Using Data to Tell a Story - Part 1

Using Data to Tell a Story - Part 1
By Kirk Harrington

I recently was given really good advice about my resume.  I hadn't realized what was mentioned to me, till this advice was given.  I was told that my resume seemed to focus highly on technical skills, but not enough on storytelling, being able to explain difficult data topics to other co-workers, and my aptitude with data visualization.  Truly, I have all these aptitudes.  I became skilled at doing them through work, lots of practice, and in sharing my presentations with other associates and business partners I worked with.  Further, in one job, I was told that my presentations were a 'standard' for the team, and it was suggested that others in the team look to my presentations as good examples of how to present analytic concepts to internal partners.

So I updated my resume a bit, however I also wanted to write this article to share some of the things I have learned in my data visualization journeys.  They have been many and several and I've enjoy data visualization and the telling of stories with my data for several years.  Here are some tips and advice I can give...

Focus on metrics that matter and that tell a story that can be acted upon

It is critical that your visualization answers the main questions and reasons for why you did the visualization in the first place.  To help with this, I will typically meet with the partner I am preparing it for and ask them specific questions so I can understand exactly what they want to know.  You might ask questions  like...
'What are you trying to measure?'
'What will this be used for?'
'Who will see this and act on it (i.e. the audience)?'
'Do you have any ideas that could help me with my analysis?'
'What are important metrics that you think should be included?'
'How will knowing the answers to your questions help our customer base?'
'Have you done something like this in the past that I can look into for more insights?'
'Do you have any ideas that could help me with my analysis?'

With answers to these types of questions, you will be able to more easily provide a visualization that tells the story that wants to be heard.

Highlight and make stand out important findings

Depending on the type of visualization done, you can bold, underline, circle, highlight, use large text...something to make a key finding stand out.  What this does is it focuses the attention of the viewer of your visualization onto what they most want to know or understand about what is happening with the underlying data.

The tool you use and the fanciness of the chart or graph is less important than how clear the visualization makes your message come across

I have seen fancy charts and graphs from many an analyst over the years.  The ones that impress me the most though are those that answer the questions that need to be answered in a clear and simple way.  In my mind...if you can look at a chart or graph and within a few seconds understand it and come away with insight...it is a good chart or graph.  On the contrary, if it takes more than a few seconds to understand the chart or graph...perhaps it looks confusing or is missing key elements...then it needs more work to be considered 'good'.  When I say 'key elements', I once looked at a chart for several minutes and was a bit confused by it.  Then I realized it was missing context and a time frame for when the data was being presented.  Without those key elements of a context and time frame, it just wasn't readable.

Follow a guideline when putting together your visualizations for a presentation

Here are my guidelines for a presentation.  I typically follow this format as I put it together.  Having a format is great because it makes things more obvious and people over time will know what to expect from you.

Title Page:  This will include a title, your name, and title in the organization.  I also mention contributors.

Highlights:  This is at the beginning and is simply a set of statements which goes over the answers to the questions being answered in the rest of the presentation.  Plus, it gives the answers to decision makers right away so they don't have to way.

Charts and Graphs of Main Findings

Charts and Graphs of Additional Findings (if any):  These are other things you discovered that can provide more value to your main findings.  I also see these as 'innovative ideas' that may not be known to help push the company forward.

Summary:  With action steps that can be taken with the findings derived.  I believe this is very important.  I have seen many presentations that focus on high level learnings, but don't provide action steps that clients can take to create a given outcome.

Appendix:  This is where you put the 'nice to know' things, detailed stats, and extra information that could be useful to answer questions when you are giving your presentation.

Make sure you include 'key elements' for your given visualization

I made mention of this already, however wanted to provide more specifics.  Here are my suggestions of key elements to use for each type of visualization...

Time series:  Date range, date labels on the x axis, key seasonal events (if any), a trend line, a context for the chart (what is it about OR what is it trying to show you? - this can also be accomplished with a meaningful title), a source

Pie chart:  Percentages, contrasting colors (so they don't appear blended), a key (to tell you what each colored piece of the pie represents), a title/context

Bar and Line Charts:  Clearly defined x and y axis, Values at each bar or line (if it does not appear too busy), labels for each bar on the x axis (or a key if the bars or lines are colored), a title/context

Scatter plots:  Clearly defined x and y axis, trend line (if appropriate), a title/context, highlight of key points or clusters of points

If your visualization looks too busy or crammed, simplify it

Visualizations will be more powerful if they are simple and have less things going on at the same time for someone trying to interpret it.  Here are my suggestions for dealing with business for each type of visualization...

Time series:  Business with these can be caused by having an extra long time window and more than 1-2 metrics followed over time.  For extra long time windows, I suggest splitting up times into time windows that make sense and having separate charts (that can be shows next to or on top of each other).  I suggest if you have more than 2 metrics over time, that additional metrics are show in separate graphs.

Pie Chart:  These can easily become busy because you have many different groups with very small percentages.  I recommend grouping categories based on similar characteristics or even by how small their percentage is (if, for example you have many groups that are <1%, combine these together for your pie, but make sure your key is clear in defining the groups included that are in that range).

Bar and Line Charts:  Including too many bars or lines can make these hard to interpret.  I recommend when you do have this problem, that you group together in their own separate charts categories that are meaningful.  You might even consider not including groups that are not meaningful or where their behavior is already well-known.

Scatter Plots:  Outliers, while sometimes useful to include, can cause these plots to look confusing and can throw off your ranges.  I suggest taking out outliers altogether and not mentioning them, or take them out and talk about them separately (outside the plot).

Scatter plots can also become busy and less meaningful when you include too many populations within the same plot that have contradicting trends within your x and y axis variables.  Trying taking slivers of your populations and seeing if trends become more meaningful.

Examples of visualizations that worked well for me

Unfortunately, I cannot provide actual visualizations in this blog for the employers mentioned...they are proprietary to them.  I can talk about them though so you can get a sense of ones that worked well and how I put them together.  I recommend keeping a record of ones that you have done that were useful so you can talk about them if you have to interview with someone.

