The Definitive Guide for Creating an Analytics Team

Analytics in development, higher education and in general

What's In This Guide?

This is the definitive guide on everything you wanted to know about starting an analytics team. The guide gives answers to questions such as "Do I need an analytics team?", "How to build a case for an analytics team?", "What type of people do I need to hire?"

This guide has information that you will not find time to research or invest resources to evaluate.

I have started my own teams and hired many people, sometimes wrong ones. Benefit from my mistakes. Save your time and money. And let me walk you through this process on how to get started in analytics.

Who is this guide for?

Although this guide is directed at decision-makers in the advancement or fundraising field, you will find value in this guide even if you are beginner to analytics.

Ashutosh R. Nandeshwar, PhD

Who am I?

Author of Tableau Data Visualization Cookbook. An award winning keynote speaker.  

I'm one of the few analytics professionals in the higher education industry who has developed analytical solutions for all stages of the student life cycle (from recruitment to giving).

I love, love, love speaking about the power of data, as well as ranting about data professionals who chase after “interesting” things.

I earned PhD/MS from West Virginia University and BEng from Nagpur University, all in industrial engineering. Currently, I'm leading the prospect development, reporting, and data science efforts at the University of Southern California.

What People Say About the Guide!

Amazing read!

Amazing read! ... The field of prospect research and advancement analytics is fortunate to have such a generous thought leader and motivator in Dr. Ashutosh Nandeshwar.

Josef Castaneda-Liles, Ph.D Senior Associate Director of Prospect Research
user-image

Wisdom

This report shares the wisdom of someone who has built successful analytics teams firsthand. In simple language, it reveals the minimal skills necessary to harvest the value of data while maintaining focus on the true prize: actionable intelligence.

Brett Lantz Bestselling author of the book Machine Learning with R
user-image

Highly recommend

I highly recommend this report if you are considering any sort of formal analytics at your organization. It contains both critical questions in designing a team and concrete tools to help a team succeed. We have used his recommendations and appreciate his expertise in both non-profit fundraising and analytics.

Susan Hayes-McQueen Director of Prospect Management, Research, and Analytics
user-image
Analytics in development, higher education and in general
The Definitive Guide for Creating an Analytics Team

Don’t have time to read the whole guide right now?

No worries. Let me send you a copy so you can read it when it’s convenient for you. Just let me know where to send it (takes 5 seconds):

Introduction

If you are reading this, most likely you heard from someone or you read somewhere that analytics can do wonders for you, that you magically will boost your appeal responses, that you will find many major donors, and that you will be a rock star!

Although there is some truth in these statements, do not conclude that you need an analytics team or office.

Performing quality analysis is a hard task, but most organizations fail in its execution. Most people are reluctant to change and selling analytics to such people is very difficult. 

In this guide, you will find information to help you decide whether you need an analytics office.

Even with the hype of big data, most companies struggle with the use of analytics. This Harvard Business Review (HBR) article provides some advice on how to achieve success.

Quick intro to analytics

This field has many names: statistical analysis, quantitative analysis, data mining, machine learning, data analytics, business intelligence and data science.

While some names have fallen out of favor, some are trending. Regardless of the size of the data, the common objective in all these fields is to learn something from the data.

It requires grit and skill, however, to learn something useful and actionable.

In recent years, the data has grown so rapidly that it has become unmanageable. Plus, management leaders and data professionals have realized the derived value of such data.

Netflix is a good example of this fact. By learning our interests, it predicts the movies we like and keeps us engaged, generating more revenue for the company. On its technical blog, Netflix says that "75% of what people watch is from some sort of recommendation." Can you imagine our good, old TV engaging us like this?

Growth of the available data and increased need of knowledge from such data have given birth to specialized tools to manage and store the data as well as to learn from it.

You won't find these specialized tools, such as Hadoop, Hive, Mahout, and other disease/animal sounding tools in this guide. You can extend the principles and methods discussed in this report to learn and use such "big data" tools.

When do you need an office?

As a believer in discovering insights from data, I am biased: I believe that data-driven decisions will make you and your organization more effective.

