Zeta Proudly Announces New Chief Revenue Officers, Jed Hartman and Kelly Kennedy >>

Utilizing Ensemble Models for Marketing Campaigns

An insightful piece on ensemble modeling from Pawan Pathak, Gayan De Silva, David Hanzelka, and Sunpreet Singh Khanuja.

Companies are improving their Machine Learning (ML) processes to increase the performance of their marketing campaigns. They’re doing this because customer behavior data is so sophisticated (i.e., complex) that it’s become difficult for humans to make sense of it on their own. As the relationships between data attributes become increasingly non-linear and complex, we need a more intelligent approach to Machine Learning that can not only highlight these relationships but learn from them to guide business decisioning. An effective method is the simultaneous application of multiple advanced ML models, combined to form an ensemble model. In this blog, you’ll see an example of utilizing an ensemble model for marketing campaigns and learn how to overcome common problems related to ensemble models.

What is ensemble modeling?

Ensemble modeling is a technique that combines multiple ML models to create a final singular prediction. The prediction may be in the form of Classification (Yes/No), Estimation (predicting future demand) or Ranking (force rank an audience based on perform a desired action).

Why use ensemble models for marketing?

Simply put, more is better than one.

Multiple ML models learn different patterns, and combining them into an ensemble model improves the accuracy and stability of the predictions.

Five things to keep in mind when designing an ensemble model…

If you’re ready to transition from legacy model processes to ensemble models, there are five key things you need to keep in mind:

  1. Type of Prediction (Classification vs Estimation vs Ranking)
  2. ML algorithms to be determined based on the type of prediction
  3. Architecture (including existing technology and deployment limitations)
  4. Model transparency (explainable AI)
  5. Performance measurement KPIs

An example—how Zeta implemented ensemble models

During the initial phase of implementation, our primary objective was setting up a semi-automated pipeline (we decided to leave the performance improvements and fine-tuning for later on).

We considered 5 ML algorithms to start:

  • Logistic regression
  • Random forest
  • XGboost
  • Gradient boosting machines (GBM)
  • Neural networks

The simplified architecture for the pipeline is shown below.

The first step in modeling is the creation of the modeling dataset. This process is automated using historical conversion data from previous marketing campaigns to create the ”Dependent” variable. This step also includes appending “Independent” variables to the modeling dataset (to be clear, Zeta owns many data sources containing demographic, affinity, location, behavioral, etc. information for 300M+ permissioned email addresses).

Steps two and three are the most critical in any modeling process, which (at a high-level) involve cleaning the data, transforming and reducing the features to retain only the most relevant (yet non-redundant) independent features for the model.

In the fourth step, the cleaned, transformed, and reduced modeling data is fed to multiple ML algorithms, each producing a model score and model object as an output.

Finally, these model scores are combined using soft voting to obtain the final ensemble model score and object.

Ensemble model scoring

Zeta’s ensemble modeling pipeline was developed in a python environment. To efficiently achieve the scalability required for scoring 300M – 800M email addresses, Zeta uses scalable SQL tools.

Each individual model used to create the ensemble is extracted as a set of rules and constraints in the form of an SQL query. These SQL queries are then executed on our 300M+ universe to create a score from each model.

Finally, the scores from each individual model are combined using soft voting to create the final ensemble model score on the entire universe.

Model validation

The performance of a model is key in determining its business impact on a marketing campaign.  Before rolling out ensemble models, we did extensive comparison with the business-as-usual modeling process. We used the following metrics to evaluate ensemble models versus business-as-usual models:

  • AUC
  • Accuracy
  • Precision
  • Recall
  • F-score
  • Confusion matrix
  • Lift charts

The results?

In general, we see 20% improvement in new customer acquisition cost for our clients using ensemble models. This evaluation was carried out in a designed experiment, ensuring statistical significance and confidence in the achieved lift.

Need more help with ensemble models for marketing—Talk to us!

If you’d like help learning how to use ensemble models to deliver more customer satisfaction, talk to us.

Thank You for Everything

Joshua offers his friends and colleagues at Zeta a heartfelt goodbye.