The benefit of new advertising to Lean Cuisine (Nestle)

For this visualization, I had done a time series model looking at contributions to volume over time.  Each key part of the volume was given its own color and what the colors represented was stated clearly in a key below.  Dates over time were clearly stated, and I also included a window from a year prior when old advertising was in play.  Further, I included enough of a window after the new advertising to show how the volume benefit was sustained.  So basically, this chart answered several things for the company...what did the new advertising contribute to volume vs. when it was not present the following year AND was that boost to volume sustained.  As an additional way to drive home what the chart was trying to say, I included the % of volume contributed by the new advertising directly on the chart.

Target vs. Walmart and how they reacted to the Affordable Care Act (ACA) (Innerscape Data)

With this one, I was asked to look at two companies and how they reacted to the ACA in terms of the number of employees covered in a plan.  I had a list of companies to choose from, and decided to choose two from the same industry that were competitors (where I suspected there was a difference in their behavior).  I chose Target and Walmart.  It turned out that this was a good choice, because their trends were very different.  Further, because the number of employees for each company was very different I decided to make separate trended charts for each of them and show them side by side when I put them in a presentation.  Because I had a long time horizon, I decided not to state each individual month...but instead just show years with points for each month.  My x and y axis were labeled with variable names, which I don't typically do but made sense to use in this case because the client was looking to purchase use of our variables.  As an additional way to strengthen and interpret the trends, I looked up news articles which spoke to the behavior of Target and Walmart after the Affordable Care Act was released so that I could show that our data was consistent with the commentary.

If You Are Not Innovating as a Data Organization, You Are Spinning Your Wheels

By Kirk Harrington

The idea for this article came from a dream I had this morning, where I was interviewing with a company and I asked one very important question...."Does your organization take time to do research, find data gaps (and seek to fill those gaps), look for opportunities to innovate with the data they have, while working on regular projects?"  After I had asked this question to the first person I was interviewing with, I then was passed from one data analyst to another.  Two analysts had given me some sort of excuse as to why they couldn't answer the question and the third said something like "Well, I don't have time to answer your question...I am very busy...and then he went on to describe an internal political matter that caused him grief."  Basically, the answer was that this company did not take the time to really work with their data...but instead became paralyzed with what they already had and the workloads and inner politics surrounding it.

If this company were real, and not a figment of my imagination created in my dream...I would suggest this to them.  To be competitive as an organization and to make work more interesting and vibrant, there needs to be dedicated time made for data.  In the various organizations that I have worked in...I have seen several practices that have helped a company be successful with this...for example:

• Engage in projects that identify data 'gaps'.  Reach out to your various organizations...find out their needs.  Find out what the data answers for them and what it doesn't.  Find out what they'd really like to know that their data is not answering for them now...and seek ways to fill these needs.
• Look for data overlaps and data 'excess'.  Sometimes, whether its through reporting or even data sets that are sitting in various repositories, overlaps and data 'excess' can be created which can cause needless slowdowns,workflow inefficiencies, time waste and data space waste.  A project to identify overlaps and data 'excess' can be useful to deal with this.  One project I did, for example, was to identify all the reports created at an organization and determine which reports were being used and which were not.  This project helped because we were able to identify which reports were not being used so they could be eliminated and it created an understanding of all the different reports being created that could possibly be used across the various teams at the organization.
• Take time to delve into the data and find innovative ways to stretch it.  One project I did once at an organization did this very thing.  I was tasked to understand all of their data from the various sources and how it could be used for analytics and modeling.  I went the extra step to teach them how to create variables from what they already had so it would enhance what they could learn from their customer base.  Further, I identified third party variables that were being paid for and found a suitable public source that they didn't have to pay for.  This effort really helped with the analytics and modeling that I was tasked to do after this project.
• Take time to understand the third party data you have purchased.  I once worked at an organization that had purchased various different third party data and when I asked what people knew about it, I didn't get straight answers.  What that told me is that they had purchased data, but didn't really know if it was useful or not.  I proceeded to test a given set of this third party data for a project I was working on and found it extremely useful to identify target populations for mailing (I was a marketing analyst).  Because of the success of the test, these variables were used and saved the campaign I was tasked to save.  And all it took was a little research time and trial.
• Look at improving processes that access and read in data that is being used for reporting and other processes.  I once did an interesting job at an insurance company where I was tasked to improve coding for various reports being created at the organization.  I was asked to look for coding inefficiencies as well as ways to improve the coding.  What this did, when done successfully, was improve processing time and create increased automation for the work they were doing.  I guess you could call this 'streamlining' and I found it to be a very useful practice in making things better.

Besides opportunities, I also found certain things held back organizations from doing the above.  And this speaks to the dream I had.  For example...

• Heavily siloed analysts that work on very specific things...that may not get the chance to work with other parts of the organization.  While this sometimes can't be helped, I found it useful to be engaged in cross-team projects with various data analysts across the organization...to work on projects that affected the entire organization.
• Company politics that focus on short term vs. long term goals.  This can be very stifling to an organization.  If all your focus is on short term financial goals and 'putting out fires', you are neglecting long term growth, competitiveness and steadiness/stability.  Recognizing that you have politics that hold you back can be very challenging as well...especially when people's reputations can be affected.
• Having an overabundance of work and overburdening your analysts can also be a source of stifling to innovation.  In an age where many companies are trying to cut costs and divide workloads among fewer and fewer analysts, while this approach can help with short term cost benefit, it will not help with long term growth and being competitive.  A balance should be found where workloads become manageable and that analysts are given some free room to innovate.  It will make their jobs more enjoyable and lead them to be more productive with the work they are already doing.

In my mind, if you don't take time to innovate and improve your data...as a company you become just like hamsters running in a wheel.  It looks like you are working, but you are not really getting anywhere.  And, you tire out the hamsters (your analysts) who probably want to do more with their careers than just round around in circles for you.

Claiming Success When your Project is Killed

By Kirk Harrington

Have you ever had an amazing project that would have enhanced the company killed by office politics?  And then your work just kicked to the wayside and you are asked to bury it?  That happened to me once.  Not the most pleasant experience.  The thing is though...what they couldn't take away from me was the success of the project and that it made an impact when it was in place.