The question is not whether you should use data or analytics, but which insights would you find most applicable.

After seeing countless news articles on big data, you may find it easy to believe that you need data scientists or that you need infographics. For simple aggregation of data or calculation of descriptive statistics, you just need an introduction to statistics class.

To do, what I consider, real analytics, you need advanced knowledge of analytical concepts and more importantly you need heightened judgment to reject questions that yield "nteresting'' yet unactionable results.

These skills come with practice and at a cost.

Research has shown that firms that use data in their decision making have 5-6% productivity higher than expected. 

In analytics, you need heightened judgment to reject questions that yield "interesting" yet unactionable results.

Click to Tweet

Analytics Maturity Model

Before you begin your journey on the analytics path, you must consider the current state of your data and organization's appetite for it.

If your data is in a bad shape due to inconsistent or irrelevant capturing of the data elements, you will have a tough time getting something meaningful out of it.

Your first priority, in such cases, should be to shore up the data and build consistent data capturing practices. Once you start regularly capturing and retaining useful data, you can begin your journey on the analytics path.

  • Ad-hoc: In this stage, you keep data in spreadsheets; and every time you are asked to provide information, you have to spend a lot of time combining data sources.
  • Foundational: You probably have a Customer Relationship Management (CRM) system, supported by a well-designed database. Finding and providing information take less time. You frequently provide basic measurable activities.
  • Competitive: You are efficiently utilizing your database and reporting solutions. Decision-makers can access data and reports easily. You can accurately measure and report past activities answering questions like ``what happened?''.
  • Differentiating: You are generating forward looking information. The decision-makers rely on this type of information to plan for the future. You are able to answer questions like ``what will happen?'' and ``why did X happen?''
  • Breakaway: You are helping automate decision-making and/or generating real-time information. The decision-makers have latest information readily available along with recommendations for next steps. You are able to answer questions like ``what should we do?''
IBM analytics maturity model shows various levels of analytical maturity. creating an analytics team.

IBM analytics maturity model shows various levels of analytical maturity.

Although you should always attempt to increase the data usage in your decision making, you need to assess whether you need an analytics team.

If you work for a smaller organization, say an organization with less than 1000 prospects, yes, you can improve your solicitation success rates or increase your retention rates using analytics. The costs of doing so, however, using a full-time analyst could be higher than any efficiency you gained.

For example, let's say the cost of an appeal is $5 per piece and you send it to 1,000 people.

Your cost of that appeal is $5 X 1000 = $5000.

Your analyst develops a model that predicts people likely to respond. Based on that model, you choose to send the appeal to 600 people.

Your cost of sending the mail now is $5 X 600 = $3000, a total savings of $2000.

Not bad.

But if these savings came at a cost of a $55,000 employee or a $15,000 model purchased from a vendor, you did not even break even on the money you spent for the modeling.

The true value of an analytics person is less in what models he can develop, but more in his critical thinking. Any analyst should be able to solve a given problem, but good analysts will ask the question nobody has asked before and provide new solutions to previously undiscovered problems.

And that ability is worth acquiring.

The true value of an analytics person is less in what models he can develop, but more in his critical thinking.

Click to Tweet

What will analytics do for you

After reading the first section, you may ask, ``But what about Moneyball?'' or ``What about all the news how `big data' is going to save the world?''

If you cut through the hype and find some problems in your organization worth solving, yes, analytics can indeed add a lot of value. Let's look at an example of how analytics can add value.

I am sick of all the hype around `big data'. Take this article for example. Everyday you see an article like this. Yes, data has the power to solve various problems, but didn't we always use that power? What is hard is finding insights and applying those insights to problems.

Lift is a common way to measure this value. Lift measures the improvement achieved by a predictive model over the standard, baseline approach.

Let's say your annual appeal has a response rate of 8%, but then you send your appeal to only a selected population using a predictive model. From this selected population, 16% people respond, thus giving you a lift of: 16% /8% = 2.

Another way to look at it is by creating a cumulative gains chart. This chart shows the response rates using traditional methods of selecting who to contact and the response rates using predictive models. Most likely, you will see improvements over your baseline methods.

