I read a16z’s “You Need a New CFO.” It was actually pretty good.

The article describes how the CFO’s job has changed: from keeping the books, to raising capital, to planning and partnering with the business. Now, the authors say, AI is changing the role again. It can assemble data more easily, do more of the work, and help finance teams operate in real time. CFO’s are now builders.

There’s a lot to that. But I think the article oversimplifies the data problem. AI has made it easier to find and pull data. It hasn’t solved the harder work of defining what data the business needs, how it should be captured, and where it belongs.

Take a key piece of information buried in an email. Yes, AI can search for it, just as it can search the volumes of data on the internet. But defining what is relevant and capturing it so it isn’t buried in the first place beats searching the forest for it with really fast search dogs.

And searching unstructured data brings its own risk. The email may be ambiguous, outdated, or incomplete. AI can extract a meaning that was never intended and present it with confidence. The answer still depends on the foundation underneath: data defined in a way that is meaningful for this business and the decisions it needs to make.

Take the sales pipeline. Sales and finance may use the same words—“lead,” “qualified,” “likely to close”—and mean different things. A promising prospect to one person may be a remote possibility to another. AI can build a very cool pipeline tool, with real-time dashboards and forecasts. But if the business hasn’t agreed on what counts as a prospect, what qualifies an opportunity, or what evidence supports a close probability, the tool can turn inconsistent assumptions into precise-looking numbers.

Those definitions belong to the business, and that’s where AI becomes transformational. AI can help us work them out before we build: map the flow, expose disagreements, compare proposed definitions with past results, and turn an agreed model into fields and controls. That is a more valuable use of AI than asking it to interpret inconsistent meanings after the data has already been collected.

This is where the “new” CFO can matter. The opportunity isn’t only to use AI to do the old work faster. It’s to use AI to help design the work before the systems are built. What information does this business need? Where should it be created? Where should it live? Who owns it? What decisions depend on it? Map the flow first. Find the duplication and roadblocks before they become part of the process.

No building is constructed without the engineers’ plan and the architect’s drawing. And physics is still physics. AI can help us plan and build faster, but it can’t make the underlying rules disappear.

Imagine a world without bank reconciliations. We could use AI to complete them faster, matching transactions across systems and flagging exceptions, but we could also ask why we even need reconciliations in their current form. Could the process capture each transaction once, preserve an immutable history, and apply controls as the transaction moves? Could those process controls satisfy the purpose of the reconciliation before the end of the month?

That’s where blockchain-type processing and AI planning might meet: a dependable, append-only record of what happened, with controls built into the flow. Corrections still happen, but the history remains visible. The goal isn’t to make the control disappear. It’s to design the process so it works as the business operates, instead of reconstructing the evidence afterward.

Collections offers another example. We build systems around the different data structures clients use to manage their portfolios of debtors. The client’s format can become the organizing principle. AI helps us turn that around: first design a coherent data model for the work itself, then translate to and from each client’s proprietary format. That translation is the supplemental build. The core is the flow of data within our own operation.

If a client needs us to track a data element we have no defined place for in our data map, that should stop us and trigger a design question. Is it truly unique to that client, or did we fail to anticipate a meaningful part of the work? A wise old mentor once told me there’s nothing new under the sun. AI can help us compare the new requirement with what we already know, identify overlaps, and catch what we missed before we build around the gap.

Define the data and put it in the right places, and operations and finance can work from the same picture: common data, common dashboards, decisions made with a shared understanding of the business. That foundation is the work of the business, not just the CFO.

The a16z article names Rillet among its portfolio examples. That context matters; the article is both an argument about the future CFO and a pitch for products in its portfolio. Rillet describes itself as an AI-native ERP with a rebuilt general ledger and accounting workflows. The interesting question is not simply how fast AI can work inside an ERP. It’s whether the business has designed the information and processes the ERP needs to make that speed useful.

Declaring victory because AI can search the data mistakes retrieval for design. The greater value is using AI to plan what the business needs to capture, define it, and build the flow that puts it where it belongs.

Search finds what got lost. Design reduces what gets lost.