What happened was this...I was tasked to create a householding methodology for the entire organization (a small bank I used to work for).  I was given leeway to create a project team, lead that team, and give them an end product that would benefit the specific group we were doing the project for (in this case it was Asset and Liability Management).  

Everything went well.  I was able to bring in a good team, given everyone manageable projects that would help with the end product, and we delivered the householding methodology (which the Asset and Liability team saw as miraculous) within a reasonable amount of time.

What happened though is that once the methodology was in place, was working well, and the client using it was happy with it....someone in the organization decided to take an ego trip (for lack of a better word) and decided that householding would be better served by having a third party do it (which by the way...that third party was notorious for not being the best to work with).  So basically, the organization spurned the advance I did...one that was done internally, that could be easily done and controlled...to a solution that was done by a third party that was less efficient with less ability to control it.  I think too the person that killed it felt embarrassed that someone like me would come in and do such a grand thing.  Again...why I call it an ego trip.  What's more...my manager felt inclined (because of office politics) to support the person taking the ego trip, so my project was basically doomed.  I was also never asked to provide project notes, coding, anything to anyone that showed that I had done the project.  In effect, it was buried.

Was I upset?  Of course I was.  Especially when I was given all the resources to do the project and it was meaningful to me because it helped the entire organization.  Further, the team I had was great and it was a very positive experience for me to work with them to create such a useful end product.

What I have determined though is this...while my project was killed ...it did do the following:

• Lit a fire under someone who put through a householding methodology for the group asking for it (even if it wasn't the best solution)
• Gave me experience managing a project team on an important project that effected the entire organization
• Gave me experience creating a householding methodology that was innovative for the company I was creating it for

So, instead of getting angry and upset because of office politics killing my project, I have decided to make it a positive.  I have included this project on my resume and its something I enjoy talking about when I speak with potential employers.  I have in effect claimed success when the project itself was killed.

Building a Magic Deck with Probability in Mind

Before I write what will be a series of articles (hopefully to be put into book form at some later date), I just want to say that I love the game of Magic the Gathering.  I have enjoyed many hours of it with friends and have come to enjoy card-based games more because of my exposure to it.  I also want to share the caveat that I am not an expert Magic player by any means, however because of my background in analytics and stats, I have been able to apply specific principles that I know to the game to help me build decks that are competitive with my friends.  I also believe these principles will be useful to anyone who wants to build their own Magic decks so they can play leisurely with their friends (and win some games).

Please rid your mind of the notion that there must be a set number of cards of each type

Other than the standard rule of not having 4 of the same card within your 60 card Magic deck, there really is no rule to how many cards you have of each type.  That said, it is important to keep in mind that you want your deck to work ‘for you’ and the ‘way you want’ within the confines and laws of probability that your deck will be built around.  And to calculate the probability of anything that you put into your deck, it is always x/60, with x being a generic number representing how many cards you have of a certain type or ability.

I like the idea of having at least 2 lands in your starting hand

The starting hand of 7 cards is very important.  Further, it’s useful to think of your starting hand as a ‘sample’ from the ‘population’ of 60 cards that you are drawing from.  Since the game focuses on having enough Mana to cast spells, I like the idea of having at least 2 lands in your starting hand.  That would come out to be 2/7 cards in your first hand, which when applied to your deck strictly would be 17/60 (7x=60 which is 8.57, 8.57 *2=17, 8.57*7=60, thus 17/60).  Keep in mind though, when you are going from a larger sample of 60 to 7, the chances of drawing at least 2/7 (or 29% chance) goes down, so I add a few more cards to this to make my chances greater (say 3-5 more cards).  So, in my example, at the minimum I recommend 20 lands, at the maximum 22 lands.

The Weighted Average Casting Cost tells you how ‘fast’ and ‘powerful’ your deck is

I honestly think they should publish the weighted average casting cost with each deck you buy.  Why?  Because it can tell you how fast and powerful your deck is.  A deck can be fast (meaning you can cast lots of low mana cost creatures and they can come out quickly), but not very powerful (because there typically is a positive correlation with casting cost and how powerful the cards are that you have).  And when I say ‘powerful’, I mean how much damage or benefit they can deal.  Further, a deck can be slow, in that it will take more time to build up mana in order to cast the cards within it.  The cards you have may be powerful, but if it takes a long time to play them, you risk dying before you get there.  This is why personally I like to keep my weighted average casting cost around 2-3.  Below two means you’re betting on less powerful cards to win and above 3 means you are betting on more powerful cards to win, however it will take more rounds for your deck to get moving.  2-3 is about average.  I will calculate the Weighted Average cost with a sample in the next sections.

For our purposes, let’s use a pre-purchased Magic deck as an example of how many cards of each type can be in a deck

I have enjoyed purchasing pre-built decks from Magic the Gathering.  I think they are, for the most part, well built.  And I expect, they are built by experts who know the game well.  So for this article, I decided to purchase the Yanling Planeswalker Deck, just so it can be used as an example.  Keep in mind though that you can totally create our own custom built deck however you want, realizing that you are at the mercy still of the laws of probability and chance as you play.

After opening this deck, I observe the following (at a high level):

There are:

25 lands, translating to the chance of having almost 2.87 cards (so 2-3 cards) in your starting hand (the math for that is 25/60=0.417, then 0.417*7=2.87.

27 creatures, translating to the chance of having almost 3.15 creatures in your starting hand (same math, except its 27/60).

4 instants, though I did notice that there are 10 creatures that have ‘Flash’ which makes it also an instant.  Keep in mind that this would effect the probability of drawing an instant (making it 14/60 instead), but does not effect the probability of drawing a creature (27/60, which is different).

3 sorceries

With either the 3 or 4 cards (or even combining them to make it 7/60), there is a chance of drawing <1 card of either type in your opening hand.  And that chance is small because of the smaller sample of 7 cards.  So I have a greater chance of drawing only creatures and lands in my opening hand.

And finally, 1 planeswalker

Calculating the Weighted Average Casting Cost for this set of cards

The Weight is calculated as the # Cards/60 for each Casting Cost type

The Weighted Average Cost is the Casting Cost * the Weight for each Casting Cost

Once you get the Weighted Average Cost for each Casting Cost, you sum this together.  For my population, the answer is 2.13.

