AI Data Analytics: AI Needs Context. It Doesn’t Need Another Dashboard.

What happens when the primary interface to your data stops being a dashboard?

For a very long time, one of the ultimate goals of nonprofit data work has been the dashboard. We bring data together from different systems, clean it, reconcile it, agree on definitions, and eventually turn it into reports and visualizations that help people understand what is happening.

I’ve spent enough of my career in nonprofit technology to remember when reaching that point felt revolutionary. In many organizations, it still is. When information has been scattered across a CRM, email platform, website, advocacy system, event tools, spreadsheets, and whatever mysterious database only one person in IT understands, seeing the organization clearly in one place is a genuine accomplishment.

Dashboards aren’t going away. Trends are easier to spot in a chart, executives need consistent measures to monitor, and teams benefit from looking at the same numbers. But lately I’ve been wondering whether we’ve spent so much time thinking of the dashboard as the destination that we’ve missed that it may actually be an interface. And that interface is starting to change.

For us, this isn’t theoretical or something sitting on a future roadmap. Clients use Claude with their data in Unite Analytics today, and other Unite clients can access this capability now. Claude is one of the tools we use, but it doesn’t have to be Claude. The same approach supports tools like ChatGPT or Microsoft Copilot, giving organizations flexibility in how they interact with their data.

Snowflake, which sits underneath Unite, gives us the ability to support that interaction through technologies such as Model Context Protocol, or MCP. Rather than building another predefined view of the data, we can give an AI tool a governed way to interact with the data foundation itself. That has me thinking differently about what the next generation of analytics might look like.

What Happens When We Can Just Ask?

Imagine that I want to understand what’s happening with sustainer retention. Today, I might open a dashboard, find the retention section, adjust the date range, filter for sustainers, compare this year with last year, and eventually notice that retention is down.

That’s useful, but it’s almost certainly not enough to make a decision. I still want to know why. Is the decline concentrated among newer sustainers? Did our acquisition mix change? Does the pattern vary by channel or source? Is what we’re seeing unusual, or are comparable organizations experiencing something similar?

I wrote recently about this in From Data to Decisions: We Have the Dashboard. Now What? The point wasn’t that dashboards have failed us. It was that getting an answer should often lead us to a better question, and that good analytics is as much about knowing how to follow an answer as it is about getting one in the first place.

AI changes how quickly we can follow that trail. Instead of navigating from dashboard to report to query to spreadsheet, I can begin by asking why sustainer retention declined. If the answer points toward newer sustainers, I can ask where they came from. If digital acquisition looks different, I can follow that. The analytical process hasn’t really changed; we’re still moving from question to answer to context to a better question. What has changed is the interface between people and data.

A Simpler Interface Demands a Better Foundation

A traditional dashboard requires people to make many decisions in advance: which metrics matter, how to calculate them, which dimensions users can explore, and which comparisons are useful. By the time the dashboard reaches the user, it has already embedded a substantial amount of interpretation.

Conversational analytics removes some of those constraints. I don’t necessarily need to know which report contains the answer or how the underlying data is structured. I can begin with the question I actually have. From the user’s perspective, that feels wonderfully simple. Underneath the surface, however, it places an even greater burden on the data.

If I ask Claude, ChatGPT, Copilot, or another AI assistant why sustainer retention declined, the particular model matters less than whether it has the context required to answer. It has to understand what our organization means by a sustainer and how we calculate retention. It needs to know where the relevant information lives, how those sources relate, and that the same constituent may appear in several systems. If I ask it to compare this year with three years ago, it may also need to understand what was true at the time rather than applying today’s attributes to yesterday’s behavior.

Because Unite is built on Snowflake, MCP can connect an organization’s AI tool of choice to its governed data environment. Organizations don’t have to wait for us to invent a proprietary ROI chatbot to work this way. Increasingly, intelligence can come to the data rather than requiring the data to move into yet another application.