By Joshua Koran, Head of Innovation Labs, Zeta

After over 2 years at Zeta (and almost 4 years including my time at Sizmek), it comes with bittersweet emotions to announce my departure. I am moving on to continue pursuing my passion advocating for a free and open web that is supported by responsible addressable media.

My Zeta adventure began with the Sizmek acquisition in May 2019. I am extremely proud of what we’ve accomplished over the past few years. In my time here, I’ve contributed to the company’s growth in innovative products, multiple acquisition integrations, becoming a public company, and what I consider to be impactful on a macro level, helping evangelize the best path for the ecosystem to pursue as we proceed toward the future of addressability.

Just a few short months after I joined, one of the first of many surprise announcements shook the web ecosystem that one of the largest browsers would no longer support digital markets’ reliance on the foundational technology standards known as “cookies.” As we all know, this was the beginning of a frenzy that has since set the industry searching for improved solutions to continue to allow small organizations to rely on supply chain partners, which require interoperable data that today is often stored in cookies.

At Zeta, we have been prominent in our positioning to maintain a free and open marketplace and ensure that marketers and publishers continue to have choice over the partners they work with. We have been leaders in educating marketers around the world, co-founded a new trade body, Marketers for an Open Web, and advocated the importance of the responsible use of interoperable data to support digital markets regulators worldwide.

This important work will continue as Zeta advocates to give their people, publishers, and marketers the best opportunities to communicate and continue to navigate the open web.

As the regulator scrutiny of internet gatekeeper conduct continues to increase in the European region, where regulators are further ahead in balancing people’s important privacy rights with the collective social rights of open marketplaces and freedom of expression, I will continue pursuing my passion to advocate for balanced solutions, rather than those that favor only the largest established players.

All in all, I am very grateful for my time at Zeta and working alongside the incredibly talented teams across the organization that are helping marketers and publishers attract, retain and grow relationships with people.

Thank you to everyone for your partnership, your friendship, and for everything we’ve done together.

Joshua Koran

5 Back to School Marketing Ideas for Retailers

Consider these 5 critical back to school marketing ideas as you prepare your 2021 strategy.

Back to school is one of the busiest times of year for parents, students, teachers, and—of course—retailers. In fact, the back-to-school shopping season is one of the busiest times of year for everyone in the retail industry. From book bags and pens, to new shoes and new clothes, it’s a time of year when consumers aren’t shy about spending money. With that in mind, here are five back to school marketing ideas to help retailers maximize their success this season.

Back to school marketing idea #1? — Partner with local businesses

Consumers report feeling compelled and motivated to “shop small” and save local businesses negatively impacted by the pandemic. 

To capitalize on this year’s preference for local shopping, marketers should consider setting aside some back-to-school budget for partnerships with smaller, independent brick-and-mortar merchants. Promote these partnerships using thoughtful email and social campaigns to cultivate awareness and promote key back-to-school products.

Back to school marketing idea #2? — Lean on “hero” products

Bundle popular products from different categories to help back-to-school shoppers meet multiple needs at once. Identify the “hero” products of the season (trendy, quick-selling items) and package them with less popular products to help move stale items off shelves. Use geofencing with social media to increase consumer interest in product bundles at specific brick-and-mortar locations.  

Back to school marketing idea #3? — Promote in-person shopping

Consumers are tired of pointing and clicking to do their shopping. Many are excited to return to stores and the in-person buying experience—they want to become tactile shoppers again.

More than 50% of consumers plan on doing some form of their back to school shopping in person, so don’t disappoint them. Even those looking to still buy online will be increasingly interested in in-person fulfillment like curbside or in-store pickup. 

When it comes to last-mile fulfillment, consumers value immediate, transparent communication above all else. With that in mind, consider making SMS notifications an integral part of the fulfillment experience this year. 

Back to school marketing idea #4? — Use sales early and often

Even with record-breaking household savings, many back-to-school shoppers will remain price conscious. In fact, 2 out of 3 consumers will plan their back-to-school shopping around a major sale

  • 35% of consumers will only be able to spend less than $100 on back-to-school
  • 23% of consumers will only be able to spend between $100 and $500 on back-to-school

Retailers should plan events and promotions centered on price sensitivities to encourage consumers to do as much buying as possible in their stores. Coupons, discounts, and other incentives for students, parents, and teachers are a MUST this year.  Lean into loyalty data to understand which products are most sought after, and implement dedicated social hashtags promoting special offers on those items. 

