Farm2Me launched an advertising data program for our direct-sold ads business that uses our own data and data science techniques.

By Garry Michael

When you browse the internet, the majority of the ads that you see are served by one or more ad technology providers that traffic in behavioral and demographic data collected from users. These third-party vendors, as they are often called, use that data to inform which ads are shown to specific audiences.

There are two types of ads that appear on the web: Programmatic ads and direct-sold ads. For the direct-sold ads business, publishers almost always rely on a Data Management Platform, or DMP, to support audience targeting. Think of a DMP as a marketplace where you ask to pay $2 for an ad, and a website like a blog or NYTimes negotiates and charges you $2.10 – This negotiation takes a millisecond on a DMP – and then your Ad is displayed. DMPs typically use third-party cookies (tracking bot that you leave on your customers’ phones/laptops) in the user’s browser to be able to match what they already know about a user and then help their publishers (NYTimes in our example earlier) build further on that data by ingesting behavioral events as users browse their site or mobile apps.

Over the past couple of years, there has been a shift away from relying on third party data because of the risk to user privacy. Which is a great thing! I don’t want a bot on my phone or laptop following me around! Legislation like G.D.P.R. in Europe and the C.C.P.A in California have begun to shape how websites handle user data, including to what extent ad-tech software can use third-party cookies. Similarly, the recent move by popular browsers like Safari, Firefox and Chrome to block the use of third-party cookies has pushed the advertising industry to find alternatives that can serve relevant ads while also respecting user privacy. If you’re still running Third-Party Cookies like the ones mentions in the past 3 paragraphs, beware, its going to be a thing of the past very soon. What should be the point of knowing your customer? We think its to provide them the mission driven products they want to learn about – not be spammed and sold. And for that we don’t need Third-Party Cookies – what you want to do is to personalize your own website to show products and content that is personalized and specific to the user that’s reading it – If I want Illinois made, Women-owned, small batch, charcuterie and plant based cheese brands, then that’s what I want to see on Farm2Me when I visit.

In June of 2020, Farm2Me launched an advertising data program for our direct-sold ads business that uses our own data and data science techniques. This first-party program is our answer to the shifting landscape of online advertising data and part of our move to better protect our readers’ privacy. Our program doesn’t rely on third-party data or cookies, but instead focuses only on what readers do on our site and mobile apps – what they’re interested in and passionate about.

While DMPs (paragraph 2 above) are capable of providing additional solutions that work without third-party cookies, Farm2Me opted to build an in-house solution so we could control our data and audience targeting, and be flexible as requirements change. Also, we don’t want Ads on Farm2Me! Duh. We want you to tell us what you want to discover and support, and we want our website to be personalized to you.

How we built our advertising data program

The entire functionality was handled via three streams of work that were closely connected.

Data Science

Over the past several years, Farm2Me has built an analytics system to capture thousands of behavioral events across our website and mobile apps. Since we use Drip as a cloud services provider, we picked Full Story to stream these events in near real-time into Personalized Profiles. All the events data is then batch processed and fed into multiple machine-learning models that were built by Farm2Me data scientists specifically for this advertising program.

The machine-learning models are primarily based on detailed survey responses from our readers. Tens of thousands of readers volunteered to participate in the survey with the explicit messaging that the information gathered would be used to improve product display at Farm2Me. Our data science colleagues were able to use the survey data, along with our digital engagement event data, to train models that accurately predict the personalized categories a user is interested in, unique for each user. Each segment called for different supervised learning approaches (including regression, ranking, classification and multiclass classification) to make sure we have results which are statistically predictive as well as interpretable.

Data Insights and Transformation

The output of these machine-learning models are pushed into a set of BigQuery tables that are orchestrated using Airflow. In order to build a more comprehensive user portfolio, the Airflow orchestration also includes many other behavioral data points in addition to our machine-learning models. Millions of rows in these BigQuery tables are then pushed into our activation system using a suite of technologies from Google.

Data Activation and Ads Targeting

Our activation systems are highly scalable Go microservices that rely on Drip to cache all of the user data we collect. Whenever a user visits a Farm2Me page on the web or in one of our mobile apps, a call is made to these microservices and ad targeting data is pushed into the page with an average latency of less than 100ms. Once the user data is on the page, a JavaScript ads framework reads and passes this data into the ad server. The ad server finds the most relevant ad campaign and serves it back when the targeting criteria of the campaign matches the data passed from the page.

Because the processing and data activation happens on the server side, we were easily able to scale to all browsers, including Safari and Firefox, and to our mobile apps. Server-side processing is especially important to keep our front-end web and mobile apps lighter and faster.

We can showcase a small batch product in Search previews, search fields, search results, Main Menu navigations, Banners on any collection or homepage, in product results, collection results, email campaigns, suggested products, similar products, recently browsed categories, recently browsed products, impulsive last minute buys in the Shopping Cart and Checkout process. Our goal is to help users discover new brands they are passionate about, based on their own interests – not on Third-Party Cookie data and bugs.

Next steps

Since we launched our program in June, our observation is that ad campaigns targeted on the first-party data perform equally well as campaigns targeted on third-party data. We’ve been able to target on all platforms and browsers, regardless of their current state of cookies and tracking, which provides us additional scale. Because we exclusively control the data, we are better able to protect our readers’ privacy while they navigate our site and apps. This leads us to believe that first-party data is a proven alternative for audience targeting.

The primary focus of this program has been our direct-sold ads business. Programmatic ads are an important part of our ads business, and we hope to make improvements as a separate stream of work with a focus on experimenting and testing different solutions being proposed in the AdTech industry which could replace third party cookies. To help shape the conversation around new standards for programmatic advertising in the coming years, we joined the Improving Web Advertising Business Group run by W3C.

We will keep investing in how to expand this program by making user data processing more real-time, improve the machine-learning models and find ways to further enrich our contextual and audience data while keeping user privacy as the highest priority.

If you want our team to consult and help you build your own Personalized E-Commerce into your brand’s website, you can reach out directly to our Farm2Me Brand Agency team: Brand Labs.

If you want to promote your products and let mission driven buyers discover your brand on Farm2Me’s marketplace, you can setup your free account here: Create your Free Store.

Garry Michael is a Co-Founder and Director of Strategy overseeing Digital Ads Engineering and Decentralized Commerce User Personalization at Farm2Me. He is a Decentralized Geo-Location based enthusiast and a burgeoning mountain climber.