But connectivity and context are different things. Giving an AI a path to the data doesn’t teach it what an active donor means, which records belong to the same person, or how two systems represent the same concept. That’s why the work behind AI Requires a Strong Data Foundation matters. The Common Data Model, identity resolution, historical context, and shared definitions give the information consistent meaning. MCP can help create the connection; the data foundation gives that connection something coherent to understand.

The Dashboard Has Been Doing More Work Than We Realize

Dashboards don’t just make information easier to consume. They also quietly constrain the questions we can ask, and that has value. If a dashboard contains twelve carefully defined metrics, someone has presumably decided what they mean, validated the calculations, and determined the comparisons make sense.

A conversational interface changes that relationship. Suddenly I can compare populations nobody anticipated comparing or explore relationships between variables that have never appeared together on a report. That freedom is exciting, but freedom and understanding aren’t the same thing. Two populations can be technically comparable without the comparison being meaningful. Two fields can have similar names while representing different concepts. Ten years of constituent behavior can be completely accurate without every event being equally relevant to understanding the person today.

When we give AI tools the ability to explore organizational data more freely, those distinctions become more important, not less. The challenge used to be making data accessible enough that people could answer their questions. We may be entering a period in which the harder challenge is making nearly unlimited access trustworthy.

Maybe We’ve Had the Relationship Backward

Historically, we’ve tried to anticipate the questions people will ask and build dashboards around them. We choose the KPIs, create the views, provide the filters, and hope we’ve built something flexible enough to answer most of what users will eventually want to know. When a question falls outside those boundaries, someone creates another report, dashboard, or spreadsheet.

Conversational analytics lets us start somewhere else. We can start with the question, explore the data, follow what we learn, and eventually discover something important enough that we want to monitor it consistently.

At that point, I may very much want a dashboard. In fact, it may become more valuable because now I know why I’m building it. Perhaps the dashboard becomes less about discovering every possible insight and more about institutionalizing the insights that matter. A conversation with the data helps identify a population worth watching, a hypothesis worth testing, or a change in behavior that deserves attention. The dashboard then gives the organization a shared way to see what happens next.

The Most Valuable Part May Be What You Can’t See

We’ve spent a lot of time lately talking about data foundations at ROI Solutions, and I realize that isn’t always the sexiest part of the technology conversation. It’s easier to show Claude answering a complicated question than to show identity resolution. A conversational interface is more immediately impressive than a Common Data Model. But the more powerful the interface, the more valuable that invisible work.

If asking a complicated analytical question becomes as easy as typing a sentence, the differentiator won’t be the text box. Everybody will have a text box. What matters is what sits behind it: whether the system understands what the data means, how the pieces relate, which records belong to the same person, what was true at a particular point in time, and where the available evidence ends.

That’s what makes our clients’ work so useful for us. We aren’t speculating about a future state where nonprofits can talk with their data. Clients are doing it today with Claude and Unite, and the same underlying approach creates opportunities to use ChatGPT, Copilot, and other AI tools. Unite clients can access this capability today.

We have an upcoming post where we’ll dig into some of those real client use cases: what they’re asking, how they’re using the capability, what’s working, and what we’re learning as people move beyond predefined reports and dashboards. The impressive part may be asking a question in plain English and getting an answer in seconds, but I’m increasingly interested in the less visible work that makes the answer useful.

I’m also interested in whether a truly useful analytical assistant should be as good at recognizing what it doesn’t know as it is at producing answers. Knowing where the available evidence ends may become one of the most important characteristics of trustworthy AI.

For now, I keep coming back to how much the interface is changing while the underlying challenge remains familiar. We spent years bringing data together so we could see more of the constituent relationship. We built dashboards to make that information accessible. Now AI gives us a more fluid way to explore it, using tools organizations are already adopting rather than forcing everyone into one prescribed interface.

The dashboard gave us a window into the data. AI increasingly lets us talk with it. But for that conversation to be worth having, the data needs more than a place to live. It needs structure, definitions, history, relationships, and enough context to understand what we’re actually asking.

AI doesn’t need another dashboard. It needs context. Let’s Talk!