The benefits of data analysis and even artificial intelligence are coming to large data-driven digital marketing programs. It’s more straightforward than ever to analyze large data sets to identify patterns in user behavior that you might use to sharpen your segmentation, targeting, and content strategy. Companies like Amazon and IBM have introduced machine learning applications to cloud platforms, and smaller companies like Arjuna now make it possible for even smaller organizations to apply artificial intelligence to fundraising program decisions.
But before you reach for the stars, you should stand on a firm foundation. Our crawl-walk-run approach will help ensure you have your user data in order and that you’re taking full advantage of the most straightforward ways that it can help you improve your program. And then you can move into more ambitious ways of using aggregated user data to drive customized content and fundraising strategy.
1. Learn to Crawl: Collect and Analyze Data to Understand Your Audience
Donors give more money when they feel that their connection to you is not just transactional, but also relational. They want you to recognize them personally, and to be responsive to their interests. And they want to know that their giving directly sustains your mission.
You can convince them of their mission value by continually showing the connection between donor support and the programming it funds. But to demonstrate recognition and responsiveness, you need data: you can’t recognize them if you don’t know them, and you can’t respond to their interests if you don’t know what those interests are. This means that finding and collecting the data you need to recognize and respond to your donors ― and to know how effectively you’re doing it ― is the foundation of an effective personalization program.
Collecting First-Party Data
Privacy considerations are increasingly leading social media companies and advertising intermediaries to limit access to the third-party data (e.g., shared cookies that reveal personal information) that most organizations have long relied on for targeting and personalization. So your organization is probably building or enhancing its compilation of first-party data (data you collect or compile from your supporters yourselves, that you own).
The first-party data available to you comes in two flavors: explicit (things your supporters tell you about themselves, e.g., their email addresses) and implicit (things you infer from their actions, e.g., what areas of your website they visit and how they behave while they’re there). Ideally you’ll use a mix of both to inform your digital strategy.
Start with implicit data, which you can collect on every website visit. Keeping track of what pages a user visits, how long they stay there, and what they interact with will give you insight into what they care about. You may also be able to infer their geographic location and language preferences by watching what site areas they choose.
With explicit data, you can fill in the picture of your users by gathering information directly from them, by asking them questions through forms or lightboxes (modals) from time to time, or offering them questionnaires or surveys on the website or in email.
With a range of implicit and explicit data in hand, you have the potential to make strategic decisions to help you target more effectively. But to have the greatest impact on your program success, you need a way to tie all these data sources together into a single system of information about users.
Centralizing Data in a CDP to Create User Profiles
Explicit and implicit user data forms a library of data that illuminates what users are interacting with and what they’re interested in. You can generate meaningful insights by viewing your various sources of data about a user side by side. But if your organization is more ambitious, and ready to invest in a system to tie them together into a holistic view, you can use a customer data platform (CDP). A CDP ingests all your implicit and explicit data sources, and matches them up into user profiles containing everything you know about each user.
Looking at a user’s aggregate profile in a CDP, you can more easily infer things about them that would be tedious to determine by manually comparing multiple data sources. For instance, a user might have told you in a survey (explicit content) that they like giraffes, and you might see from their web activity (implicit content) that they spend a lot of their website time enjoying content about zebras. From these two data points together, you can hypothesize that they might donate in response to an email about your wildlife conservation activities in East Africa ― where both giraffes and zebras live.
2. Learn to Walk: Use Your Data to Suggest Tactics and Testing
When you look at what you know about each user in aggregate ― whether by using a CDP to create holistic profiles, or by laying out your separate data sources side by side ― you can build some initial user personas for audience segmentation in your fundraising appeals. And as you accumulate more data (by, among other things, sending out appeals and tracking how individuals respond to them), you can further refine your targeting by focusing attention on particularly valuable segments, and developing tests to build actionable knowledge about behavior. (The more data you want to include in your model, the more complex the calculus, and the more benefit you get from investing in a CDP.)
For example, imagine you’ve successfully identified segments of donors and non-donors who are interested in content about East Africa wildlife conservation. Within those segments, you can focus further on targeted content for subsegments, and develop a testing strategy to learn the best way to further program goals like these:
- Informing mid-tier and major gift officers with additional insights about their prospects interests and actions
- Upgrading the most generous low-tier donors (e.g., those who have given $500-999) into the middle tier
- Changing the content mix to better retain mid-tier ($1000+) donors
- Converting those small donors who give frequently into recurring donors
- Converting highly engaged non-donors into donors
- Increasing the average size of recurring gifts
Let’s take the last two goals as examples, and imagine how you might draw on the data in your CDP to help establish a testing strategy:
Converting highly engaged non-donors into donors: Suppose you identify a group of supporters who haven’t given money, but who open most of your biweekly newsletter emails, click through to your website frequently, and spend at least two minutes (an eternity, in Internet time) reading updates about your giraffe conservation program. You might test serving these supporters a modal or interstitial page each time they land on your site, asking directly for a $10 gift to go directly to giraffe conservation.
