Why preserving history is only half the challenge of building a modern nonprofit data strategy and foundation.
Our systems have gotten remarkably good at remembering.
That’s something we’ve spent decades working toward in nonprofit technology. We wanted complete giving histories, campaign responses, event attendance, volunteer activity, membership changes, advocacy actions, preferences, relationships, and all the other interactions that help us understand someone’s connection to an organization. As the technology has evolved, we’ve also realized that much of the relationship doesn’t live in the CRM at all. It’s scattered across email platforms, websites, advocacy systems, event tools, content platforms, membership systems, and many other places.
That realization helped lead us to build Unite Analytics. If we can bring those sources together, resolve identity across them, and represent the information consistently, we can see much more of the constituent relationship than any one system can show us. Unite is designed around governed analytical structures and analysis-ready data rather than simply putting multiple source systems in the same place.
I’ve written before about what happens when more of that history becomes visible. In Maybe We Should Stop Calling It a Donor Journey, I discussed that what fundraising calls acquisition may actually be year ten from the constituent’s perspective. Seeing more of the past can fundamentally change how we understand when the relationship began and what it has meant along the way.
Lately, though, some of our work on the Common Data Model has me thinking about time from another direction. The more complete the history becomes, the more important another question becomes. Does remembering everything that was true about someone necessarily help us understand what is true about them now?
Preserving What Was True Then
Our clients ask us questions about time constantly, even if they don’t describe them that way. They want to compare year over year, month over month, and sometimes day over day. They want to understand how the constituent file changes, how people move between segments, how membership and sustainer populations evolve, and what an audience actually looked like when a particular campaign ran.
Answering those questions correctly requires more than preserving transactions. We also have to preserve the context surrounding them.
Imagine that someone belonged to one segment when they responded to a campaign two years ago and belongs to a completely different segment today. If the current segment simply replaces the old one, the record may accurately represent who that person is now. But when we go back to analyze the campaign, we can end up viewing the past through the lens of the present. We still know the transaction occurred, but we’ve lost some of the context that gave it meaning at the time.
This is one reason we’ve accounted for slowly changing dimensions, or SCDs, within the Common Data Model. The terminology is pure data warehousing, but the value is wonderfully straightforward. When something about a constituent changes, sometimes we need to preserve what it used to be rather than simply replacing it with what is true today.
That can apply to a constituent’s segment, membership level, sustainer status, household, assigned gift officer, engagement classification, or many other attributes. To understand how relationships and populations evolve, we sometimes need to know both what is true now and what was true at a particular point in the past.
This is basic data-modeling work, and I mean “basic” as a compliment. There is enormous value in being able to confidently answer a seemingly simple question like What was true on this date? It’s the kind of capability that makes year-over-year analysis meaningful, not just convenient.
It also shows why we keep investing so much in the Common Data Model within Unite. Bringing disparate sources together is only the beginning. The model creates a consistent analytical structure for constituent, donation, campaign, interaction, engagement, and other data that may originate in systems that represent the same concepts very differently. The more consistently we can represent that information, including its historical context, the more confidently we can understand how things actually changed.
But as we’ve been thinking about how to preserve historical context, I’ve become increasingly interested in almost the opposite problem.
When History Is Accurate but No Longer Current
Suppose someone responded to an emergency appeal eight years ago. That response belongs in their history, and there may be very good reasons to preserve the context surrounding it. But does that gift still tell us that emergency response matters to them today?
Maybe, but we shouldn’t assume so simply because the data is still there. The same question applies to other signals we collect. Someone may have attended an event in 2019, clicked repeatedly on content about a particular issue in 2022, or been identified as a major gift prospect five years ago. Those facts don’t become inaccurate simply because time passes. What changes is how confidently we can use them to understand the person in front of us today.
This is where I started thinking about two clocks. Good data needs one clock that preserves history and another that measures relevance.
The first clock allows us to understand the past without rewriting it every time the present changes. Slowly changing dimensions do exactly that: they preserve enough historical context to know not merely that something happened, but what else was true when it happened.
The second clock asks a different question. It asks how much a historical signal should influence our understanding of the constituent today. That turns out to be much harder, because different information ages differently. A click may lose relevance relatively quickly, while repeated engagement with the same issue over several years may tell us something more durable. An inferred interest may deserve less weight than a preference someone explicitly gave us. A planned giving conversation from five years ago could remain enormously important, while a behavioral signal from the same year may tell us very little about what someone wants today.
No universal expiration date applies to constituent data because people don’t work that way. A signal’s significance depends on what it represents, how it was collected, what has happened since, and what else we now know.
That last part may matter most. Suppose years of behavior led us to infer that someone cared deeply about one part of the mission, but yesterday they explicitly told us that something else matters most to them now. The new information doesn’t make the old history wrong. It changes the context in which we should interpret it.
In other words, preserving historical truth and understanding current relevance are two different jobs. One prevents new information from erasing the past. The other prevents the past from overpowering the present.
