The Most Important Algorithms for Marketing Data Analysts to Understand

Algorithms matter little to the average consumer, working behind the scenes of the technology and social platforms they use every day (Facebook, Google, etc.). To marketers, though, they matter a great deal, playing the deciding role in how visible a brand’s content is to the target audience, how relevant the products recommended are to a prospect about to hit the check-out button, and a variety of other determinations that weigh heavily on the success or failure of a marketing strategy, as well as on the company’s bottom line.

With so many algorithms at work, marketing data analysts are faced with a sea of knowledge to acquire. Which algorithms matter most to today’s marketing data analysts? What algorithms should analysts focus on understanding that will offer the biggest benefit to their work? To learn more about the various algorithms that marketing data analysts are faced with today, and which among those algorithms analysts should focus on understanding, we asked a panel of marketing professionals and data analysts to answer this question:

“What are the most important algorithms for marketing data analysts to understand?”

Meet Our Panel of Marketing Pros and Data Analysts:

Brady Keller

Marisa South

Ken Gilbert

Swapnil Bhagwat

Eagan Heath

Ryan Barnes

Benjamin Cogburn

Gopi Suvanam

Gene Caballero

Monica Georgieff

Dan Roberge

Max Page

Paul Burke

Kean Graham

Khalid Saleh

Kyle Hearnshaw

Donna Duncan

Brandon Seymour

Justin Yek

Arsalan Jabbar

Flynn Zaiger

Linda Allen

Keep reading to discover what our experts had to say about the most important algorithms every marketing data analyst should master.

Brady Keller

@AtlanticNet

Brady Keller is the Digital Marketing Strategist for Atlantic.net, a trusted hosting solution for businesses seeking enterprise-class data centers.

“Data Analysts should be up to speed on…”

Google’s recent release of the Penguin 4.0 algorithm.

The result of this update is that Google is now devaluing bad links instead of penalizing sites. One of the big changes here is that updates are made in real time. As a result, marketers must understand that link acquisition is about quality, not quantity. Having ten high-quality links can be more valuable than having 1,000 low-quality links. The best way to achieve high-quality links is to create high value, relevant content that other website owners will want to link to. If you’re able to consistently create valuable assets for your web property, the links will follow.



Marisa South

@MarisaSouth

Marisa South is the General Manager of Vet & Pet Jobs, a career resource for employers and job seekers in the veterinary industry.

“2017 is upon us, and the most important thing that marketers need to understand is that…”

Google is steadily approaching a mobile-first world. Millennials are moving away from their desktops, and they’re searching on-the-go from their phones and tablets instead. Google recently announced that they would be implementing a mobile-first index, making it critically important to develop and maintain a mobile-optimized website. Their algorithms will now be focusing on mobile sites first, followed by desktop sites. So if you haven’t done it already, create a responsive design website, implement structured data in your mobile version, consider user experience and conversion optimization, and plan out your link building strategy. If marketers are able to prepare and implement these strategies correctly, Google’s mobile-first index should be welcomed.



Ken Gilbert

@UTKnoxville

Kenneth Gilbert, professor emeritus of the Department of Business Analytics & Statistics at the University of Tennessee’s Haslam College of Business, holds a Ph.D. Gilbert has published in Management Science, Decision Sciences, IIE Transactions and the Journal for the Society of Computing Machinery and has consulted for numerous companies.

“One important algorithm that marketing data analysts must understand is…”

Time series analysis is an important part of the marketing analysis toolkit. From time series models we can (1) glean information from the history of the process (2) have an accurate baseline to evaluate the impacts of promotions and other intentional interventions and (3) have the quantified data necessary for developing a supply chain strategy for dealing with variation.

Insights from Historical Data: The insights provided by historical data are of two types. The first is understanding the process that generates sales. For example, a by-product of fitting historical sales data to a time series model is a yearly profile of sales, e.g. 40% of the sales of this item occur in December. The second type of insight is provided by the anomalies that point to unique events that affect the process. These events are signaled when the outcome falls outside the range of variation predicted by the time series model. One amusing example of this is the spike in the sales of strawberry pop tarts that occurs in coastal cities in advance of hurricane.  Understanding these events allows marketers to better plan for similar events in the future. It also prevents marketers from mistaking these events for sustained changes in the level of demand.