Understanding what can be affected in gameplay first, before I run out the probabilities for abilities across cards

Before I go into calculating the probabilities of abilities across cards, I want to step back and talk about the aspects of the game that can be affected by the cards you have.  This will provide a framework, in my mind, of understanding how the deck you build (and the probabilities of abilities across cards) will play out.  And it will make it easier for someone to customize their decks and get their deck to do what they want.

Aspects of play affected (high level):

Your own creature(s) or someone else’s creature(s)

Your graveyard or someone else’s graveyard

Your draw deck or someone else’s draw deck

With a planeswalker deck, your planeswalker (or someone else’s if they also have one)

Your own spell or someone else’s spell (other than a creature)

Your land or someone else’s land

Your casting cost or someone else’s

Your hand or someone else’s hand

This is pretty much it!  And, any ability(ies) you have on a given card will affect one of these aspects of gameplay.  

Here are some examples from the deck I purchased:

Flying:  Affects my creature

Flash:  Affects my land (because I have to use a land to cast it)

Return target creature to owner’s hand (for simplicity, I will call this ability ‘Return creature’):  Affects someone else’s creature

Draw cards:  This affects my draw deck (makes it go down and makes cards available to my starting hand)

Search the deck for the Planeswalker and put it in your hand:  Affects your hand

Understanding the probabilities of your abilities across cards and why that’s important to getting your deck to do what you want it to do

Calculating the probabilities of abilities across cards is easy.  Getting your deck to do what you want and have a higher chance of that is ideal.

For our example, I will use the pre-purchased deck again.  And just go ability by ability across cards.  And realize there may be more than one ability on a given card, however for our purposes, we will calculate the probabilities separately (irregardless if there is more than one ability on a given card).

There are 16 creatures with Flying.  That can be given a couple of probabilities.  16/60:  The probability of there being a creature with Flying in the deck overall, 16/27:  The probability that at least one of your total number of creatures has flying. 

There are 10 creatures with Flash.  That translates to 10/60:  The probability of there being a creature with Flash in the deck overall, 10/27:  The probability that at least one of your total number of creatures has Flash.

There are 6 cards with Return Creature.  That is 6/60 overall, and for the Planeswalker it is 1/1, for creatures it is 1/27, and for Instants it is 4/4.

There are 10 cards where it affects positively my own creature’s strength and/or defense (+X/+Y)  That is 10/60 overall, 10/27 for Creatures.  Please note than I am combining, just for simplicity, whether is affects your own creature or someone else’s.

There is 1 card that negatively affects someone else’s creature’s strength and/or defense.  That is 1/60 overall and 1/1 for the Planeswalker.

There are 3 cards that Tap another player’s creature.  That is 3/60 overall and 3/27 for the creatures.

There’s are 2 cards that can put the Planeswalker into your hand.  That’s 2/60 overall.  2/27 for creatures.

There are 3 cards that Tap a target creature.  That is 3/60 overall.  3/27 for creatures.

There are 6 Draw Cards abilities.  That is 6/60 overall and 3/27 for creatures and 3/3 for sorceries.

There are 3 cards with Casting benefit.  That is 3/60 overall and 3/27 for creatures.

Get out of your mind that there is a ‘perfect’ deck and instead focus on increasing your chances of the deck doing what you want it to do

At this point, it is helpful to create a visual (kind of like a score card) of your abilities across cards.  Here is this deck’s based on the probabilities I just calculated…

Weighted Average Casting Cost:  2.13

Observations – This deck is fair, it could TOTALLY be better

There are several observations I have when looking at the numbers for this deck.  As follows:

• You shouldn’t have problems being able to cast most spells within the next 10 rounds.  If you have 2/7 in your hand in the first round, you will have 2 mana out within the first two rounds and then have a 23/53 chance (The denominator is 53 because you are taking out the 7 in your hand) of drawing a land on the second round (so about 2.3 cards out of the next 5.3 should hopefully be a land – Note I am purposely not taking into account the margin of error).
• This deck is creature heavy.  You won’t have a problem drawing creatures.  And, since the weighted average casting cost is 2.13 and since your starting hand will have 2-3 mana, you will be able to cast a good number of your creatures within the first several rounds (because you should have at least 4 mana in the first 5 rounds – 2.3 within the next 5.3 after your first round).
• Although there are a good number of creatures, it’s important to note that a good percentage of them are in the 1-3 mana cost range (because your Weighted Average Casting Cost is 2.13).
• Flash is a nice ability.  And in this deck, while 37% of your creatures have it, only 17% of total cards have it.
• This deck seems to want to focus on Flying, Flash, and +X/Y, but these probabilities are getting dragged down by other abilities that may or may not prove useful in supporting those (especially given their low probability scores within the deck).
• Unless the creature of the other player is a flier, the return creature ability may or may not be useful.  Further, if the other player’s creature is a flier, the tap creature might be useful…but what if it isn’t?  Do you really need Tap Creature, especially since you have so few cards with that ability?
• Given that you have a decent number of creatures and a good weighted average casting cost (and a good chance of drawing mana in a decent number of rounds), do you really need the ability to draw more cards?
• The Retrieve Planeswalker cards are very nice, however there’s only a couple of them.  This is crippling your ability to bring out this powerful Planeswalker.

My suggestions to improve this deck:

• Have more cards in the deck where you are seeking for the Planeswalker to put into your hand (Yanling’s Harbinger).  You’re allowed 4 of the same type anyway, and this Planeswalker is pretty powerful.  At the least, you can put in 2 more of Yanling’s Harbinger.
• I don’t necessarily like 27 creatures in a hand, only because I enjoy using spells more, so I would have less creatures, and more spells in the decks I build.  And if this were my deck, I’d add more Retrieve Planeswalker spells and +X/+Y spells to increase your chances of boosting your creatures more.
• I would remove cards (unless they are dual) that Tap or Return Creature or Draw Cards.  And add cards with abilities as I suggested.  This would increase the probabilities on your ‘leading abilities’ and remove cards that may make less sense to play together.
• Because you will have a good amount of mana in a small number of rounds, you could afford to increase your weighted average casting cost by removing a few less powerful creatures with lower casting costs, and adding a couple more powerful ones.  For me, the obvious choice would be to get rid of a Spectral Sailor and a Warden of the Evos Isle and add two more Air Elementals.  All else equal, that would make the Weighted Average Casting Cost 2.22.