People Change, Even When Their Records Don’t
This may sound like a data-modeling problem, but at its heart it’s a very human one.
People change jobs, move, have children, retire, lose loved ones, become more financially secure or less financially secure, and develop new interests. Issues that were deeply personal ten years ago may matter less, while something they barely thought about then may become central to their lives.
Their relationships with our organizations change too. A volunteer becomes a donor. A donor becomes a sustainer. A listener becomes a member. Someone who was deeply engaged may drift away for a while and later return. Even the same behavior can mean something different depending on where someone is in their relationship with us.
Our databases are extraordinarily good at accumulating evidence of who someone has been. The more sources we connect, the more complete that evidence becomes. That is enormously valuable, but completeness introduces its own risk if we begin treating everything we know about someone’s past as equally representative of who they are today.
The fact that something happened is permanent. What it tells us about the person shouldn’t necessarily be.
That distinction matters for human decision-making, but it becomes even more consequential as we ask machines to help us interpret these increasingly complete histories.
A Strong Data Foundation Also Needs a Sense of Time
Karen Taggart recently wrote about why AI Requires a Strong Data Foundation. AI needs consistent structures, shared definitions, identity resolution, and governed data if we expect it to understand what our data actually means.
The Common Data Model is an important part of that foundation. By creating consistent structures across disparate sources and combining them with identity resolution, we can reduce ambiguity that otherwise gets pushed downstream to analysts, reporting tools, and eventually AI.
But I think time belongs in that foundation too.
Imagine that we’ve done everything else right. We’ve resolved the identity and know that the donations, email activity, event attendance, advocacy actions, and other interactions scattered across different systems belong to the same human being. We’ve represented those interactions consistently, preserved important historical context, and created a beautifully structured ten-year history of the relationship.
Then we ask an AI what that person cares about today. Every fact available to the AI could be completely accurate, and it could still reach the wrong conclusion if it doesn’t understand how those facts relate to time. A preference explicitly stated last week probably shouldn’t carry the same weight as an inferred interest from five years ago. A recent pattern of sustained engagement may tell us more about current interest than a burst of activity surrounding one event years earlier.
The problem isn’t inaccurate data. It’s accurate data without enough context to understand its present meaning.
That feels like an important extension of the data-foundation conversation. We need structures that help us understand not only what our data means, but also when it meant it and how its meaning may have changed.
Remember the Past Without Getting Stuck in It
A lot of attention right now is on the exciting end of the data stack: AI, predictive analytics, conversational interfaces, next-best-action recommendations, and increasingly sophisticated ways to activate what we know about constituents. I’m excited about those things too.
But our work with Unite keeps reinforcing how much those capabilities depend on getting some decidedly less glamorous things right underneath them. Can we confidently understand what was true on a particular date? How has a constituent or population changed over time? Can we preserve the context in which something happened, rather than letting every change in the present quietly rewrite our understanding of the past?
Those are some of the problems we’re solving for in the Common Data Model. Slowly changing dimensions may never get the biggest reaction in a product demonstration, but preserving historical context is basic, powerful, high-value functionality. It makes the analytics built on top of that data more trustworthy and gives us a stronger foundation for everything we want to do next.
At the same time, preserving the past is only half the challenge. Once we have that increasingly complete history, we also need to become more thoughtful about how much of it should influence our understanding of the present. That becomes particularly important when we move from understanding a relationship to deciding what to do next. I’ve argued before that Your Next Best Action Might Be No Action at All. Making that kind of judgment depends on understanding what the relationship needs now, not simply accumulating everything we’ve ever known about it.
That’s why I keep coming back to the idea of two clocks. For analytics, we often need a clock that preserves what was true at the time so we can understand how things actually changed. For engagement and decision-making, we need another that helps us understand how recent a signal is, whether its relevance has faded, and whether newer information has changed what that history tells us.
We don’t need our systems to forget. We need them to become better at understanding that remembering something happened and believing it still describes the person are not the same thing.
We’ve spent years teaching our systems to remember more of the constituent relationship. As we continue building the Common Data Model in Unite, preserving that history remains enormously important. But the more complete our memory becomes, the more important it is for our data to develop a sense of time.
Good data needs two clocks: one that preserves history and one that measures relevance.
If we can get both right, we won’t simply have a better record of who our constituents have been. We’ll have a much better chance of understanding who they are now.
Are you ready to explore your own data time machine? Let’s Talk!
Key Takeaways
- Building a modern nonprofit data strategy involves more than preserving historical data; it requires understanding current relevance.
- Unite Analytics helps consolidate data from various sources, providing a clearer view of constituent relationships.
- Slowly changing dimensions are essential for retaining historical context while analyzing current data trends.
- Good data strategies need two clocks: one for preserving history and another for assessing relevance over time.
- Effective decision-making relies on understanding how past behaviors influence current engagement, ensuring that organizations adapt to constituents’ evolving needs.
