Essentials of data science for marketers

What is data-analytical thinking?

Algorithms show you the fastest route to work, tell you when to go to bed, which books to read and soon they’ll drive you to your hobbies as well.

We already live in a data-driven world.

As a marketer, your choice is between getting driven by data or learning how to use it to drive smarter decisions.

Data-analytical thinking is learning how to develop solutions that take advantage of data.

Data-analytical thinking is a skill that you can master. A skill that will ultimately help you make smarter decisions by using data and data analysis correctly.

Is data-analytical thinking an essential skill for marketers?

You bet it is.

The same way digital tools and platforms revolutionised marketing, the availability of data and data analysis tools is creating new roles while leadings others towards redundancy.

Marketing is a complex and specialised skill and for some marketers, the idea of learning data science can be quite daunting. Terms like Hadoop, MongoDB, R, SQL, TensorFlow and PyTorch, just to name a few, are enough to make anyone’s head spin.

Rest assured that mastering data-analytical thinking isn’t the same as becoming a data scientist.

For a marketer, regardless of where you are in your career, learning to think critically about data and the tools available for extracting value from it can be both professionally and financially rewarding.

You’re already responsible for data

As a marketer, chances are that you’re either a part of a team or lead one where you’re expected to deliver data-driven marketing campaigns, understand how different algorithms impact performance and report on how your campaigns perform against specific metrics.

If you’re lucky you also have a data analyst on the team or tools that offer data analysis and dashboards that help you keep up with your campaigns and marketing investments.

Take a good look around yourself; you’re already responsible for using data effectively.

Back in 2011 McKinsey predicted that there will be a shortage of necessary talent required to take advantage of big data [Full PDF Report]. In 2019 that shortage is a reality and it’s not limited to data scientists alone but also business managers who are skilled at using data for making better decisions.

Working with data science teams

I often hear marketers explain: ‘… but we already have a team of data analysts, what’s the benefit in me learning how to do their job?’

You’re right to evaluate the benefit of investing your time in learning a new skill versus finding an expert in data analysis. I give you that.

Undoubtedly your resident data scientists are experts in their field but are you selling yourself too short?

As a marketer, you’re still an expert in your domain, your customers, your products and your particular brand of business problems. Knowing how to translate a business problem into a data science challenge is the key to working more effectively with your data science team.

The same report from McKinsey also argues that while businesses can have a central data science team to support multiple functions, to get any real benefit from data science, managers within those functions must understand how to leverage data properly.

Essential data science concepts for marketers

You don’t have to be a Data Scientist to benefit from data analysis. Undoubtedly, you might need an expert to help you solve complex statistic and mathematical problems.

As a marketer, you can save a lot of time and effort by learning how to correctly frame a problem, understand the basic terms and techniques as well as the application of the most commonly used data science tools.

Let’s get started with the basic terms.

What is machine learning & data science?

In traditional programming, a user (programmer) writes a specific set of rules or instructions (logic) to define how a program should behave.

For example, you can create an email spam filter by creating a logic that filters emails from a specific sender or messages with certain subject lines.

In Machine Learning (ML) instead of providing strict logical instructions, algorithms are used to teach a program about both inputs and outputs.

For example, to create the same email spam filter, the algorithm is given a list of emails that have been tagged as spam and not spam. Using this supervised data the algorithm creates the required set of instructions (logic) all by itself to filter out spam.

Machine Learning can be both supervised and unsupervised. (We’ll cover the difference in a bit.)

Data Science is the knowledge of extracting knowledge from data using techniques derived from Computer Sciences, Statistics, Mathematics as well as specific domain level expertise such as Finance, Marketing or Engineering.

As a discipline Data Science includes Data Collection, Machine Learning and organising data to exact knowledge required to solve a specific problem.

What is data mining and data analysis?

Data Mining is the process of discovering patterns in data that data analysis wouldn’t normally uncover. Data mining involves using specific methods from Machine Learning and Statistical Models to improve knowledge discovery.

Data Analysis is the processes of ordering and organising data in order to discover useful insights from available data. Data analysis is an umbrella term that includes data mining. Data analysis tasks use techniques from computer sciences, statistics, mathematics, domain-specific knowledge as well as machine learning.

What is supervised and un-supervised machine learning?