For example, in this chart you see that if you contact 20% of population, using your baseline method, you will get a response of 20%. With a predictive model, if you contact 20% of population, you will get a response of 50%. You gain efficiency by contacting people likely to respond as predicted by the models.

Life curve shows the improvement created by an analytics model over a baseline.

If applied thoughtfully, insights generated using analytics can help your organization to do all of these better:

  • find more prospects
  • build a stronger prospect base
  • retain more donors
  • increase giving
  • manage the right prospects
  • recommend giving options to your donors
  • find volunteers
  • recruit gift officers
  • invite people to events
  • `up-sell'' online giving
  • create stewardship articles
  • staff the right geographic regions
  • assess campaign readiness
  • scale prospect research
  • measure performance

How to build a case for an analytics team

The best way to build a case for an analytics team is to report on the Return on Investment (ROI) on analytics as applied to your organization's existing problems. 

Are there any problems that worry you about the future of your organization?

Problems such as

  • how to retain donors,
  • how to find new donors,
  • how to increase giving,
  • how to increase participation rates,
  • how to focus your gift officers' efforts,
  • how to provide timely information on your prospects to your staff or
  • how to know that you have enough prospects to reach a campaign goal?

Here is one approach to build the case.

Create a table with five columns.

In the first column, list all questions that worry you about the future of your organization. Don't put the lack of your promotion as a worry. Numbers will always be against you! 

In the second column, record your thoughts about solving those problems.

Take a break.

Go over the list again and add any other information that you can think of.

In the third column, make notes of any problems that you think can be solved by data-driven decision making.

In the fourth column, make notes of any outcomes such as improved processes, saved time, or increased giving.

In the last column, enter estimated savings or earnings.

Once you complete the list, you may find that an outcome of fixing a single problem could be worth thousands, if not millions of dollars, to you. If that is the case, congratulations! You just built a strong case for your analytics team.

Summarize all the outcomes, provide the estimated dollar amounts in savings or new income, and present the findings to your management team. When leaders see significant risks or opportunities, they are more likely to invest and support the idea.

When leaders see significant risks or opportunities, they are more likely to invest and support the idea.

Click to Tweet

Problems

Solution

Using Data

Outcomes

Amount

acquire more prospects

purchasing lists

yes, we can purchase lists based on profiles of our existing donors

1,000 more prospects in the database

$5,000

increase retention

learn interests of donors

yes, we can build a dataset of donor interests

improved response and retention rates

$10,000

shortage of donors

find new markets

yes, using geographic modeling, we can find new regions

new opportunities

$500,000

If your leadership is unwilling to invest in a new team or you just don't have resources to do so -- though you have a solid case -- start doing.

Doing is far more powerful than saying.

Lead with an example.

Tackle a reasonable problem and generate an analytical solution to it.

Show the results and projected outcomes to a potential consumer of your information. Be very picky about choosing your first consumer. This consumer should be your champion and should be able to communicate the power of analytics to other people, including your leadership.

Differentiate actionable from "interesting"

It is very easy to think that you can apply analytics to every problem -- true, yes, you can -- but the bigger challenge is separating ``interesting'' from actionable.

For example, social network analysis is quite cool and you may apply it to your data to find network graphs. Yes, the network graphs look good and interesting; how to put them to action, however, is a bigger challenge. 

LinkedIn has one of the largest relationships datasets in the world. Even LinkedIn put a stop to its network mapping tool in September 2014. These maps did not add anything to our knowledge.

That is why it is important to think first of the biggest problems or questions that your organization is facing. By solving these problems, you could provide a new direction for your organization.

If you think of solutions before the problems, forget implementing; you will have a hard time creating a buy-in and ``selling'' your solutions.

If you have read this far, I assume that you want to build this analytics team, and I anticipate your next question might be: "What type of people do I hire for such positions?''

I consider the following qualities, which make up the mindset of analysts, critical for the success of such a team.

Mindset

Curiosity

Some of the world's biggest inventions happened because someone was curious about something. It would be very nice if we could describe the problems with all the parameters to our analysts and then ask them to find solutions. You know this: it doesn't work that way. What worked in college or graduate school hardly works in the working world. In a school setting, you solve a given problem, whereas, in the working world, you interpret problems.