Extra topics I may explore in future articles:

Multiple color decks – How many of each type to use?

Shuffling deck techniques (to make things more random)

Suggested ability segmentations

Scoring a deck with arbitrary numbers

Taking into account powerful combo cards

Keeping Your Passion While You're Unemployed

Being let go from Nestle was a challenge for me.  It was a hard time for me personally (with various health and family challenges at the time), but also I was extremely disappointed that I wasn't able to continue in that position...which truth be told was not the best fit for me from a 'level' standpoint.  

To give you a brief synopsis...I had worked about as far as I could in the banking industry.  I was the most senior analyst (Analyst Level 3) and I was one step before Manager.  When I took the job at Nestle (in CPG, a completely different industry) I came in at a very high level manager position...too high I think...which caused me a bit of career whiplash (i.e. I think it would have been more beneficial for me to have come into CPG as an analyst so I could have learned the ropes better).  As an aside, it was one of the reasons I was let go...Nestle and I both agreed that I did not have enough experience to be able to perform optimally in that specific position.

This left me wondering what to do once I became unemployed.  I was able to take care of my personal and family challenges during that time, which was good.  While in the position, I did very well at the analytics projects I was asked to do.  I was able to continue to explore my passion for marketing and consumer behavior analytics.  I also discovered in the role that I had a skill at negotiating with 3rd party vendors and helping to onboard them to an organization.  

I also really enjoyed working in CPG.  I learned an immense amount in the 2 1/2 years with Nestle (and something I'll be grateful to them for).  Besides learning basic metrics used to measure consumer products in the marketplace, I also learned how to do time series modeling and how to incorporate digital media measurement into my analysis and predictive modeling.  I was also able to attend a seminar in California on creating digital content with Buzzfeed.
  
And I have since learned two valuable lessons, that have helped guide me through these different waters that I am in (and I'd like to talk a bit about each)...

1. Take the Best From What You Learned in your Last Job and Do Something With It

2. Don't Give Up On What You Enjoy


Take the Best From What You Learned in your Last Job and Do Something With It

I knew leaving Nestle that I still had a passion for Analytics and Marketing.  I also knew I still had a passion for learning about consumer behavior and applying analytics to it to improve performance over time.  Further, I really enjoyed the new aspects of what I learned in the CPG Marketing space, particularly with digital and online content creation.

And I was doing something in my spare time while unemployed by myself or with friends that I've been doing all my life and am good at...namely gaming (whether online, on a computer, or socially with friends).  I even wrote a lengthy article while unemployed that I plan to publish which looks at probability and Magic the Gathering deck creation (once I figure out how to get it off a hard drive of mine that died - or I will just rewrite it).

So, I decided to marry this spare time activity to things I learned in my last job...I began to create online content for gaming and stream my own digital gaming online.  And I know what you might be thinking...but please hear me out.  I think it's easy to think that I am just finding an excuse to 'play games in my spare time', but I don't see it as this at all.  In fact, my overall goal with this is to have enough followers with this entertainment product I am creating so that I can actually make money at it (which many actually do).  Further, I am building a network with other online content creators through cross-promotion activities (for example) as I build it out.  I have done amateur acting in the past...so this has been a way to explore that again also.  And, I have invested money into capital that I use to make my product better, like video and audio equipment, props, wardrobe, and lighting.  Truth me told, it's really like a side business where I can pour my knowledge and passion into it to drive it.

If I had to make a list of the things I've done (from a business perspective) in this venture (and incorporated into it as I've rolled it out), it is as follows...

• Audience targeting for digital content (through paid promotions)
• Analysis of Facebook and YouTube metrics to improve content and increase engagement
• Networking and information sharing with other online digital content creators for sharing of best practices (and to obtain technical advice)
• Use of a variety of platforms (YouTube, Facebook, Twitter, Instagram to promote consistent messaging)
• Writing, blocking, performance, visual, photography/photo-shoots, sound, and lighting tech work
• Creating a unique product that is different than others out there, so I can remain competitive
• Cross promotion of other consumer packaged goods during my streams (i.e. snacks, beverages, skin care products)

In this treatment of it as a side business with a goal to drive engagement and profitability...I've been able to have some great experiences with it.  Some highlights so far have been...

• I was able to do two photo shoots in West Virginia (for Bethesda's Fallout 76) to enhance and drive interest in the other content I also created.
• I have been engaged in watching how others do their online streaming and have done work on cross-promotion with other streamers and content creators.
• I have taken a class on Google Analytics, which I plan to incorporate into improving the effectiveness of my various online presences.
• I have attended and learned things at Conventions where people with like interests exist.  At one of these conventions, for example, I took a couple classes with a voice actor who discussed the technical aspects of improving sound quality (and I purchased some sound equipment to help with this).
• I've been able to consistently have a loyal group of consumers that has not changed too much over time (which is a big accomplishment in this space) with the efforts I have been incorporating.
• I have had people come back to me and tell me my content is very good and they look forward to seeing it again.
• I have noticed increased engagement over time as I've done analysis and made improvements to my product and what I present.

Don't Give Up On What You Enjoy 

And I'm still learning...and that's ok!  I think it's important to realize...and as many of you know as statistical modelers and analysts...no model or analysis is perfect and will always have errors.  It is something I have certainly come to realize as I've done this venture.  I have definitely had my share of frustrations (mostly with technical challenges and not being as organized at times like I'd want to be).  But what's important is that I haven't given up on what I'm passionate about....analytics, understanding and adjusting to consumer behavior, and being engaged with a product (gaming in this example) that people can relate to and enjoy.