In Supervised Machine Learning, the algorithm learns from data (known as Training Data) that already contains the ‘right’ answers.

In Unsupervised Machine Learning, you do not provide any answers. Instead, the algorithms work on their own to discover the best way to find the right answer.

Both methods of Machine Learning require expertise in Data Science. However, Unsupervised Machine Learning is often more complex and the results can be more unpredictable.

What are dependent and independent variables?

In programming, a variable is a quantity that can change based on the instructions given to the programme.

In statistics and Data Science, there are two types of variables Dependent and Independent.

An independent variable is a value that is changed or controlled to measure its impact on a dependent variable.

For example, revenue from your e-commerce website is a dependent variable. Independent variable such as your online marketing budget, marketing channels, and promotions can be optimised to have higher or lower revenue.

Fundamental methods of data science for marketers

Now let’s dive into the fundamental methods and statistical models most commonly used in Data Science to solve business problems.

I hope you didn’t skip the last section because understanding the basic terms will make it easier to follow along.

Classification

Classification is a statistical technique used to predict a class or a group for an individual when you have two groups or segments available.

Classification algorithms use training data. In this case, training data can be a record of known business and private customers and the products they’ve purchased in the past.

Using the known customer types available in the training data, a classification algorithm can segment unknown customers into the two customer segments i.e. Business and Private Customers.

Classes or segments used in Binary Classification, as in the example above, are often mutually exclusive. Either you belong to one group or the other but never both simultaneously.

What if you’re interested in figuring out the likelihood of a customer being a Business or a Private Consumer? That’s where Class Probability Estimation Models come in.

Class probability estimation or scoring

Class Probability Estimation or Scoring Models instead of a classification predict the likelihood (quantified value or a probability) that a customer belongs to one or more classes.

Classification and Class Probability Estimation are very closely related. Often a model used to do one can be modified to do the other.

The important thing to remember is the key difference between the two. Classification places an object into a group and Scoring estimates the likelihood of an object belonging in a group.

Regression

Regression is a statistical measurement used for discovering cause and effect relationship between one dependent variable and one or more independent variables.

Regression is the most commonly used method for forecasting e.g. using Regression you can answer questions like, ‘which independent factors such as price, features or seasonality have the highest impact on an independent variable such as sales?’

A well-fitting regression model can predict outcomes that closely resemble observable results.

Causal modeling

Causal Modeling is a statistical technique used to understand which events are really influenced by certain actions.

To some Causation and Regression modelling might sound similar but they’re not.

Regression models discover dependence between variables present within a model but can’t always imply causation. For example, crop yield might increase with the amount of rainfall. While there’s a relationship between the two, still the yield itself doesn’t affect how much it rains in a year.

An important rule to remember is that in statistics, a relationship doesn’t imply causation.

Similarity matching

Similarity Matching is a data science technique used to identify objects within a dataset that match the known qualities of other objects based on known data.

Training data sets created through similarity matching can support regression and classification tasks.

Clustering

Clustering, on the other hand, is used to discover natural groups within a dataset without using any known attributes.

Co-occurrence grouping

Co-occurrence Grouping also known as Market-Basket Analysis tries to find associations between objects based on transactions e.g. a purchase.

Co-occurrence Grouping and clustering have one major difference. While clustering groups objects together based on shared qualities, Co-occurrence Grouping groups objects that were present in the same transactions.

Profiling

Profiling is used to characterise the typical behaviour of a person, group or a population. It’s often used to establish norms that can be generalised to describe the behaviour of a population, a small group or an individual.

Link prediction

Link Prediction is a method used to predict the presence and strength of the connection between objects. Social networking and e-commerce website use link prediction to recommend new connections and products.

Appreciating your creativity & intuition

Data-Analytical Thinking doesn’t replace creativity. In fact, using any analytical tools for problem-solving requires appreciating your own creativity and intuition.

This might sound counterintuitive to some since data, especially in marketing, is often used to counter the rampant use of intuition.

Learning to appreciate the value of your own intuition means taking the time to explore a problem. Tap into your intuition to better understand the challenge you’re dealing with and think creatively to break large problems into simpler steps.

Learning how to do this every day is perhaps the most important skill for mastering data analytical thinking. If you’re new to structured thinking, I recommend taking the time to explore Structured Problem Solving or the famous McKinsey’s 8-Step Framework for it.