Your analysts first and foremost must be curious. Curious to ask questions, curious to wonder whether there is a better way of doing things, curious to find information, and curious to talk to people and understand their problems.

Balanced Skepticism

To succeed in this type of a role, one needs to have balanced skepticism toward existing practices, available data, current conclusions, and cultural biases.

I'm suggesting a careful and objective point of view toward everything and not becoming a ``devil's advocate'', which I think people use as a shield while bringing down other people's ideas..

Asciteauthor{o2013being} suggests in her book , ``A skeptic is someone who maintains a consistently inquisitive attitude toward facts, opinions, or (especially) beliefs stated as facts.''

Skepticism is further helpful in balancing the belief ``Data can solve every problem'' with ``I don't know whether data can support that question, but I will find out.''

Skepticism is further helpful in balancing the belief ``Data can solve every problem'' with ``I don't know whether data can support that question, but I will find out.''

Click to Tweet

Persistence

Real world data is messy. Cleaning and preparing such data takes a lot of time.

Some comment that 70% of any data analysis project is spent on data preparation.

When you add the learning curve, intricacies and sheer difficulties of using specialized tools, the whole process no doubt frustrates you.

Just then your underlying question changes, newer data becomes available, or you are asked for something completely different.

To survive through this and still succeed, one needs persistence, a lot of it. I have seen many talented professionals quit (not only quit projects, but quit their jobs) because they wanted quick results and did not persist through the messiness of our business.

Hunger to learn and improve

As data grows, tools available to gather, manipulate and analyze data are changing, too. It is hard to keep up with the newer technology.

Practitioners of data ``science,'' however, should willingly give up an inefficient tool for a better one.

Doing so means regularly reading and learning about the field and picking up relevant tools. A good analyst will separate herself from an ordinary analyst with such a mindset.

Continuous improvement of processes, tools, methods, and, most importantly, of oneself, should be the cornerstone of an analyst's mindset.

When you immerse yourself in similar fields and you constantly read and learn about such fields or industries, innovation happens.

When you neglect the other fields, you don't innovate, you simply repeat.

Continuous improvement of processes, tools, methods, and, most importantly, of oneself, should be the cornerstone of an analyst's mindset.

Click to Tweet

Motivation

Research shows that an intrinsically motivated person is better at learning than an extrinsically motivated person citep{ryan2000self}.

People who are intrinsically motivated enjoy their learning, have persistence and are creative compared to those who are extrinsically motivated.The extrinsically motivated expect and await rewards.

It is hard to keep an extrinsically motivated employee happy in a job requiring nimbleness, curiosity and continuous improvement.

Portfolio approach

While tinkering with data and developing various data products, a good approach is the portfolio approach. As Adam Grant, a Wharton Business School professor, explained in his talk, while looking for breakthrough ideas, it always is the best approach to generate lots of ideas and work on them simultaneously.

One of them is likely to be a winner.

Selling

Although you may disagree, it is a fact that at all times you are selling something. In every conversation, you are explaining your perspective or convincing others to accept your idea. Selling is critical when you want your users to take action on your insights and recommendations. They will not take an action if they can't trust your models, theories, or, worse, you.

You need to explain your processes using stories or analogies such that you don't need to hard sell, but they understand that you are solving their problems. Selling becomes easier if you clearly communicate that you are solving your users' problems and that you explain your methods without confusing your listeners.

Here's a thought for you, paraphrasing Wayne Gretzky: ``You waste 100% of your analysis that your readers don't take action on."'

You waste 100% of your analysis that your readers don't take action on.

Click to Tweet

Technical expertise

You may ask "Why did you emphasize mindset and softer characteristics first and technical expertise second?'' "Aren't technical skills more important than the softer skills?''

Yes, they are important because one would be unable to do his job if he didn't have the technical skills, but he would be unable to succeed if he didn't have the softer skills.

Plus, one can be trained in technical skills, but it is very hard to extrinsically cultivate the softer skills.

