningful conclusions – and yet DrRosling did
using visualanalytics.
He found that the notion that his students held about the nature of life indifferent parts of the world was fundamentally
flawed. He created a visual map that showed the correlation between ‘children per woman’ versus ‘life expectancy’ for
countries across the globe and animated the chart to move through the years. Watching the
visualanimationofthedata,viewerscouldtellthatinitiallylookingatdatafrom1950 the definition was largely accurate but by 2007 it
simply wasn’t true any more. And yet here were young students being taught this definition as though it was still a hard, fast
and accuratedefinition.
Granted there were still countries such as Afghanistan where that definition still
heldtruebutthevastmajorityofcountrieshadsignificantlyreducedfamilynumbers and were living much longer than
theirgrandparents.
Thatisthepowerofvisualanalytics:itallowsmind-bogglingdatasetstobecome genuinely useful and can help us to change out
mindset about what is really hap- pening in ourbusiness.

Tips and traps

When creating the visual analytics make sure that the key strategic question the data is answering is stated clearly on the
page or screen. This will help to focus the reader’s attention on the data’s purpose so they don’t get lost in the graphic or
visualrepresentation.
The key danger with visual analytics is that you can end up becoming obsessed with the visual part of the equation and
slice and dice the data a thousand ways. Visual analytics is extremely useful for bridging the gap between the data and the
insights but only when you stick to what’s needed when it’s needed. Just because
avisualanalyticsprogramcanpresentandmanipulatethedataathousanddifferent ways doesn’t mean you need to present and
manipulate the data a thousand dif- ferent ways. Stay focused on what needs to beanswered.

Further reading and references

For more on visual analytics see for example:

● Tufte,E.(2001)TheVisualDisplayofQuantitativeInformation,2ndedition, Cheshire,CT
● http://www.visual-analytics.eu
● http://www.sas.com/visual-analytics
 Correlation analysis

What is it?
Correlation analysis is a statistical technique that allows you to determine whether
thereisarelationshipbetweentwoseparatevariablesandhowstrongthatrelation- ship maybe.
This type of analysis is only appropriate if the data is quantified and represented by a number. It can’t be used for
categorical data, such as gender, brands pur- chased, or colour.
The analysis produces a single number between 11 and 21 that describes the
degree of relationship between two variables. If the result is positive then the two
variablesarepositivelycorrelatedtoeachother,i.e.whenoneishigh,theotherone tends to be high too. If the result is negative then
the two variables are negatively correlatedtoeachother,i.e.whenoneishigh,theotheronetendstobelow.
So, for example, if (as a hypothetical example) correlation analysis discovered
thattherewasacorrelationof10.73betweenheightandIQthenthetallersomeone was the higher the likelihood is that they also
have a higher IQ. Conversely, if that correlation was discovered to be 20.64 then the taller someone was the more likely he or
she was also to have a lowIQ.
A positive score denotes direct correlation whereas a negative score denotes inverse correlation. And zero means there is
no correlation between the two vari- ables. The closer the score is towards 1 – either positive or negative – the stronger the
 correlation is. The result is considered ‘statistically significant’, i.e. important enough to pay attention to if the result is 0.5 or
above in either direction.

When do I use it?

Correlation analysis is most useful when you ‘know’ or suspect that there is a relationship between two variables and you
would like to test your assumption orhypothesis.Forexample,youmaybelievethattemperatureisaffectingsales.

An ice-cream seller will definitely sell more ice cream in hot weather but is there a
correlationbetweenyourproductandserviceandtemperature?Correlationanalysis would allow you to work thatout.
Alternatively you can use correlation analysis when you want to know which of several pairs of variable shows the
strongest correlation. So you may want to see whether temperature affects sales more than time of year for example.
And finally you can use this type of analysis speculatively on quantifiable data sets to see what emerges. Sometimes
correlation analysis will highlight anunex- pectedrelationshipthatcouldwarrantfurtheranalysisandpotentialexploitation.For
example, Walmart discovered an unexpected relationship between the purchase of Pop-Tarts and a hurricane warning.
Apparently when there was a severe weather warningintheUS,thesaleofPop-Tartsincreased.ThisknowledgeallowedWalmart to
position Pop-Tarts at the entrance of the store following a hurricane warning, further pushing up sales. An unexpected
correlation was also discovered between beer sales and nappy sales in the United States. Presumably the father sent to buy
nappies would be reminded that he wouldn’t be going out this weekend and bought some beer instead. These types of
insights can of course be extremely useful and lead to even higher sales with a little in-store productpositioning.