Lessons I Learned From Building a Successful Online Campaign - DDTTRH

The story about how I built a successful online campaign starts with a trip I made to a restaurant in Cleveland.  They had there, in a frame, a t-shirt with a list of the Rock and Roll Hall of Fame inductees over various years.  I noticed that my favorite was not on the list.  I couldn't believe it.  I then remember searching online to verify the fact, and it was true!  Duran Duran, though amazing and successful artists...were not in the place that honors such artists.  This had to change!  And it was then that I started with a commitment...to see them inducted.  And, I would do everything I could to raise awareness to how special they are and how they have made an indelible impression on our music history and popular culture.

And, with a good team I did!  Over the course of 3+ years I and the team created a petition that garnered 2500 signatures from a fan base from whom it was difficult to get to sign.  Further, the signatures came from all around the world which was a feat of its own.  Several of those signatures were from celebrities and friends close to the band, like Nile Rogers.  We also created a Twitter and Blogger site to raise awareness.  Our Blog site received almost 27,000 views from all over the world!  We interviewed bands, artists, and Duran Duran super fans as well...inviting them to talk about Duran Duran and their contribution to music and culture.  Further, we made local and national press (including the Barrie Examiner in Canada, The Sun Press in Cleveland, Yahoo News, and the Washington Post - links at the end of this article).  Since starting this effort, there has also been a rise in talk about Duran Duran and the Rock Hall, as evidenced by several articles that came out after our effort started (see also links below).

We also had a lot of fun.  We created a logo and a brand, DDTTRH.  We even released a few t-shirts.  Further, I created a partnership with The Snow Leopard Trust of Seattle to purchase stuffed leopards to sell to the fans (and raise awareness of the plight of the snow leopard) - I did this to honor the release of the song 'The Man Who Stole a Leopard' when it was released in 2010.  DDTTRH petitioned Glee to sing a Duran Duran song and they did (we never got a letter from them, but I don't think the timing of our petition and the song being sung was coincedence).  There was also a Mercury News mock vote for the band as well...and thanks to our effort and the help of Duran Duran, we won the vote.  Finally, while the effort is not focused any longer strictly on the Rock Hall (in fact we took the 'Rock Hall' out of our name for various reasons), the effort catapulted us into what is now Duran Duran Worldwide...a worldwide fan outreach that is focused on bringing fans together, and on raising awareness to Duran Duran and their impact on culture and popular media.  One of our current projects (which is very exciting) is to find references to Duran Duran in books, TV shows, and movies from the year 2000 on.  The idea is to prove that Duran Duran is not 'just a band from the 80's', but they are something special that has had an indelible impression on our society.  Eventually, I plan to release a book that chronicles this, which is also an exciting undertaking.

What I did and things I learned...

• I started with a commitment to something.  Having a commitment and a vision was important as things developed out.
• I was never afraid to experiment...and I did regularly.  I did many promotions, did Facebook analytics to improve engagement with our site, and continually engaged directly with fans.
• I had a specific target audience on the outset, but was not afraid to expand that audience as we grew and awareness was raised.  For example, the effort started mainly in US and Canada, but expanded its outreach to Latin America and Europe as well.
• It was important for me to realize at an early stage that I couldn't do a grand effort like this on my own.  I was able to recruit like-minded and talented people to the effort.  We had someone helping with our tech, an editor, publicists, someone who did our interviews, someone who focused on our petition, city representatives that worked to spread the word where they lived, a video creator and editor, artists who made our logos.  These people were committed and also became good friends.
• I started by creating a Facebook page, but soon realized that other online entities and ways to communicate with our fan base were needed.  We had a blogger site for our articles.  We had a Twitter site to share quick news.  And it wasn't just important using these sites...learning how to best use them given their limitations and strengths also was critical.  Social media builds and feeds on itself.  The tools you use facilitate that feeding.  
• I learned by making mistakes.  Especially when it came to celebrities.  Have a publicist saved me several times.  Going around the publicist and doing things the way I thought was best never worked and wound up making the effort look bad.  There is something to be said for a good reputation...especially when it comes to social media and outreach.  A good reputation can really take you places.  A bad one can really shut doors and make things sour.
• There are crazy people and trolls out there.  In dealing with them, I found it best not to give them too much attention and to distance yourself from them as much as possible.  Engaging them only feeds them and makes you look insecure.  I could tell you stories...
• We had fun.  And, we learned what the fans of the effort found fun and we geared ourselves towards that.
• It was helpful that friends of the band and many very close to them supported my effort.  Several of these people became good friends as well.  When I felt down about the effort and wanted to give up, they were there to encourage me on.  It was nice.
• One crucial thing I learned is to give credit to where it's due.  This was especially important as things got more expansive.  To that end, here are some special mentions...

Special mentions...

+Jason Henry:  Our first staff member from Canada
+Nikki Glista:  Also a big help with promoting in Canada
+Christian Helwig:  Our amazing tech dude
+Josie Beaudoin, our illustrious editor
+Kandice Purpura:  The Petition Duchess
+Anitra Delorenzo (aka Ladyaslan):  Did our early artist and band interviews
+Kitty Page and Sharmila Tredger:  Helped as publicists
+Jake Crawford:  Our video man from Down Under
+Nathan Stack:  From the band Autohypnosis, did a major part of our petition write-up
+Katy Krassner:  One of the band's media representatives.  Although the band could not be directly involved with the Rock Hall effort, she enjoyed it when we made press and there was one time she was able to offer us signed posters of the band for one of our Duran Duran Worldwide promotions when that effort was launched.
+Nile Rodgers:  Because he signed our petition and was happy to do so
+M. Douglas Silverstein:  A producer and a huge fan of the band.  He interviewed with us and supported the effort.
+Nabeel Shahid and Nuno from Portugal:  Two of our artists.  Nabeel did our original logo, Nuno did the one after that.
+The Scissor Sisters:  Fans of Duran Duran and provided words of support
+Your Vegas:  Ladyaslan interviewed them and they were more than happy to provide their positive voice to the idea of Duran Duran and the Rock Hall
+Jen Chaney:  Washington Post writer.  We interviewed each other.  That was fun.
+Pranav Chandrasekhar:  Owner of Radio Creme Brulee and key in helping promote us
+Jeff Piorworski:  For giving us the press in Cleveland - home of the Rock Hall
My family who became my first fans and support