Areas of importance

Main areas importance in data science: data mining, data visualization, database management, programming and storytelling

Main areas of importance in data science.

Data mining/machine learning/statistics

Data mining, machine learning or applied statistics.

Whatever you call them, these skills are the foundation of analytics. Data mining is a general name for the process of finding patterns from data.

Machine learning is a field of computer science that focuses on using various pattern detection algorithms. Some machine learning algorithms are association rules, nearest neighbors, decision trees, random forests, Bayesian methods and neural networks. Some methods from applied statistics have also made their way into machine learning. Multiple linear regression, logistic regression, and Bayesian methods are the most used techniques from the applied statistics field.

Data visualization

In this infographics crazy world, it is easy to dismiss graphics. I know I do.

Bad data visualizations take up the whole space to describe a very few data points (think people, flags, buildings, exploding pie charts), whereas, good data visualizations get out of your way and actually show the underlying data (think tables, simple charts, patterns).

If carefully crafted, data visualizations can tell powerful stories. This NYT graphic is a good example of how to tell different stories using a data visualization.

The key is to avoid the trap of making them overly beautiful but hardly actionable.

I believe it was Noah Iliinsky, a data visualization expert, who said that "data visualizations are advertisements, and not art.''

Improve Data Visualization In As Quick As 5 Minutes With These 20+ Special Tips

Expert Advice To Create Data Visualization Like Pros


Your main objectives are: make the visualizations tell your story, let the data/patterns stand out, and not distract your reader.

If you follow the principles of effective data visualizations, you will more than likely make your visualizations actionable, yet good looking.

Data visualizations are advertisements, and not art - Noah Iliinsky @noahi

Click to Tweet

Database management

Of all the other processes in an analytics project, data gathering and manipulation takes the most time. If you are unable to get the required data in a structure suitable for analysis, you spend even more time manipulating the data.

Structured Query Language (SQL) is handy in such cases. Most likely your data is stored in a database management system such as Oracle or Microsoft SQL server. There are three things you must know to efficiently get the data out of such systems:

  1. Database structure: knowing which data elements are stored in which tables and how the tables are connected to each other
  2. Understanding the theory and principles of relational databases will help you get the required data faster and with accuracy.
  3. SQL: writing queries to access the desired data.?

Programming

You may complete various analytics projects without writing a single piece of code, but programming offers tools to become efficient. The other benefits of using a programming language are: reproducibility, repeatability, and readability. Reproducibility helps you track your steps when someone asks you how you got to a certain number. Repeatability helps you modify your process when someone asks you to make some changes to your analysis. Readability helps you and others to understand the logic of your analysis. Open source and free statistical and scientific programming languages such as R and Python are helpful in our analytics pursuit as both languages can use countless libraries on various topics. Plus, they both make data manipulation and analysis very easy.

Communication Skills/Storytelling

Imagine yourself speaking in front of the consumers of your analysis.

You want to describe how your predictive models performed. You can show them the "confusion matrix'' i.e. the errors and accuracy of your model.

Or,

you can describe a single person (and his characteristics) from your data and how those characteristics impact your model.

Which version do you think your audience will most likely understand, remember and trust?

I am willing to bet on the second one. Even the most serious scientists enjoy good stories.

I take huge inspiration from The Economist articles: these often start with a story of one person and later, describe a wider phenomenon with detailed statistics.

?

?Why Communication Skills Are Important?

Why are communication skills important? Why can’t we just be happy with great analysis?

Well, here’s the thing. Without communication of your analytics, your overall effectiveness is close to zero.

Jerry Allyne?? ????from Boeing talked about the overall effectiveness of analysis. He presented a venn diagram. In this venn diagram, we have two circles for: the quality of analysis and the acceptance of analysis.

The effectiveness of analysis is directly dependent on both of these i.e. Q times A = E.

Here’s how he talked about it.

Let’s give a maximum of 10 points to the quality and 10 points to acceptance. Then the highest effectiveness is 100. So, let’s say you build a fantastic model which we will give an 7, but it wasn’t accepted widely and we give it a 3. What’s the effectiveness? 7 times 3 = 21.