Evaluating the Impact of Interventions:  In order to evaluate the impact of marketing changes, changes by the competition or changes in the market itself, we must first know the predicted range of variation of sales absent any change. Time series model provides us this predicted range. In order to say with confidence that an intervention had impact we must observe an outcome inconsistent with the range of outcome inconsistent with the range of outcome that could have happened with the status quo.

Developing a Strategy for Dealing with Variation in Demand: One of the biggest mistakes made in operations planning is to forecast, analyze and plan and operate as though the future is certain. The outcome is a constant mode of crisis management dealing with the inevitable fact that the outcome will differ from the forecast. Ultimately this strategy results in lost revenue increased operational costs.

Time series models of demand combined with analytics supply chain models provide an alternative approach for dealing with uncertainty. In the planning process these models provide a means of predicting the performance of a supply chain under various assumptions. This makes it possible to compare various sourcing strategies, production scheduling philosophies and distribution strategies. It allows marketers and operations managers to understand the value of real time point of sale information, flex capacity and strategically placed inventories.

On the other hand, it is necessary to respond quickly to true changes in the level of demand. Failure to do so allows the gap between production and demand to accumulate into overstocks or stock-outs. This not only drives up cost by causing bourgeoning swings in the level of inventories, productions and shipments.  It also drives down revenues by causing lost sales due to stock outs and markdowns due to stock outs. This phenomena is so common that it has a name, the bullwhip effect.



Swapnil Bhagwat

@swapnildigital

Swapnil Bhagwat is Senior Manager – Design & Digital Media and implementing marketing, social, content, digital, web and design strategies for the group companies. He is an MBA graduate with work experience in the U.S., U.K. and Europe. Swapnil has worked for more than a decade across a range of businesses for the global markets.

“The need for various types of algorithms…”

Such as search engine algorithms or social media algorithms, start with one basic premise, and that is to provide the most relevant and interesting content first. Be it Facebook’s
newsfeed algorithm or Google’s latest ones, they’re doing the basic job of balancing out the enormous amount of data produced every day and making sure that the most relevant information doesn’t go unnoticed.

This makes social algorithms such as Twitter’s ‘relevancy over recency’ algorithm, which boosts relevant content irrespective of it being organic or paid, quite important for marketing analysts to understand.



Eagan Heath

@GetFoundMadison

Eagan Heath owns and runs an online marketing agency in Madison, WI called Get Found Madison.

“I would suggest two algorithms from the major online players…”

  • Google’s suggested searches in their search box, searches related toat the bottom of the results page, and Google keyword planner suggestions. When doing keyword research for SEO or AdWords campaigns, it’s invaluable to know what keyword phrase variations people search regularly so that you can target those phrases with web page content and Google Ads.
  • Facebook’s lookalike audiences are a marketer’s dream. The tool findstargeted prospects in minutes rather than the weeks or months it might take a human to find thousands of similar consumers. For best results, try plugging in the addresses from your email list for Facebook to find even more potential subscribers.


Ryan Barnes

@Ryan_N_Barnes

Ryan Barnes holds a PhD in economics with a focus in econometrics. After completing his PhD, he founded Barnes Analytics. Barnes Analytics is a data analytics and machine learning consulting firm and data analysis training company.

“Marketing analysts need to have a lot in their tool belt…”

But nothing is more important than fundamentals. Most data analysts in the marketing field will be spending their time doing A/B tests, as such they need to really dig deep into t-tests. I know that they are kind of boring, but they were all the rage about 100 years ago. T-tests were designed to do A/B testing, so you absolutely need to understand this simple test really well. Everything else that you will do as a marketing data analyst is probably some variant of a t-test.