What business questions is it helping me to answer?

Essentially,correlationanalysiscanhelpyoutomakeconnectionsbetweenquantifi-
ablevariablesthatcanhelpyoutomakebetterdecisionsandimproveperformance. It can help you toanswer:

● Are our most loyal customers also our mostprofitable?

● Do customers purchase more when the price islower?
● Does pay influence length oftenure?
● Does number of annual holidays influenceabsenteeism?
● Is there any relationship between factor X and factorY?

Correlation analysis can be essential for testing assumptions prior to alterations in strategy or product mix.

How do I use it?

Ifyouarefeelingbraveandyouhaveascientificcalculatortohandthenyoucanuse what is known as ‘Pearson’s
correlationcoefficient’®.

1 First you need to gather your data for the two variables you want to analyse. You can calculate the correlation for any quantifiable
dataset.
 2 Create a spreadsheet or table that lists the data sets vertically in columns. In the first column, labelled x, add all the data for your
first variable (x) and in the second column, labelled y, add all the data for your second variable(y).
3 Labelcolumnthree,fourandfive‘xy’,‘xx’and‘yy’respectively.
4 Performtherelevantcalculationsincolumnthree,fourandfive,i.e.‘xy’5x
multipliedbyy,‘xx’5xmultipliedbyx,and‘yy’5ymultipliedbyy.
5 Add all the values in each column and add the total at the bottom of each column.
6 Insert the numbers into the equation to establish the correlation between the variables underinvestigation.

Alternativelyyoucanusesoftwareandtherearemanycorrelationtoolsonthemarket.Youcanmakeyourlifealittleeasierbyusingdesktopso
ftwaresuchasMicrosoftExcelthatcontainspre-installedformulastocalculateyourcorrelations.Thereare
many simple online tutorials available to explain how you use it.

Practical example
Say you wanted to find out whether there was a relationship between the price you charged for your product and the number
of units sold at that price. Often the assumption is that the cheaper a product is the more units of that product you are likely
to sell, but that hypothesis does not always hold true. Considering how important price and sales are to revenue and growth
you decide it’s time to actually establish if that assumption is true ornot.

x (price) y (units sold) xy xx yy

5.00 56 280 25 3136

7.50 54 405 56.25 2916

10.00 50 500 100 2500

15.00 40 600 225 1600

37.5 1785 406.25 10152

Sxy5 (5)(56) 1 (7.50)(54) 1 (10)(50) 1 (15)(40) 5 1785

Sx5 5 1 7.50 1 10 1 15 5 37.5
Sy5 56 1 54 1 50 1 40 5 200
Sx2 5 25 1 56.25 1 100 1 225 5 406.25
Sy2 5 3136 1 2916 1 2500 1 1600 5 10152

Thisresultindicatesthatthereisnostatisticallysignificantcorrelationbetweenprice and unitsold.

Tips and traps
If you already have the data, you might like to try some speculative correlation analysis to see if you can find unexpected
relationships or connections that you could exploit for additional sales.
If two variables are correlated that does not imply that one caused the other it simply means there is a relationships
between them. Don’t be caught out by assumingcausation.Equally,justbecausetwovariablesarenotcorrelateddoesnot mean
they are independent of eachother.
Remember, establishing a correlation between two variables is not a sufficient condition to state categorically that there is
a causal relationship between the two. A business experiment would help to clarify if a causal relationship doesexist.

