The finance team loves spreadsheets, so we didn't build a spreadsheet-heavy workflow. They built them because the systems they paid millions of dollars to buy couldn't answer the questions they needed answered.
Over the past 20 years, corporate finance software has gotten very good at storing data. ERP, billing systems, financial planning platforms, and financial infrastructure tools serve as systems of record for nearly all financial transactions within a company.
But much of the real work in finance is done elsewhere.
When executives ask why revenue fell short of expectations, which customers caused churn, why margins were compressed, and whether hiring plans are on track, finance teams rarely answer these questions directly in ERP. Export the data to a spreadsheet. They bring the information into Snowflake or BigQuery. Build dashboards in Tableau, Looker, or Power BI. They create financial models in Excel that are external to the systems where the underlying data originates.
Financial software has become the source of truth, and spreadsheets, warehouses, and BI tools have become the place to interpret truth. Over time, financial organizations built entire operating layers on top of their systems of record, many of which were held together by organizational knowledge rather than formally defined definitions.
Last mile financial stack
Most finance leaders can probably sketch this architecture from memory.
Financial data originates within operational systems. Teams can move that data into warehouses, spreadsheets, dashboards, and reporting environments where they can manipulate it, combine it, and answer business questions.
Businesses rarely operate the way software vendors expect. Revenue recognition is customized. Predictive models evolve. The department develops company-specific indicators. Board reporting requirements change. Acquisitions introduce new definitions and processes.
Finance teams responded by building their own layer on top of their systems of record. Spreadsheets offered flexibility that packaged software couldn't provide. Warehouses have made it easy to combine information from multiple systems. BI tools now allow you to distribute reports throughout your organization.
Although the issue was resolved at all tiers, it also created an opportunity for inconsistency at all tiers.
Over time, financial logic became distributed across spreadsheets, dashboards, warehouses, and planning systems. Organizations have adapted by creating processes to coordinate information and keep teams aligned even when the underlying systems are not in place.
This coordination exists because the information needed to run a business has become fragmented across dozens of tools and workflows.
Why AI changes the equation
AI agents introduce a different operating model. Instead of pulling information from a system and navigating through a series of spreadsheets, dashboards, presentations, and emails, agents can directly query the underlying system, perform analysis, and return answers.
This distinction is important because much of the last-mile financial stack exists to move information from one place to another. Analysts extract data from one system, transform it in another, visualize it in a third, and distribute it to a fourth.
When software can directly access the underlying records and perform analysis on demand, some largely manual workflows start to look less like permanent requirements and more like workarounds accumulated over time.
Finance teams will continue to use spreadsheets, warehouses, and dashboards. What can change is how often information needs to move between them before someone can answer a question.
The bigger change is that analysis begins to occur closer to the system of record itself. In environments with strong governance, finance leaders no longer need to know which systems contain the answers. When they ask questions, they can trust that the software will retrieve relevant data, apply consistent definitions, and return reliable results.
The problem wasn't the spreadsheet.
Spreadsheets have become a convenient villain in financial discussions, but they weren't the root problem.
Finance teams use spreadsheets because they provide flexibility and transparently auditable logic. Analysts can test assumptions, build models, investigate anomalies, and answer questions that packaged software cannot.
The real problem isn't the spreadsheet. It's a lack of definition of governance across finance, sales, operations, and leadership.
Assumptions for revenue, churn, ARR, customer counts, and forecasts often exist in multiple systems with different interpretations, ownership, and business logic. Over time, these differences become embedded in dashboards, planning models, warehouse transformations, and spreadsheets.
Ultimately, no one can explain where all the numbers came from without manually tracking the entire chain.
AI doesn't solve that problem. Often it exposes it.
If finance, sales, operations, and executive management all maintain different definitions, agents won't be able to reliably answer financial questions.
Governance moves to the center
For years, finance teams have compensated for inconsistent systems with human effort.
Analysts knew which spreadsheets contained approved metrics. Financial managers knew which reports represented official numbers. Institutional knowledge bridged the gaps between systems.
Agents do not act on the basis of organizational knowledge. They operate based on a written context of definitions, permissions, and data structures.
If finance, sales, and operations have different definitions of revenue, automation won't resolve the discrepancy. You can simply get answers faster than before.
Finance teams have spent years solving data collection problems. Agents can automate much of that work. What cannot be automated is governance.
Someone needs to decide what counts as revenue, how churn is measured, which forecasts are reliable, and which versions of metrics belong in board materials.
As automation capabilities increase, these questions become even more important.
What CFOs should evaluate
Much of the evaluation of financial software still centers around dashboards, reporting features, and user experience.
While these factors are important, finance leaders increasingly need to evaluate a variety of questions as well. for example:
- Can the system expose data cleanly?
- Can you support consistently managed definitions across departments?
- Can you explain how you arrived at your conclusion?
- Can you provide the controls you need for financial oversight?
- Can it serve as a reliable source of truth for both humans and machines?
These questions are important because agents are starting to plan their analysis rather than just doing what humans have planned.
Finance built a last-mile stack because systems of record alone can't answer the questions a company needs to answer. Warehouses, spreadsheets, dashboards, and planning models filled the gap.
Agents attack the same problem from different directions. Rather than moving between systems, agents use managed definitions to perform analysis across multiple systems.
Whether that eliminates a large portion of the last-mile stack remains to be seen. What is already clear is that automation increases the costs of definitional mismatches. Human analysts can usually tell when two reports don't match. Agents confidently return answers provided by the underlying system.
As such, the definition of governance, ownership, and finance will be at the center of the conversation. Finance teams have always been responsible for these things. AI makes it much harder to ignore the consequences of mistakes.