Regression analysis is the next. Again, this seems like it could be a bit of a yawn, when you could be talking about support vector machines, deep neural networks, random forests and so forth. However, the humble OLS regression is where these other algorithms started life. It also has several benefits that these other algorithms just don’t have, namely, interpretablity. Going back to an A/B test, you can perform a much more careful calculation of the effect that your campaign had using a regression. For example, you can control for age, gender, and other demographics so that you isolate the effect of your campaign using regression analysis. It is also very flexible, you can look at time-series, panels of individuals, and you can even do a naive version of classification.

That brings me to the last two algorithms, logistic regression and k-Nearest Neighbors. Logistic regression is useful when you want to spend marketing dollars more effectively. The goal here is to stack rank individuals to expose to some sort of campaign. This is incredibly important when you are doing something that is expensive, like sending mailers, you want to send those mailers to the person that is most likely to respond. Other algorithms apply to this concept because you aren’t so concerned about the why, rather the who, and you usually go with the algorithm that gives you the best lift. I call out logistic regression specifically though because it is also useful for doing conjoint analysis. k-Nearest Neighbors is great for helping you to build out personas for the rest of the team to think about. With it you can find the top five clusters of your customers and the average values for them on any number of socio-economic variables that you have access to.

Of course, there are really cool, sexy, machine learning algorithms out there that you should learn, but you need to start with the basics and really master those.



Benjamin Cogburn

@ONTRAPORT

As ONTRAPORT’s Traffic Manager, Ben Cogburn is ONTRAPORT’s digital advertising guru. In addition to statistics and video games, Ben is an avid geology enthusiast and is known for his robust rock collection. Ben graduated from the University of California, Santa Barbara with a degree in Environmental Studies.

“The most important algorithm to understand for marketing data is the…”

Statistical Confidence algorithm. For some background, on Facebook, we run multiple variations of adsets and ads each with their own variables such as image, headline, body copy, audience, etc. Essentially, we split test almost every variable possible and split them all into their own adsets to ensure fair testing. After having the ads run for a few days to gather a large enough sample size (in order to strengthen our confidence), we plug the numbers into a spreadsheet. This spreadsheet has multiple tabs which then automatically calculates, with a specified level of confidence (we use 85%), which of the ads will perform better in the long run with the data provided. We can then turn off or pause any adsets that are underperforming so we are not wasting any money or we can even increase the amount of money in the adsets performing well to get more leads at a better price.

You can find these calculators easily enough online but many only allow for two datasets to be calculated at one time. We expanded on it and tweaked it to what will work best for our situation. In doing so, we are able to compare the Impressions, Impressions-to-conversions, click through rate and other data against other adsets. So understanding the algorithm and knowing how it compares the data is essential to our marketing data analysis.



Gopi Suvanam

Gopi Suvanam is an entrepreneur with a demonstrated history of working in the financial services industry. He is skilled in Machine learning, Analytics, Financial Institutions, Capital Markets, ALM and Derivatives. A strong professional with a Master of Business Administration (MBA) from IIM Ahmedabad and a Bachelor in Computer Science from IIT-Madras, Gopi runs a data analytics and AI solutions firm focused on financial services sales, G-Square Solutions.

“There are two types of algorithms used in marketing data analytics…”
Supervised and unsupervised. Supervised models are typically used for predicting customer behavior, and unsupervised models are used for grouping customers together to form homogeneous clusters. Naive Bayes, Logistic regression, linear regression and decision trees are the algorithms used for supervised models and k-means clustering is the most popular algorithm in the unsupervised category. Each of these algorithms have different uses.

Some sample use cases for the algorithms are:

  • Naive Bayes is used for classifying customers into different groupsfor predicting behavior, like which product a customer is likely to buy.
  • Logistic regression is also used for classifying customers, buttypically only into two groups. For example, customers with high likelihood of attrition and customers with low likelihood of attrition. Logistic regression also gives a score between 0 and 1 based on likelihood.
  • Linear regression is used for predicting numeric values, for example,the potential quantity a customer will buy.
  • Decision trees are also classification algorithms like NaiveBayes or Logistic regression, but can model even more complicated
  • K-means clustering is used to group customers behaving similarly inhomogenous groups so that marketing activities and products can be designed around these groups.