Back to school marketing idea #5? — Talk to kids, not just parents.

Don’t forget that back-to-school marketing isn’t just about connecting with parents. Students play a pivotal role in the buying process, from influencing their parents’ purchase decisions to spending their own money, students can’t be ignored. Analyze signals (behavioral, content, geographic, etc.) related to Generation Z and Generation Alpha across channels, and use any insights gleaned to develop more effective marketing campaigns that zero-in on clear interests. 

Roughly 13% of this year’s back-to-school buying will be done by the students themselves.  

  • Teens will spend approximately $37 of their own money this year
  • Pre-teens will spend approximately $26 of their own money 

Looking for more back to school marketing insights?

If you’re interested in getting even more granular with your back to school marketing—talk to us. From our Zeta Data Trends to our back to school marketing guide, Zeta has everything retail marketers need to stay ahead of the game.  

Why Using Retina Images for Email Is Crucial for Marketing Success

Increase subscriber engagement and trust by adding retina images for high-DPI displays.

If you’re an email marketer, you might be wondering why using retina images for email is so important. Well, the truth is, using retina images for email can help you increase subscriber engagement.

Moreover, it can help you improve the authority of your brand—nothing legitimizes an email like crisp, clear images.

In this blog, we’ll explain the need for retina images for email marketing success. We’ll also explain how you can use retina images more effectively in your marketing, and share some key considerations before getting started.

The need for retina images

High-DPI displays, often referred to as retina displays, have the power to enhance any email campaign. On device screens, DPI refers to the number of pixels a manufacturer can fit into an inch of the screen. The higher the DPI, the more detailed and clear images and text on that screen appear.

Example of using retina images for email
Non-Retina (left) vs Retina (right)

While subject lines, copy, and cadence all play an important part in creating an effective marketing email, the quality of your images as a marketer cannot be overlooked.

If you want your subscribers to look forward to your emails, you need visuals that people admire. After all, when you open an email, your eyes always jump to the visuals before they go back to the copy and call-to-action.

Moreover, if high-DPI displays are not accounted for (in other words, if you’re not using retina images for email), you risk of looking careless to subscribers. Because of the way high-DPI displays work, non-optimized images end up looking blurry and pixelated on retina screens (see the example above).

Using retina images for email

How exactly are images optimized for retina screens?—The answer is quite simple.

Roughly speaking, high-DPI displays have twice as many pixels per inch than their traditional counterparts.

Therefore, for the images to look good on those screens, they need to be twice as large so that when they are scaled down in the email. Doing so ensures there are more pixels for retina screens to display.

For example, if the emails are designed and coded at 600 pixels wide, the images will need to be 1200 pixels wide in order to display properly for retina display. For non-responsive email code, the height and width of the image will remain the same, in order to still restrict the email to stay 600 pixels wide. If the code is responsive, a max-width of 600 would need to be applied to the image styling in order to constrain Outlook.

Example: Original image is 600 x 100 pixels. Retina should be 1200 x 200 pixels. Due to Outlook being very literal with HTML, if the image is coded at 1200 x 200 in the HTML code, it would stretch to be 1200 pixels wide. If it is coded at 600 x 100 pixels, it will display the retina image with double the amount of pixels.

Additional items for consideration

As the dimensions of the image are increased, the file increases in size as well.

That increased file size can cut into subscriber’s data, causing an email to load slowly. So, keep in mind that not every campaign or image within the campaign needs to be retina.

Prioritize what images are going to be retina versus non-retina. Any images that include text or small product/image detail should always be chosen as retina first. Small divider images and spacer images can always be non-retina.

If retina images are needed, there are two techniques to help avoid large files.