Increasing the average size of recurring gifts: Imagine you have a group of recurring donors who give $10, and you’d like to try to upgrade them to $25. You can view your $25 recurring donors in your CDP and try to identify patterns in their behaviors and interests that you can take advantage of in messaging to your $10 recurring donors. For example, if your $25 recurring donors tend to click through more frequently than average to your pages featuring employee stories, that might mean that the employee story content has significant stewardship value. Perhaps you should test a short email series to your $10 recurring donors that specifically highlights the valuable work your employees do, with some personal stories, and then begin greeting them with a modal or interstitial on your site inviting them to upgrade their gift to support the important mission work your employees do,
3. Learn to Run: Personalized Messaging Based on Data Analysis and Machine Learning
Once you’ve mastered the basics, there are multiple ways in which you can apply analysis to large sets of user data ― including via artificial intelligence ― to add even more sophisticated logic to your program strategy.
Developing user insights from a user’s early interactions: Even before you know who a user is ― before they’ve identified themselves by providing an email address ― you can identify behavior patterns among similar users via your CDP, and begin to target content based on the rules they imply. For instance, if people who visit your website from Southern California tend to be interested in content about desert ecosystem preservation, you might lead with desert-related content when an unidentified user who is IP-located to Southern California arrives at your site.
Identifying large cross-system patterns: Data analysis can identify patterns within the totality of your first-party data that might not be obvious to the naked eye. Some of them might not even make intuitive sense: for instance, the data might show that the people most likely to make a supplemental gift outside of your main campaigns are those who are in their first or second year of giving. Why would that be the case? Well, it doesn’t matter why it’s the case! Once you know it is the case, you can make adjustments to your targeting and messaging and generate some incremental revenue.
Cross-testing successful tactics: Suppose you find that a specific messaging tactic improves retention within one segment of donors (e.g., among those who gave last year, but in an amount less than they did in the previous year). Data analysis of information within your CDP may be able to surface other traits that the members of that segment tend to have in common. (For instance, perhaps those people are disproportionately likely to have been on your file more than five years; or to have made at least five gifts for a total amount of $1000 or more; or to have attended at least two Zoom events in the past 18 months. Then, you can test the successful tactic on those members of your broader file who have a combination of those other traits.
Identifying patterns that correlate with desirable or undesirable consequences: Looking at your data in aggregate might yield revelations like these (both fictional, but also plausible):
- People who make 50% fewer site visits within 90 days than they did in the previous 90-day period are twice as likely to unsubscribe (undesirable consequence) in the next month.
- People who have visited the site twice in the past 30 days have a 20% higher average gift amount (desirable consequence) than people who haven’t.
You can use these kinds of insights to target people with supplemental messaging (like email kickers pushing them to make more frequent site visits) that may improve the chances that they take desired actions (give more) or refrain from taking undesired actions (unsubscribe). Note that in a situation like this, it may not even matter in which direction the causality runs: even if someone is inclined to unsubscribe because they have become less interested in your mission, pushing them back to your website might have the effect of renewing their interest.
Using algorithmic learning to improve segmentation and targeting: The kinds of learnings you develop by drawing hypotheses from the patterns in your CDP and then testing them on donor segments are similar to machine learning. You’re identifying the patterns and drawing insights by viewing your donors through the lens of your CDP; you’re testing those insights through programming; and then you’re feeding the results back into the CDP to sharpen future insights. If you want to go further, you can feed a selection of your CDP data into a machine learning algorithm designed to explore a specific question (e.g., “which of our lapsed donors are most likely to reactivate, given a particular type of reactivation series”).
Using artificial intelligence to refine individual donor strategy: Although the broad application of artificial intelligence to digital strategy is still a ways in the future for many organizations, centralizing your implicit and explicit data in a CDP means you’ll be ready to use new services and tools as they become available to sharpen your segmentation and targeting. And even now, some services exist that use artificial intelligence to improve your donors’ experiences at the individual level, leading to more giving.
For instance, Arjuna’s ExactAsk uses your donors’ transaction and interaction history as inputs to an A.I.-driven algorithm that generates customized ask arrays for each donor and each campaign, based on the totality of knowledge you have about that specific donor’s history of behaviors and responses. Arjuna claims that by using artificial intelligence to model each donor’s current level of commitment and giving interest, using ExactAsk consistently will typically lift revenue by 12% to 18% within the first year.
Looking To the Future
The ideal digital content and fundraising program would entail a content experience that was both fully holistic and individualized ― in which everything we served to a particular user, across all channels, took advantage of machine learning to ensure that we were delivering exactly what they wanted to see, right now, under the parameters most likely to drive them to make a gift or take another mission-valuable action.
For all but the very largest organizations, that level of precision is still in the future. But if you deploy the crawl-walk-run strategy outlined here, by the time you get to “run,” your program has taken on an impressive degree of complex personalization, and you’ll be well positioned to adopt new machine learning products and approaches as they come to make economic sense for an organization of your size.