Why the AI Conversation Keeps Starting in Finance

The question usually arrives sounding like an AI question. Can we use it for forecasting, for the board packet, for answering the same staff questions every week? Then the conversation turns, and within a few minutes it has become a question about the numbers. Which system has the right one. Why two reports disagree. Who the person is that knows how the month actually closes.

That turn is healthy. It means the organization has found where the real work sits.

2N2 was founded in financial operations. As our clients grew, the work grew with them: into data, then automation, and AI. What stayed constant through every one of those shifts is the layer underneath. Each new capability is built on the same data, and the quality of that foundation decides how far the next step can go.

Is our data organized enough for AI to be useful?

Most leaders ask this expecting a yes or a no. The more useful answer is that some of it is ready now, and it is usually more than people expect.

The sector is in the middle of this work. In the 2026 State of Nonprofit AI survey from NTEN and The Bridgespan Group, about 24% of responding executives said data organized enough to support AI was in place now, and about 47% said it was in development.

For a first use, AI does not need every system cleaned. It needs a few sources the team already trusts and a clear definition for each number that matters. The practical question is narrower than "is our data ready." It is which data this one decision depends on, and whether that slice is ready. A decision about monthly cash position touches a handful of accounts and one close process. That is a scope a team can get right in weeks.

A model answering from data nobody has reconciled will still answer, quickly and confidently. The foundation is what makes the answer worth acting on.

Our systems don't talk to each other. How do we get one view of the numbers?

Every system in a growing organization was added to solve a real problem at the time. The accounting platform, the CRM, the membership or program system, the payroll tool, the spreadsheet that fills the gap between two of them. Each was a good decision. Together they form a map that nobody drew on purpose.

Replacing all of it is rarely the answer, and it is almost never the first answer. The path that holds up is connecting what you already run so the numbers flow into one shared model on a schedule, without someone exporting a file at the end of the month. That shared model is the part the organization owns outright, which keeps the underlying vendors swappable and the logic in your hands.

Some organizations stop at one connected view for one function, and that is a complete outcome. Others build toward one source of truth across functions when they are ready for that level of integration. The sequence matters more than the ambition: connect the systems a named decision depends on, prove it, then extend.

Why do different departments report different numbers for the same thing?

Usually because each team is right. "Revenue," "active member," "client," and "closed month" each get counted a little differently depending on which system you ask and what that team needed the number for. Everyone has quietly adapted, and the totals never quite tie.

This is where finance tends to lead, and for good reason. Finance already reconciles. It already works to a calendar, already has definitions that auditors and boards have tested, and already knows which figure wins when two disagree. That discipline is exactly what every later capability needs.

The fix is small and specific: agree on one definition for each key number, write down which system is authoritative for it, and build that definition into the data once. After that, meetings spend their time on the decision in front of them rather than on reconciling the figures.

Signals the foundation is the next step

• The same question gets a different answer depending on who runs it. That is a definition problem, and no tool will settle it for you.
• A critical number depends on one person's monthly export. The process works, and it is also a single point that can quietly stop.
• Board materials take days to assemble. Most of that time is usually spent reconciling sources, which a connected model does once.
• A pilot produced answers nobody felt safe forwarding. The model worked. The data under it had not been agreed on.
• Nobody can say which system wins when two disagree. Writing that rule down is often the highest-value hour of the whole effort.

Start with three numbers

Pick the three numbers that appear in every board or leadership packet. For each one, ask the people who produce it for their definition and their source. Where the answers differ, that is the first piece of foundation work, and it costs an hour of the right people's time.

That list also tends to show where AI earns its place first. The decisions resting on numbers everyone already agrees on are the ones ready for automation and AI now. The rest have a clear, sized step in front of them.

We work inside that foundation with our clients over time, because the second thing worth building is usually something the first one made visible. If there is a number your team keeps reconciling, that is a good place to start the conversation, at 2n2solutions.com.

Sources

1. NTEN and The Bridgespan Group, State of Nonprofit AI Adoption and Governance, published September 10, 2026. nten.org. 917 nonprofit staff and executives responded in summer 2026. Executive responses n=404. On "data is organized enough to support AI use (clean, accurate, accessible)": 23.94% "in place now," 47.13% "in development." Verified at the NTEN publication page 2026-09-24. Figures are from the executive breakdown shown on that page; the full PDF sits behind a sign-in and was not opened.