  1. Use compressive images—images saved at huge dimensions, but very low quality settings.

To create a compressive image, design the image at something around 4X the intended size and save it at an extremely low quality setting. While the image looks terrible when viewed at full-size, after it is scaled down to its intended size in an email, the artifacts shrink down and go unnoticed. The low quality settings ensure that your file size is kept to a minimum.

2. Use a dedicated program to further compress the images after saving them.

Applications exist for every operating system that allow you to process and compress images after they are created (ImageOptim, JPEGmini, TinyPNG, Compressor.io, etc.).

Need help using retina images for your next email campaign?

If you want to avoid slow load times, decrease file sizes, but still deliver high-quality images through email, talk to the experts at Zeta or learn more about the Zeta Marketing Platform.

Google Presses “Pause” on Canceling Third-Party Cookies

Google delays the deprecation of third-party cookies until late 2023 as they look to support publishers while protecting users’ privacy.

Google is canceling third-party cookies

As reported in the Wall Street Journal, Google will not be canceling third-party cookies any time soon. Originally slated for retirement in January or February of 2022, Google announced they will refrain from canceling third-party cookies until late 2023. In a statement released by the company, Google explained this timeline extension is intended to “allow sufficient time for public discussion on the right solutions, continued engagement with regulators, and for publishers and the advertising industry to migrate their services.”

Publishers, ad-tech companies, and marketers are largely applauding Google’s decision as one that supports the preservation of an open web. Moreover, they agree this new timeline will give them the time required to find suitable alternatives to the third-party cookie.

Google’s new schedule for canceling third-party cookies is as follows:

  • Phase 1: Starting in late 2022, Google will begin intensely testing its planned Privacy Sandbox initiative.
  • Phase 2: In mid-2023, Google will start phasing out support of third-party cookies by shortening their in-browser lifespan from months to weeks to days.
  • Phase 3: By the end of 2023, all support for third-party cookies will cease, effectively canceling third-party cookies in their entirety.

Where Zeta stands on canceling third-party cookies

Nothing in Google’s decision will impact Zeta’s overarching strategy—Zeta’s technology and solutions are built to thrive in a cookie-less future.

Having said that, Zeta supports Google’s decision in that it is in the best interests of consumers and publishers alike. Because cross-publisher identifiers provide marketers with the accurate, scaled, real-time information needed to achieve optimal personalization in advertising, the delay in canceling third-party cookies will give people more time to discover and implement media identifiers (such as Zeta’s proprietary ID) that offer more in the way of transparency, choice, and accountability.

Additionally, this delay will give marketers the time they need to strengthen their first-party data strategies. Marketers can accomplish this by using partners like Zeta to enrich owned data in existing customer databases to obtain more valuable consumer insights that can be activated across channels.

For more information on this issue, please visit the Zeta Resources page, or simply contact us here.

Zeta Announces Pricing of its Initial Public Offering

NEW YORKJune 9, 2021 /PRNewswire/ — Zeta, a cloud-based marketing technology company that empowers enterprises to acquire, grow, and retain customers, announced today the pricing of its initial public offering of 21,500,000 shares of its Class A common stock at a public offering price of $10 per share comprised of 14,773,939  shares of Class A common stock offered by Zeta and 6,726,061 shares of Class A common stock offered by the selling stockholders. The gross proceeds from the offering, before deducting underwriting discounts and commissions and other offering expenses payable by Zeta, are expected to be approximately $215 million. Zeta’s common stock is expected to begin trading on the New York Stock Exchange on June 10, 2021 under the ticker symbol “ZETA.” The offering is expected to close on June 14, 2021, subject to the satisfaction of customary closing conditions. In addition, certain selling stockholders have granted the underwriters a 30-day option to purchase up to an additional 3,225,000 shares of Class A common stock at the initial public offering price, less the underwriting discounts and commissions.

Morgan Stanley, BofA Securities, Credit Suisse and Barclays will act as joint lead book-running managers for the offering. William Blair, Needham & Company, Oppenheimer & Co., Canaccord Genuity and Roth Capital Partners will also be co-managers.