Gene Caballero

@YourGreenPal

Gene Caballero is the co-founder of GreenPal, which has been described as Uber for lawn care.

“One KPI marketing data analysts should understand is…”

LTV or lifetime value of a customer. Basically, this is the estimated value of what you will receive from a customer over the lifetime of the relationship with your company.

Knowing this number will allow you to make better operational decisions, like marketing campaigns, so one can ascertain how the long-term health of the business looks.

Other than customer acquisition costs, LTV is the second most important KPI.



Monica Georgieff

@KanbanizeInc

Monica Georgieff is the Head of Marketing at Kanbanize. She grew up in Canada, has a BA in literature and works (and lives!) where creativity and software intersect. She is an evangelist of Lean digital marketing.

“In order to understand the numbers behind their marketing strategies, marketing managers must…”

Apply algorithms to the way they predict the outcome of their own process within the team. Historical data about the internal marketing workflow, such as throughput, cycle time of tasks, and, equally importantly, efficiency is crucial to the continuous improvement of any marketing operation. A marketing analyst can benefit from tracking the flow metrics of their own team and then use them to generate algorithmic predictions about how fast they will go through a similar campaign or cluster of assignments. Monte Carlo simulations, for example, could work well to achieve this. Marketers should not underestimate the importance of the metrics they can collect from their own process and use in tried and tested algorithms in order to predict their results in the future.



Dan Roberge

@MaintenanceCare

Dan Roberge is the President of Maintenance Care, a free computerized maintenance management system designed specifically to increase the efficiency of word orders, preventive maintenance, and asset management.

“Marketing analysts should be aware of Google’s RankBrain algorithm…”

Which involves machine learning and artificial intelligence when determining rankings in search results. RankBrain is now thought to be the highest ranking factor for Google and its main purpose is to provide the most relevant search results for people using the popular search engine. Every time someone enters a search query, this algorithm converts massive amounts of written language into mathematical entities, making it possible for your computer to understand. The future of marketing and the future of search will rely heavily on AI and machine learning.



Max Page

Max Page is the Founder of CouponHippo.

“Marketing data analysts should understand…”

How any search engine algorithm works that brings you traffic.

Whether you are optimizing for Google Search, Amazon Internal Search or even your own website Internal Search, you should understand how your desired result can come up higher. Figuring out how and why you get users to find and click on your product or page in any search results set is key to influencing those results. You may have to build more links to a page or re-write the product description. The key is to read case studies on how other people have done this and then to experiment by changing copy and external levers till you get it right.



Paul Burke

@Renthoop

Paul Burke is the CEO and Founder of RentHoop, a mobile app that connects renters looking for roommates – often seen as the ‘Tinder for Finding Roommates’ to millennials. He has immense knowledge of the rental market and the effects of rising rent prices in big cities like Seattle, SF, NYC. He’s heard hundreds of stories about roommates – good and bad – and can provide context and best practices for navigating those relationships.

“In terms of the algorithms most important for marketing data analysts to understand…”

I’ll say there’s only two that matter: Google and Facebook.

Every marketer needs to understand the factors that increase your company’s visibility through those two platforms.



Kean Graham

@monetizemore

Kean Graham is the CEO of MonetizeMore a leading ad tech firm that is a Google Certified Partner.

“The most important overall algorithm marketing data analysts must setup is…”

Daily tracking of their company’s sales funnel. The algorithms must automatically pull data from their Analytics, sign-up form software, CRM and referral technology into a consolidated platform. Zapier tends to have many integrations that can enable these algorithms to pull the data. This will enable the marketing and sales teams to gauge their performance on a daily basis.



Khalid Saleh

@khalidh

Khalid Saleh is the co-founder and CEO of conversion optimization company Invesp, a leading provider of conversion rate and landing page optimization solutions.

“There are several algorithms that a marketing analyst needs to understand and implement to derive meaningful and actionable insights from data…”

At the outset, a marketing analyst should have a solid foundation of statistical terminology and its implementation. Concepts such as confidence level, standard deviation, probabilities will impact the day to day operation of many marketers. One common challenge I see is the lack of real understanding of what these concepts mean and, instead, marketers rely on repeating common definitions for some of these terms that do not relate into the real application of these statistical concepts.