Further reading and references

Correlation analysis is a basic statistical method that is covered in more detail in most statistics books and websites. See for
example:

● Urdan, T. (2010) Good books are Statistics in Plain English, London: Routledge
● Rumsey, D. (2011) Statistics For Dummies, Hoboken, NJ: WileyPublishing
 Scenario analysis

What is it?
Scenarioanalysis,alsoknownashorizonanalysisortotalreturnanalysis,isamethodofprojection.Itisananalyticprocessthatallowsyouto
analyseavarietyof possible future events or ‘scenarios’ by considering alternative possibleoutcomes.
Byplanningoutthedetailrequiredtoimplementaparticulardecisionorcourse
ofactionyoucanobservenotonlythefinalpotentialoutcomebutalsotheviabilityofthepathleadingtothatoutcome.Oftenit’sonlywhenyo
ureallyconsiderwhatwouldbeinvolvedintheactualimplementationofanideathatyoufullyappreciatethescopeofthatidea.Scenarioan
alysisthereforeallowsyoutoimprovedecisionmakingbyfullyconsideringtheoutcomesyouexpectandtheirimplementationimplicatio
ns without the cost and time involved in actual real-worldimplementation.
Scenario analysis does not rely on historical data and doesn’t expect the future to be the same as the past or seek to
extrapolate the past into the future, rather it tries to consider possible future developments and turning points.

When do I use it?

Use scenario analysis when you are unsure which decision to take or which course of action to pursue. It can be especially
useful if the implications of the decision are significant. For example, if the decision would cost a great deal of time or money
to implement or if the ramifications of getting the decision wrong could be fatal for the business then scenario analysis can be
a very powerful tool.
Itcanbeusedtoassessthepossiblelikelyfutureofdifferentstrategicchoicesor
itcanbeusedtogenerateacombinationofdifferentscenariosthatlookatthesame scenario but from three different perspectives –
the optimistic version of events,the pessimistic version of events and the most likelyscenario.
Itisalsoaveryusefultoolifyouareunclearaboutwhatisgoingtobeinvolvedin the execution of a strategy or decision as the
process required pushes you toreally
engage with the scenario you are testing. This amplified engagement can help to anticipate more of the pros and cons of
each scenario therefore reducing risk and directing you to the best choice.

What business questions is it helping me to answer?

Scenario analysis can help the decision-making process by looking at the likely implications of that decision and how it might
or could pan out in the future. It can help you to answer:

● Which strategic direction do wetake?

● What are the best countries to expand our businessinto?
● Do I open in a new location or upgrade or expand the retail stores I currently have?
● Do I invest in a new market or seek to increase market share in the one I’m alreadyin?

Scenario analysis can help to prevent errors of judgement and direct strategy.
How do I use it?
Essentially, what you are doing in scenario analysis is attempting to work out if the world would turn out a certain way if
certain conditions were met. The process usually consists of a five-stage process:

1 Define theproblem.
2 Gather thedata.
3 Separate certainties fromuncertainties.
4 Developscenarios.
5 Use the outcome in yourplanning.

Define theproblem
Obviously the only reason you would use scenario analysis is if you were trying to gain insight into a particular challenge.
The first step is therefore to define the problem you are trying to solve or gain greater understanding of it so that you can
make the bestdecision.
It is also important to think about the time horizon. Most decisions need to be
madewithinatimelinesomakesureyouhaveenoughtimetoconductthescenario analysis before the decision needs to bemade.

Gather the data

So what is going to affect or influence the scenario you are considering? Identify what data and information you need to make
the analysis as realistic as possible.
You may, for example, consider trends and what uncertainties exist around your scenario.
You could use PESTLE analysis as a guide in gathering data – what couldaffect the outcome when you consider politics,
economy, social, technical, legal and environmental issues? Also seek to identify the key assumptions on which the plan
mightdepend.

Separate the certainties from the uncertainties

You will invariably come into the analysis process with a host of assumptions and
preconceptionsabouthowtheanalysiswillturnout.Itisimportantthatyoubecome
awareofwhatthoseassumptionsaresoyoucanreallyshinealightofenquiryonto those assumptions and separate the certainties
from theuncertainties.
Take a moment to challenge all your current assumptions and decide if they are certainties or uncertainties. It is always
best to err on the side of caution; that wayif the outcome is better than expected it’s a bonus, whereas an outcome worse than
expected could be a disaster and would negate the whole purpose of running the analysis in the firstplace.
List the uncertainties in order of priority with the largest, most significant uncer- tainties at the top.