DDTTRH Wiki:
https://duranduran.fandom.com/wiki/Duran_Duran_to_the_Rock_and_Roll_Hall_of_Fame

Press links:

DDTTRH specific:
Barrie Examiner:
http://www.stcatharinesstandard.ca/2010/05/15/duran-duran-fan-with-a-plan

Sun Press, Cleveland: article: http://www.cleveland.com/sun/all/index.ssf/2010/09/university_heights_duran_duran.html

Washington Post:
https://www.washingtonpost.com/blogs/celebritology/post/rock-and-roll-hall-of-fame-the-effort-to-get-duran-duran-and-other-snubbed-bands-inducted/2011/09/28/gIQABdxy4K_blog.html?utm_term=.033e24f6cc88

Yahoo News:
https://www.yahoo.com/news/blogs/cleveland/two-cleveland-fans-boost-support-duran-duran-weird-182634444.html

Related:
https://www.mercurynews.com/2017/07/08/vote-now-does-duran-duran-belong-in-the-rock-and-roll-hall-of-fame/

https://edgeinducedcohesion.blog/2017/05/01/why-arent-they-in-the-rock-roll-hall-of-fame-duran-duran/

http://www.goldminemag.com/blogs/duran-duran-the-cure-for-what-ails-rock-hall-of-fame

https://books.google.com/books?id=B8BLCAAAQBAJ&pg=PT292&lpg=PT292&dq=rock+hall+petition+duran+duran&source=bl&ots=mxAuV9XnZf&sig=AriwLeuwFKI2QrgNOaZ7NuK_z0E&hl=en&sa=X&ved=0ahUKEwicmvPasZbZAhUKVK0KHe3xAw0Q6AEIWTAJ#v=onepage&q=rock%20hall%20petition%20duran%20duran&f=false

http://www.futurerocklegends.com/artist.php?artist_id=duran_duran

https://www.stereogum.com/1000982/the-11-biggest-rock-roll-hall-of-fame-snubs/franchises/list/attachment/duran-duran/

https://www.iheart.com/content/2017-10-05-20-bands-snubbed-by-the-rock-and-roll-hall-of-fame/

https://www.yahoo.com/music/the-pressures-off-duran-duran-dont-need-the-153052592.html

https://www.mercurynews.com/2013/04/02/rock-and-roll-hall-of-fame-duran-duran-kiss-the-cars-depeche-mode-make-the-final-four/

https://www.mercurynews.com/2013/04/02/rock-and-roll-hall-of-fame-kiss-vs-duran-duran/

http://www.metroweekly.com/2015/04/top-25-artists-that-belong-in-the-rock-roll-hall-of-fame/

http://www.kissfaq.com/forum11/viewtopic.php?t=85793&p=1829972

Our original logo on a t-shirt (Artist:  Nabeel Shahid):

A banner we used with our newest logo by Nuno:

Some banners used on our sites:

Our newest Duran Duran Worldwide logo:

That signed Red Carpet Massacre poster we gave away as a prize:

Know what you know

 by Kirk Harrington, for SAEG (The Statistical Analyst Effectiveness Group)

Something that has helped me as an effective analyst is to keep a record of trends, insights, and key analytic observations I have had over the time in my career.  Keeping a record of this can be helpful to 1)  remind yourself of things you learned in the past 2) provide talking points when you are interviewing for your next position, and/or 3) be helpful to other analysts you work with or that are in your same field.  As an exercise, I have done this based on my experience and share that with you here. My background as a modeler and analyst has been in Credit Risk, Marketing, and Asset/Liability Management.  Here are some examples of things I have learned over the years from working with these three areas:

Credit Risk

On scoring models
Scoring models should be regularly checked to ensure their output is reasonable.  I once discovered a model was creating appraisal values higher than expected.  This knowledge came from line by line proofreading of code and comparing output of the model to actual appraisal values.  In looking through the code, I found several errors that led to inaccurate predictions and brought those to the attention of the vendor to improve the model's ability to create more accurate predictions.

Multivariate modeling lends to more flexibility
Roll rate and forecasting default dollars based on money movements (alone) throughout the year makes the assumption that dollars move the same way as in the previous year.  Modeling these behaviors (say using logistic, linear, or decision tree methods), however, creates more flexible  forecasts which are based on the underlying populations and (if included in the model) could include the economic environment, credit quality, and pricing environments relevant to these populations.  Simple forecasts can be useful with populations whose dependent variable does not shift much over time and the populations are relatively stable (i.e. not diverse, for example..if most of the population consists of a conservative high FICO customer).  These forecasts will break down however when this is not the case.

 The balance between risk and profitability
Marketing and Credit Risk working together is essential to ensure offers strike a balance between risk and profitability.  Here is a diagram which illustrates this:

If the circle depicted represents the path of a consumer and how they affect an organization, this diagram illustrates that the more risk is taken (more weight of decisions on the risk side), the higher the potential profitability.  There can come a point though where greater risk leads to risk loss (associated with events like default, foreclosure, loss of credit quality).    The diagram also shows that the less risk is taken (movement of the fulcrum up on the left, down on the right), the more this leads to the zone of profitability loss and potential cash flows.  Its important to state though that this diagram inherently assumes that an organization engages in risk-based pricing, which prices customers higher the more risky they are found to be (a positive correlation between interest rate and their presumed credit risk).

 Zones leading to a loss event
Three factors which can be examined leading up to a loss event are credit quality, line utilization (in the case of open ended credit products like equity lines and credit cards), and payment behavior.  Models which predict a credit loss event can be gauged at different periods around the event.  The closer to the event, the more deterioration is evident in all three of these factors;  whereas the farther from the event, the less deterioration there is with these three factors.  Illustrating this on a time diagram is useful:

What this shows is that the closer to a loss event, the more obvious the deterioration would be in factors leading up to the loss event.  For the factors I mentioned, deterioration of credit quality, payment behavior, and line utilization are more evident.  The farther you go back in time the more deterioration is sporadic and less evident.  Models which are looking for 'clues' to a potential future loss event and that are trying to identify customers that could use help to avoid moving to the zone of obvious deterioration could be built around the zone of sporadic deterioration (i.e. Triggers models, models to place consumers in relief call groups).  Models which predict the inevitable loss event and make allowances for loss (i.e. ALLL models) would be built on the zone of obvious deterioration.  This zone is characterized by consumers that have reached a point of 'no return' in terms of their deterioration in loss factors...loss is inevitable and the chance of loss is highest.