In addition, a marketing analyst should have a good understanding of two more broad concepts and their algorithms. These are: 1. Data mining and 2. Predictive analytics.

Data mining algorithms help marketers dig deeper into data looking for user patterns. Perhaps the most common algorithms we use in data mining are C4.5 (taking an input of classified data and taking new input and predicting how it will be classified) and EM algorithm (clustering data to look for similarities).

Predictive analytics uses existing data and how users responded in the past to marketing messages to predict future behavior. It basically builds on the insights from data modeling to use them in predicting future actions. There are many algorithms used in predictive analytics but I believe that a marketing analyst should understand the general concepts as opposed to specifics of how these algorithms are implemented.



Kyle Hearnshaw

@conversion_com

Kyle Hearnshaw is the Head of Conversion Strategy at Conversion.com, the UK’s largest conversion optimization agency.

“While the importance of an algorithm would depend on your industry and available data…”

My general recommendation would be to pay close attention to collaborative filtering algorithms. When companies like Netflix are ready to offer $1m to anyone who can re-build it for them, collaborative filtering is no joke. Essentially, collaborative filtering algorithms are what power recommendation engines of the most tech savvy companies across the globe. Amazon, another giant, is said to generate up to 30% of additional revenue from recommendations alone. It will be a key opportunity for marketers who strive to deliver bottom-line impact.

2017 is also the right time to get started…

Personalization is on the minds of every marketer as the next big thing. People think of personalization as being explicitly based on the user’s own choices, “If the user has previously purchased shoes – let’s show them shoes on the homepage when they return.” However, arguably more effective personalization comes from looking at the behavior of other users that are similar to your target user – enter collaborative filtering.

We can use information from a larger pool of data to infer things about our users based on what other similar users have done or liked. This allows us personalize the experiences even for users about whom we know very little – just based on some very early actions they take and using our algorithm to predict things about that user. Returning back to the shoe example, we might infer that, from the types of shoe that this user is browsing on our site, and our existing knowledge of users that browsed a similar selection, that this user is likely to be a very price-conscious user. We could then serve them a discount offer when they reach the key decision point in their journey. Measure the impact, learn from visitors’ behavior, update our knowledge, iterate and repeat.​



Donna Duncan

@beseenontop

Donna Duncan is the SEO / Content Marketing Consultant for B-SeenOnTop LLC.

“The most important algorithm marketers need to understand is…”

The one that Google uses to decide who ranks where in search results. Google’s ranking algorithm is called Hummingbird, a name that means precise and fast and represents the standard of perfection that Google wants marketers to help them achieve – the delivery of sought-after, relevant, fresh, unique, specific, engaging, and valuable information that quickly helps people do what they need to do.

The inner workings of Google’s Hummingbird ranking algorithm are top secret, but there are best practices readily available from search engine optimization (SEO) specialists on the Internet that can help you construct a marketing plan that better ensures your chances of earning your way to the top.



Brandon Seymour 

@Beymour

Brandon Seymour is the founder of Beymour Consulting, a Florida-based digital marketing agency, specializing in search, branding, and reputation management.

“Back in 2011, Google’s Matt Cutts said…”

Don’t chase after Google’s algorithm, chase after your best interpretation of what users want, because that’s what Google’s chasing after. Now that machine learning has become the driving force behind Google’s search algorithm, marketing analysts will need to double down on user intent. As search engines become more sophisticated, it will become increasingly difficult to fake your way to the top of search results. In some ways, Google’s machine learning still has miles to go, but if you want to stay ahead of the algorithm, you’ll need to focus less on what Google wants, and more on what the user wants, since these will eventually be one and the same.



Justin Yek

@justinyek

Justin is a partner at Altitude Labs, a digital transformation consultancy. He builds technology that converts retail data into sales with predictive marketing and personalization.