What if you went back and built even better model and now you got an 8, but the acceptance stayed the same. 8 times 3 = 24. but what if the acceptance was high let’s say 6 and your analysis was a 5.

How much effectiveness did you get? 6 times 5 = 30.

Effectiveness is dependent on quality and acceptance of analytics

Effectiveness of your Analysis = Quality of the Analysis * Acceptance of the Analysis

Think about your every analysis that you completed, that you worked hard on, but the reader just tossed it. What was the effectiveness of your analysis? Zero.

I find this very useful because it reminds me that it is not only about the analytics. Action must succeed analytics, for your success is defined by action.

Action must succeed analytics, for your success is defined by action.

Click to Tweet

Training Guide

Get an online version of the training guide to prepare you for successful analytics projects. Or,

Download this Free Guide to become a data scientist

Get this FREE self-study guide as a beautiful, high quality PDF.

Where to Find Talent

You may look for one person with all of the above skills, but you also may be able to build a team with complementary skills.

I would like to see organizations create another important position that I call the insights manager.

Although the analysts themselves can communicate the results to the stakeholders, we will see better results if we have a dedicated person to: work with the management team members, listen, ask questions, and formulate data questions for the analysts.

Once the analysts complete their analysis, the insights manager then builds a plan to put the analysis in action and makes sure that the analysis is used in decision making. This person frames the right questions, applies the analysis and internalizes that a mediocre analysis that is used is more effective than the excellent analysis that sits on a desk.

A mediocre analysis that is used is more effective than the excellent analysis that sits on a desk.

Click to Tweet

Now that you know the technical skills and the required mindset of a sound analytics team, the next logical question you may have is where to find this talent.

Two obvious choices are:

Grow the Talent

This could be a good choice if a person interested in this field is interested for the right reason i.e., not to make a quick buck, but to learn various tools and their uses.

If an employee, in your mind, has already passed all the mindset tests, it is quite easy to put her on the above training or ask her to complete related data science courses on coursera.

If you are lucky to work near universities, you could also look for interns who major in computer science or other quantitative fields.

Hire

Although hiring the almighty data scientist or the measly data analyst may seem like an obvious choice, both are quite hard to find, let alone hire. If you go about hiring, you could look at recent graduates from the applied statistics or analytics programs such as North Carolina State University's MS program or you could work with a recruiter specializing in analytics.

Most likely, your best hires are passive candidates who are already doing well in their current job. Wherever you find them, I recommend testing the technical skills and thinking/analytical skills of these candidates. The hardest qualities to test, I have learned, are perseverance, patience and hunger to learn. The following job description may help you create your own job posting.

Sample Job Description

We are looking for an experienced data analyst who enjoys working with messy datasets and finding patterns of business significance from them. An ideal candidate would have a graduate degree or equivalent coursework in a technical and quantitative field along with strong programming skills.

Job responsibilities

Data manipulation, enrichment and analysis (80%):

  • Manage, acquire, clean, and manipulate data to support analyses and reporting
  • Use machine learning and advanced statistical techniques to draw meaningful and actionable recommendations from various data sources
  • Use various software and tools (R, SQL, Weka, Python and others) to analyze and present the data analysis

Concept development and learning (10%):

  • Build and keep up with the knowledge of literature, practices and techniques in data science, business intelligence, and management communities
  • Develop product ideas and solutions to increase our operational efficiency.

Outreach (10%):

  • Promote data-driven culture
  • Build and share analytics expertise

Qualifications

  • Strong critical thinking and project management skills along with curiosity, passion for learning and balanced skepticism
  • Graduate degree or equivalent course work in a technical or quantitative areaitem
  • Strong computing and programming expertise

Conclusion

I hope this guide helped you think about the importance of analytics, whether you need a team, how to build a case for an analytics team, and what to look for when hiring. If you enjoyed reading this guide, please share this guide with your network.?

Analytics in development, higher education and in general
The Definitive Guide for Creating an Analytics Team

Don’t have time to read the whole guide right now?

No worries. Let me send you a copy so you can read it when it’s convenient for you. Just let me know where to send it (takes 5 seconds):