Develop scenarios
Starting with the top uncertainty – what would you consider to be a good outcome for that uncertainty? What would be a bad
outcome? Once you’ve done this develop a story of the future around each that marries the certainties with the outcome
you’ve chosen.
Do the same with each of the major uncertainties you’ve listed.

Use the scenarios in your planning

This process will give you much more knowledge and clarity around the situation you face and you can use the scenarios to
influence and guide your planning.
There are commercially available scenario analysis tools on the market that can make scenario planning much easier.

Practical example
Scenarioanalysisisessentiallyaplanningtoolthatcanallowyoutoidentifyvarious factors that could affect a proposed plan and
 assess how those factors may play outinthefuturesoyoucanseewhichalternativeismostlikelytoworkoutwell.
Say you are planning to start a new business that helps clients implement aspe-
cificnewsoftwareprogram.Youwantthebusinesstobeturningover£1millionwithin fiveyears.Butisthatfeasible?
Afriendsuggeststhatyourunsomescenarioanalysis tohelpyougetaclearpictureofthatchallengeandhowlikelythatoutcomereallyis.
You gather data on trends and current realities. Among other things, you dis- cover that people tend to hold off on buying
new hardware and software during a recession and you are currently in a recession. That said, economic pressure also
increasespotentialcustomers’desiretoincreaseproductivityandyoursoftwarecan meet that need. The software vendor is also
working on an upgrade that is already at beta testing, so your clients could potentially reap even greater rewards. The only
possible challenge is that your software is quite new and innovative so youare unsure how quickly you could recruit
consultants to implement thesoftware.
Next stage is to separate certainties from uncertainties. The current economy is an uncertainty, but the recession has
definitely increased the number of people looking for work so you can feel more confident that you can find the necessary
employeestomakethiswork.Yourealiseyouhaven’tactuallyseenthebetaversion of the upgrade so you visit the vendor to see
what’s in the pipeline and are certain that the new version will be even more beneficial to your clients. What is uncertain,
however, is whether or not any other software company is working on anything
similarorbetterthatcouldseriouslyundermineyourplanningandpotentialoutcome.
Based on what you’ve discovered you create three scenarios to test:
● Best-case scenario assumes that the economy pulls out of recession and grows steadily over the next five years. It
also assumes that you’ve chosen the right software and that no big company swoops in and surpasses your software.
● No great scenario assumes the recession continues for at least another two years, which would probably mean that at
least 25 per cent of your clients would choose to defer their investment.
● Worst-case scenario assumes that a global software giant does enter your market and establishes itself within two years. This
would definitely put pressure on your newbusiness.

Having looked at the scenarios and considered their planning implications you realise that most of the risk is in the short term.
While the economy is a challenge, you could take advantage of this by educating clients in the productivity improve- ments
that software will deliver. The bigger issue is the possibility of another player entering the market with better software, so you
decide to make the business flexible by hiring a mix of full-time and contract workers so you can scale up and scale down
quickly depending on what actually happens. Plus you need to keep a very keen eye on what rival software companies are
doing so you can cross-train personnel if necessary.
Byappreciatingtheveryrealthreatarivalsoftwarecompanyposestoyourbusi- ness you can anticipate that challenge, and are
well positioned to monitor it closely sothatitdoesnothavethechancetode-railyourplans.

Tips and traps

While the outcome of scenario analysis can greatly facilitate better decision making and help to reduce the risk of big
decisions, the process of the analysis is probably
moreimportant.Bycreatingafewscenarios–eventheoutofleft-fieldscenario–you can end up uncovering new information or
insights that were previously unknown that can not only assist the outcome but what’s happening in the business
rightnow.
That said,it’s easy to assume that what you think you know about the current situation is true and jump to conclusions, or
automatically assume that you are certainaboutsomeoftheinfluencingfactors.Scenarioanalysiswillalwaysyieldthe best results
when you challenge all your assumptions in theprocess.

Further reading and references

A lot of the information in this section is based on the great information provided on the Mind Tools website:

● http://www.mindtools.com/pages/article/newSTR_98.htmOther useful sitesinclude:

● http://www.wisegeek.com/what-is-scenario-analysis.htm
● http://www.scenarioanalysis.net/