Finance
AI Exposes Gaps in Financial Data Infrastructure
Most FP&A teams are ready for AI. Their data infrastructure isn’t.
Ben Pierce
General Manager, Workday Adaptive Planning
Workday
For most of my career, the quiet truth of financial planning was that operational data always arrived late and incomplete. FP&A teams reconciled first and analysed second, pulling late nights on Thursday to iterate ahead of a Friday board meeting.
AI changes that dynamic. An AI model does not pause to ask whether a figure looks reasonable, nor does it route around data gaps the way an experienced analyst reflexively does.
Guide
CFO Playbook: Top 3 Outcomes of Agentic Finance
AI will not fix a broken data foundation. It will expose it.
For most finance teams weighing an AI investment, the AI model is not the primary constraint. The data infrastructure underneath is. AI will not fix a broken data foundation. It will expose it.
How the Workaround Became the Workflow
Every FP&A leader knows the muscle memory of the workaround. The ERP closes late, delivering high-level summary actuals rather than the underlying operational drivers – volume, backlog or unit-level changes – that actually drove the result.
As a result, analysts scramble across systems to fill in the gaps, requesting manual extracts from operations or pulling sales metrics from platforms never architected for financial rigour. Analysts force period discrepancies and category mismatches in order to reconcile actuals with general ledgers. Spreadsheet models become the default glue simply because they are flexible enough to force conflicting systems into the same table.
But the glue doesn’t stick. When finance, sales and operations arrive at an operating review with three conflicting versions of the same metric, meetings devolve into data reconciliation. Teams either retreat into offline modelling or make decisions on stale assumptions.
This was never an intentional strategy; the workaround became the operational workflow. Planning software historically assumed analysts would perform this heavy manual lifting before data reached the model. But as enterprise software multiplied across ERP, CRM, HRIS and supply chain, the tolerable became unmanageable.
As enterprise software multiplied across ERP, CRM, HRIS and supply chain, the tolerable became unmanageable.
AI Inherits Every Data Flaw
When data feeds an offline spreadsheet, an error remains contained. When data feeds an AI engine, that error scales instantly.
The quality of AI output directly reflects the data it can see. Fed a narrow or fragmented dataset, AI produces confident, narrow and potentially flawed analysis faster than human review can catch it. AI cannot transform bad inputs into good decisions, nor does it possess the human instinct to double-check an outlier.
Validation must happen in the infrastructure layer before data ever reaches the model.
AI Needs Context, Not Just Storage Volume
When AI models deliver generic or inaccurate results, the default reaction is to build a larger pipeline into a cloud data warehouse. That approach misdiagnoses a contextual problem as a storage volume problem.
What AI actually lacks is financial context – a semantic map of what numbers represent. Standard cloud data platforms require IT teams to flatten relational tables into static summaries, stripping away sub-ledger detail during traditional ETL processes.
As a result, an AI model can retrieve a line item without recognising that a cost spike represents a one-time GL reclassification rather than an operational trend. Without business logic, accounting hierarchies and metric definitions embedded directly in the data layer, AI reasons fluently over relationships it fundamentally misunderstands.
Building this semantic layer directly into the core infrastructure, rather than trying to bolt it on afterwards, is the critical requirement for modern planning architecture.
What’s Possible With an AI-Ready Data Foundation
A few finance organisations are already building this kind of data foundation, and the results are a useful gauge of what’s possible.
YES Communities manages more than 300 communities and 83,000 housing plots, a scale that had pushed spreadsheet-based planning well past what it could reliably handle. After rebuilding its planning and data environment, the company gained real-time visibility into operations at the community level, saving managers more than 10 hours a month they had previously spent chasing down reports.
Primient took on a different version of the same problem. A corporate spin-off had left it with a complex SAP environment and a four-week forecasting cycle. After rebuilding its data foundation, the company cut its forecasting cycle from 28 days to about four hours, raised costing accuracy to 99.7% and eliminated roughly $1 million in legacy software costs.
The Foundation Comes First
AI doesn’t replace human financial judgment – it just moves up when that judgment is needed. When context runs thin, models make assumptions, requiring finance leaders to validate variances and jump in to course-correct.
The newest, and most effective, planning tools don’t require that handholding. They unlock advanced capabilities with natural-language querying and AI-driven scenario modelling because the quality, accessibility and meaning of the underlying data is sound. The next era of planning won’t be won by the smartest model. It will be won by the team whose data the model can trust.
The finance teams that build a unified, context-rich data foundation first will be the ones leveraging AI to guide business strategy. Everyone else will remain stuck reconciling inputs.
The next era of planning won’t be won by the smartest model. It will be won by the team whose data the model can trust.
Ready to start building the AI-ready foundation your business needs? Explore Workday AI for FP&A.
Guide
CFO Playbook: Top 3 Outcomes of Agentic Finance
A practical framework for CFOs choosing AI-ready finance platforms, with the ROI evidence needed to build a business case.