“The most important algorithms for marketing data analysts to understand are ones that tie their teams’ actions to more sales…”

In today’s context, these actions are ones that predict customer behavior and personalize consumer experiences to maximize expected returns to the business. This comes in the form of having meaningful conversations with their customers by sending the right content, at the right time, consistently across devices and channels.

Marketing algorithms fall within these categories:

Predictive churn rate: To identify active, at-risk and lost customers which helps customize marketing. Models: Logistic Regression, Bayesian Inference and Pareto/NBD model, and inputs include frequency, recency and time between purchases. Recent techniques include Q-learning.

Predictive customer lifetime value: Gives an estimation on how much you can expect to earn from a customer over his lifetime. This is useful in knowing how much you should spend on acquisition. Models: Gamma-Gamma models and hidden Markov chains models. Inputs include predictive churn rate and the average amount of purchases.

Customer personas: Group customers with similar buying patterns using the previous purchased products, so you can target customers with particular offer. Models: Non-supervised learning algorithms such as k-means.

Replenishment: Identify the right time that a customer will need to reorder a product again. Models: Time Series analysis, Monte Carlo Markov chains and probabilistic models.

Recommendations: Suggest products that customer are most likely to buy, based on his purchase history or on the product he is currently viewing. Models: Collaborative filtering and content-based recommendations. Recent techniques add a layer of reinforcement learning.

Share of wallet: Percentage of a customer’s expenses for a product that goes to your store. Models: Quantile nearest neighbor and quantile regression are used.

Affinity analysis: Used to identify groups of products that are bought together. Models: A Priori algorithm.



Arsalan Jabbar

@Gaditek

Arsalan Jabbar is working as Data Scientist at Gaditek, A Digital Agency with Global Alliances & Partnerships. He has hands-on experience in NGO, Tech & SaaS industries.

“The most important algorithms for marketing data analysts are…”

1) Regression

  • Linear (SLR, MLR & GLM)
  • OLS (Ordinary least squares)
  • Logistic
  • Multivariate

Regression algorithms can be used to analyze marketing campaigns – deducing effect of independent variables over dependent ones and to gauge other cause-and-effect relationships.

2) Bayesian 

  • Naive
  • Gaussian
  • Multinomial naive bayes

Bayesian algorithms provide data insights in decision making, campaigns, pricing decisions and new product development.

3) Time series

Assists in forecasting and decision making.

4) kNN (Nearest neighbor)

Used for predictions based on historic data.

5) Principle component analysis

An efficient algorithm to be used for market research – allowing researchers to deduce observations into highly correlated or non-correlated components to focus on.

6) Clustering

Another efficient algorithm for market research – allowing researchers to identify market segments or groups with similar interests or characteristics, hence, providing target market insights.

7) Decision tree 

A graphical model that enables a researcher to deduce possible outcomes of a decision and thus identifying the threshold of risk involved.



Flynn Zaiger

@FlynnZaiger

Flynn Zaiger is the Founder and CEO of Online Optimism, a digital marketing agency located in New Orleans that’s the youngest company on the city’s Best Places to Work list the past two years.

“One of the most important algorithms for marketing data analysts to understand is…”

Information Fuzzy Networks. Yes, it has a slightly silly name. But it’s crucial as a data analysis to understand it and its uses. Info Fuzzy Networks provides for helpful construction of decision trees. Rather than actual tree, they’re constructed directed graphs, which can be utilized in a manner similar, but different than trees, due to the vertices being connected by edges. These nodes make all the difference.



Linda Allen

@DigitizeFZC

Linda Allen is the Vice President of Customer Experience for Digitize.

“It’s crucial for marketing data analysts to understand…”

Eigenvalues. Google uses it for page rankings, Facebook for the news feed, Google+ and Facebook friend suggestions, and LinkedIn for job suggestions and contacts, Netflix and Hulu for movies, YouTube for videos: each of these has a different goal; however, the eigenvalue-math is the same.

What algorithms impact your marketing efforts most, and how do you leverage your knowledge to make it work for your company?  

Already a member?
Download the Report

The Forrester New Wave™: AI-Fueled Speech Analytics Solutions, Q2 2018

GET IT NOW