The proposed offering will be made only by means of a prospectus.  When available, copies of the preliminary prospectus relating to the proposed initial public offering may be obtained from Morgan Stanley & Co. LLC, Attention: Prospectus Department, 180 Varick Street, 2nd Floor, New York, NY 10014; BofA Securities, Attention: Prospectus Department, NC1-004-03-43, 200 North College Street, 3rd Floor, Charlotte, NC 28255-0001, by email at dg.prospectus_requests@bofa.com; Credit Suisse Securities (USA) LLC, Attention: Prospectus Department, 6933 Louis Stephens Drive, Morrisville, NC 27560, by telephone at (800) 221-1037 or by email at usa.prospectus@credit-suisse.com; and Barclays Capital Inc., Attention: Broadridge Financial Solutions, 1155 Long Island Avenue, Edgewood, NY 11717, by telephone at (888) 603-5847 or by email at barclaysprospectus@broadridge.com.

This press release shall not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.

About Zeta
Zeta Global Holdings Corp. is a leading data-driven, cloud-based marketing technology company that empowers enterprises to acquire, grow and retain customers for a lower cost than they can achieve without us. The Company’s Zeta Marketing Platform (the “ZMP”) is the largest omnichannel marketing platform with identity data at its core. The ZMP analyzes billions of structured and unstructured data points to predict consumer intent by leveraging sophisticated artificial intelligence to personalize experiences at scale. Founded in 2007 by David A. Steinberg and John Sculley, the Company is headquartered in New York City. For more information, please go to www.zetaglobal.com.

Myths About Addressable Advertising

Addressable advertising and the loss of cross-publisher IDs are hot topics. In this blog, we talk about some of the most common myths…

There is a ton of confusion swirling around the topic of addressable advertising at the moment. Specifically, there’s confusion around what can replace the loss of cross-publisher IDs. Fortunately, it’s easy to make sense of the chaos. In this blog, we’ll explain how each alternative will support marketers’ advertising effectiveness. 

Contextual targeting

Let’s begin with contextual targeting. 

The notion that contextual offerings can replace the loss of IDs for addressability is misguided. Contextual targeting is very useful in improving how marketers match content to consumers. But it offers nothing in the way of measurement. If marketers can’t accurately measure their campaigns then they can’t do their job. 

Contextual targeting only delivers on engagement, making it an incomplete replacement at best.


A second proposal often mentioned is the use of encrypted emails. Many large publishers require people to authenticate or sign-in to their web properties. The email associated with “logging-in” can be used to generate a consistent, cross-site identifier. The problem with the alternative is that it doesn’t scale across the web. 

Sure, some of the larger publishers like the New York Times or the Washington Post will be able to use authentication because they are established brands with sizable existing audiences. 

But what about the small publishers who make up the long tail of the web? Well, they’ll get crushed operating under an authentication model. As the New York Times’ Senior Vice President of Ad Innovation, Allison Murphy, notes:

“While a differentiator and I’m thrilled about it, this isn’t a path available for every publisher, especially not local who don’t have the scale of resources for building from scratch.”  

More importantly, an authentication approach won’t work for most major brands. As an example, consider Nike, one of the most iconic, consumer-facing businesses in the world. Nike won’t require visitors to Nike.com type in their email address. But without that email identifier to provide view-through attribution, Nike can’t measure their campaign results across publishers, even if they use authenticated targeting on the publisher sites themselves. 

Marketer’s only get to see what’s working and what isn’t in a world where cross-publisher IDs exist. In that world, Nike can conduct addressable, authenticated, people-based targeting across publisher websites, then measure campaign performance in an unauthenticated, “logged-out” state. 

In summary, authentication remains a partial solution. It will not help with measurement or optimization without the existence of cross-publisher IDs.  

Aggregate reporting APIs are often mentioned as a replacement for the real-time feedback loops marketers’ programmatic buying platforms use today. Some OS providers suggest providing reporting APIs with a built in 24-48 hour delay. Yet, given marketers will be wasting their spend proportional to any delay in feedback, the longer the time delay the greater the reduction in effectiveness which leads to lower publisher CPMs.


Last but not least, there are those who would like to replace marketer-specific audiences with generic clusters or “cohorts.” 

Cohorts assign a single audience segment to the browser, while today most browsers belong to thousands of marketer-specific audiences. 

