How a University Fundraising Team Used HubSpot’s Data Agent to Score 100,000+ Donors
HubSpot has been adding AI features steadily over the past few years, and most of them follow a familiar pattern: they help with content generation, email drafting, or summarizing records. The data agent is different. It is not a writing tool. It is a data processing tool, and what it does is meaningfully different from anything else in the HubSpot ecosystem.
If you have heard the term and are not sure what it actually means in practice, this is a plain-language explanation of what the data agent is, how it works, and what kinds of problems it is well-suited to solve.
Where the Data Agent Lives
The data agent is part of HubSpot’s Data Hub, which is a newer set of capabilities focused on data processing and visualization. Data Hub is where you go when the standard CRM objects and properties are not enough to do what you need — when you need to aggregate information from across the system, transform it, or visualize it in ways that the core CRM does not support natively.
Within Data Hub, the data agent is specifically the feature that allows you to bring AI into that data processing layer. It is not a chatbot you talk to. It is a tool you configure, by writing a prompt, to process contact data and produce an output.
What the Data Agent Actually Does
The core function of the data agent is this: it takes information about a contact, including information from related records, feeds that information into a prompt, runs it through a large language model, and writes the output back to a field in HubSpot.
Each of those steps is worth unpacking.
The information it uses is not limited to what is on the contact record itself. This is one of the most important things to understand about the data agent. Using Data Studio, you can create an aggregated view that pulls together contact details alongside data from associated records — giving history, event attendance, family relationships, communication logs, program participation. That full picture gets fed into the prompt, not just the surface-level contact data.
The prompt is a set of instructions you write. It tells the data agent what to look at, how to evaluate it, and what to output. A well-written prompt reflects your organization’s actual criteria for whatever decision you are trying to support. For donor scoring, that means the prompt encodes how your best development officers think about high-potential donors — what signals matter, how to weight them, what disqualifies a contact from a high priority designation.
The output gets written back to HubSpot as a property on the contact record. That means the result of the AI processing is not locked in a separate system or a spreadsheet. It lives in HubSpot, where it can be used to filter lists, trigger workflows, build reports, and inform outreach decisions. How to build a donor scoring system inside HubSpot goes into detail on how that output gets structured for a donor prioritization use case.
The Smart Column
The output mechanism in the data agent is called a smart column. The easiest way to think about it is as a column in a spreadsheet that is powered by AI.
In a standard spreadsheet, a column holds a value you enter or calculate from other cells. A smart column holds a value that the data agent produces based on the full context of that contact’s data and the instructions in your prompt. You define what you want the column to contain — a score, a summary, a classification, a recommended action — and the data agent populates it for every contact in the view.
For a fundraising team, the most useful version of this is a simple classification: high, medium, or low. That output is easy to read, easy to filter on, and easy to act on. A fundraiser does not need to interpret a complex score or read a paragraph of AI-generated analysis. They need to know who to call today. High means call. Medium means cultivate. Low means hold or deprioritize.
The simplicity of the output is intentional. The data processing behind it is complex. The result the fundraiser sees should not be. Why your fundraising team is spending more time on admin than donor relationships explains why that simplicity matters operationally.
What Makes It Different From Other HubSpot AI Features
Most of HubSpot’s AI features are generative. They produce content: email copy, meeting summaries, suggested responses. They are useful, but they are not doing anything with your CRM data in a structured way.
The data agent is analytical. It is reading your data, applying logic to it, and producing a structured output that lives in your CRM and affects how you work. That is a fundamentally different kind of AI capability.
It is also configurable in a way that most HubSpot AI features are not. The prompt you write determines everything about what the data agent does. Two organizations using the data agent for donor scoring might have completely different prompts, because their criteria for what constitutes a high-priority donor are different. The tool does not come with a fixed definition of donor quality. You supply that definition, and the tool applies it at scale.
This is what makes the data agent particularly well-suited to use cases where expert judgment needs to be systematized. What your senior fundraisers know that your CRM doesn’t yet covers the challenge of capturing that judgment in a form the system can use.
What the Data Agent Is Not
It is worth being clear about what the data agent does not do, because there are reasonable assumptions that turn out to be wrong.
It is not a real-time tool. The data agent processes contacts in batches, not continuously. The frequency with which scores get updated depends on how the system is configured. For most fundraising use cases, running the scoring on a weekly or monthly basis is sufficient, but it is not a live feed that updates the moment a donor makes a gift.
It is not a replacement for data quality work. The data agent synthesizes what is in HubSpot. If giving history is incomplete, or event attendance is not being tracked, or related records are not properly associated, the output will reflect those gaps. Is your HubSpot data clean enough to use AI on? covers what to assess before building.
It is not a black box. The prompt is visible and editable. If the scores the system produces do not match what your team expects, you can look at the prompt, identify where the logic is off, and adjust it. That transparency is one of the things that makes the data agent practical for organizations that need to trust the output.
Where It Fits in a Larger System
The data agent is most useful as one component of a broader donor prioritization setup, not as a standalone tool. It produces the score. The rest of the system — the list segments, the workflows, the reporting views, the outreach assignments — determines what happens with that score.
When FSM built a donor scoring system for a university fundraising team, the data agent was the engine that produced the high, medium, low classification for each of the 100,000-plus contacts in the database. But the value of that classification depended on the list views that used it, the workflows that acted on it, and the reporting that gave management visibility into how the pipeline was distributed. How a university fundraising team got a unified donor view in HubSpot covers how those pieces fit together in practice.
A recent episode of AI in Action walks through this implementation in detail, including the specific data points that went into the scoring prompt and what the team was able to do with the output.
Listen to the full podcast episode here.