Marketing

 On control groups
Control groups created during experimental design need only be created with the underlying populations in mind, not with treatments in mind.  To illustrate what I mean, consider the following diagram:

The diagram on the left assumes a control is needed for each treatment group.  This would only be needed if the population for each treatment is different (typically not the case).  The diagram on the right shows a more appropriate design, when the population for treatments A-C are from the same population.  While this sounds basic enough, I have seen designs done by treatment group.  This approach complicates the design and makes it harder to work with for tracking and measurement purposes on the back-end.

On finding consumers who would open a check card
The check card (aka debit card) is a popular way that consumers pay for transactions.  This card typically carries a Visa or Mastercard logo and is directly tied to a consumer's checking account.  To determine which type of customer is likely to open a check card (say in the case of trying to create a marketing campaign to encourage people to open check cards), I have found looking at check writing behavior and use of online banking useful.  In studies I have done, the more people write checks (and the higher number of checks they write) and they less they use online banking, the less likely they are to open a check card.  Further, the less checks they write and the more they use online banking, the more likely they are to open a check card.  This implies that the check card (in the views of consumers that don't have them and newly open them) is considered a useful technology to improve their transaction experience.

On striking a balance between response rate and approval rate
When creating marketing campaigns for new credit products, consideration should be given to striking a balance between response rate and approval rate.  Typically, consumers with lower credit scores, who are riskier as a whole, will seek more credit.  This 'seeking' will increase their chances of applying for a credit offer that is sent to them.  This will in turn increase the response rate.  However, while this is so, it does not mean (unless in the case of a pre-approved offer) that they will be approved for the offer and the credit product will make it to the financial institution's books.  This implies finding a group of consumer likely to apply AND likely to be approved once they apply.  If creating a model, the dependent variable could thus be defined as someone that not only responds, but is approved after they respond.

Find the 'warm' prospect
Prospect Marketing can be tough.  It is usually associated with lower response rate, higher cost per account, and more difficulties in finding the population highly likely to respond.  What makes it even more difficult is knowing less about a prospect before you mail or advertise to them, which makes modeling on the population a further challenge.  One population that I discovered in my prospect marketing work that was found to be highly likely to respond is what I like to call a 'warm' prospect.  This kind of prospect already knows something about (or has had some experience with) the product or service that is being offered.  One group of these populations can be found with prospects that live at the same address of current customers.  These prospects likely have heard of the product before through interactions with the customer at their residence (and hopefully interactions were positive).  To find these 'warm' prospects, I have created a match key containing the address and comparing the match key to addresses on record for current customers.  Care must be taken to ensure that the match keys are standardized similarly between the two sources so that the process of flagging these prospects is accurate.

On measuring 'new money effect' for a deposit campaign
Let me start by saying that this is by no means a perfect way of measuring new money effect, however it will get you close to what needs to be measured.  If the campaign is for current deposit customers (i.e. they already have a checking, savings and/or certificate of deposit account) and the purpose is to increase 'new money' to the bank, how do you measure it?  One way I found to do so is to first group products and customers by household first.  Household relationships (based on how I define them or have seen them defined) share the following characteristics:  1)  Customer is at the same address as someone in the same household (it is assumed that money is shared among that household, and 2)  Customer is on the same account as another customer (irregardless of address).

Once customers are placed into a given household, associated products and account balances should then be appended to these households.  For each household, total deposit balance should be calculated.  This is appended at two different points in time based on the marketing event that occurred.   To illustrate...

Once Total balance is appended to each time period, calculate change in total balance by taking Total Deposit Balance at T2 - (minus) Total Deposit Balance at T1.  Then, if the campaign was set up using experimental design (test and control group) you can perform a hypothesis test of the average change in total balance of the test group vs. the control group.  If the test group change is significantly different than the control group change, there has been some new money effect to the campaign.  Its important to recognize that this effect can be positive or negative (calculated as the different between the average changes).  To calculate the estimated total effect, take this difference multiplied by the number of incremental (or test less control) 'new' accounts times this average balance difference.

Other considerations:  As I mentioned, this method is not perfect in measuring new money effect, but it can be useful.  Any results presented should be stated as 'estimated', not exact.  Further, it is possible to have customers switch households from one period to another.  To adjust for this in measuring the effect, I use only households that have not changed household key from one period to the other.  Also, in measuring average differences, it is wise to check and adjust for outliers prior to performing any hypothesis testing.  This will provide a more refined result in the end.

Asset and Liability Management

Modeling on a down rate environment creates problematic forecasts in an up rate environment
I once validated a prepay model that was built during a down rate environment, specifically during the period leading up to the most recent mortgage crisis and during the crisis.  The reason more time periods could not be included was because of availability of data.  By the time the model was complete, the interest rates had already begun to rise for some time since.  When asked for predictions vs. actuals reports on a more current environment, it was noticed that there was an over-prediction of prepay effect on the more current population.  This was due to the fact that the model had not included a period of rising rates (to balance out the coefficient estimates).  The question is, how was this problem of over-prediction dealt with?  One could not simply create more data (since it wasn't available) and it was not in the company's best interest to wait several more years to rewrite the model to include an up rate environment.  The solution used (which I consulted on to create) was a prepay multiplier to deflate the over-stated prepay rate.  This multiplier was used against any model predictions so that they could be brought more in line with current rate environment prepay behavior.

~~~~~~~~ end of examples

In conclusion, these are just a few of the examples of things I've learned while working in my career.  As evidenced, these matters are a product of trial and experimentation and come through experience working with specific data for specific purpose.  Further, these findings will represent either innovations in your field or knowledge that is supported by general trends.  Irregardless, findings you write down should be important to you and important to company and industry you serve.