Given millions of people will be in any cohort, they do not offer marketers the ability to understand if any of the people in a given cohort who saw advertising ever arrived at their website or store. 

Cohorts also cannot answer reach, frequency, or effectiveness questions. Marketers need those answers to allocate their budgets or change the prices they pay. The “95% claim” often repeated in the press resulted from a cohort experiment where the real-time machine learning of the DSP was allowed to use cookies to determine when to talk to a given browser belonging to a specific cohort in a given context as well as to calculate what price to pay. 

The truth about cookies and cross-pub IDs

Cookies aren’t perfect.

There are discrepancies between the buy side and the sell side related to counting systems that create problems. 

A common cross-publisher ID will eliminate these discrepancies. It will make life better for both buyers, sellers, and consumers. It will do all this by making it easier for auditors to hold bad actors accountable. 

Leading this charge is the Partnership for Responsible Addressable Media. This organization was created by the Association of National Advertisers (which consists of the world’s largest brands such as Procter and Gamble or Disney). 

PRAM and ANA support Addressable Advertising

Those advertisers are concerned about their ability to engage consumers without cross-publisher IDs. Instead of being able to reach consumers through any number potential publishers, they’ll be forced to choose to go through walled gardens. 

To learn more about this issue, please read Zeta’s blog: Explaining the Turbulence Around Addressability.

Why Relying Too Much on Existing Customers Is Dangerous

Marketers love existing customers, but do they love them too much? Here’s why you can’t overlook the acquisition of new blood.

When it comes to growing a business, many marketers are quick to focus on existing customers. If ‘a bird in the hand is worth two in the bush’ as the old expression goes, it should be easier to squeeze additional value from previous purchasers by way of reactivation, category expansion, or an accelerated buy rate. 

However, the appetite for growth that most companies have exceeds what existing customers can satiate. Here are five reasons why relying too much on existing customers is dangerous.

They get lured away by competitors

Brand loyalty isn’t what it used to be. About half of all consumers will switch from one brand to another, if the new brand can offer them better service, better pricing, or a customer experience that better suits their expectations. Unless a brand can create a massive moat for its products and services, it’s bound to lose retention and reactivation opportunities to competitors willing to slash prices or offer a more appealing customer experience. 

They have less to spend

Cart sizes are shrinking and average order values aren’t what they used to be. According to a McKinsey study, up to 40% of consumers decreased spending on discretionary categories in the last 12 months due to financial concerns that won’t likely be relieved for several years as the global economy struggles to recover from the impact of COVID-19. Constrictions on disposable income limit what consumers can do for a brand. It means they’ll be less likely to increase their buy rate, less likely to try new (or more expensive) products, and less likely to reactivate if they’ve lapsed. 

Existing Customers have less to spend

They get tired of being pushed

At the end of the day, consumers aren’t numbers on a balance sheet—they’re people, and people can get tired of being pushed. When customers are constantly peppered with promotions to buy more (or buy new) products and services they’re bound to grow cold at some point, and when they do, they won’t be able to contribute to revenue growth. 

They start spending time elsewhere

Consumers aren’t static and they can’t be expected to stay active on one channel forever. If a brand isn’t set up to reach them no matter where they go (desktop to mobile, social to email, brick-and-mortar to ecommerce, etc.), and do so in real-time, its connection with them—along with the consumers’ loyalty for the brand—will wane. As the connection withers, it will become more and more difficult to entice current customers to spend more or step into new categories.

They’re gone for good

All customers lapse, but not all customers lapse for the same reason. Some lapse because of economic circumstances (e.g., job loss). Some lapse because of personal reasons (e.g. kids are finally potty trained). Some lapse because of a negative experience with a brand, and it’s these customers who are all but impossible to reactivate. About 90% of customers who have a bad experience with a brand will never come back, and if a customer won’t come back, they can’t contribute to revenue growth.  

You can lose Existing Customers to competitors

Want to learn more about growth, existing customers, and acquisition?

Read our Micropaper, Why Does Real Growth Always Come Down to New Customer Acquisition?

In The Know

Get Fresh Marketing Insights Delivered Right To Your Inbox.