The 44/21 gap
In a McKinsey survey of 102 CFOs, 44% reported using generative AI across five or more finance use cases in 2025McKinsey, “How finance teams are putting AI to work today” (2025). 44% of 102 surveyed CFOs used generative AI across five or more finance use cases in 2025, up from 7% a year earlier.—up from just 7% a year earlier—yet Deloitte finds that among finance functions that have fully deployed AI, only about 21% can point to measurable valueDeloitte 2026 Finance Trends (via CFO Dive). Among finance functions that have fully deployed AI, only ~21% believe the investment has delivered tangible, measurable value to date.. That gap is the most important number in enterprise finance right now, because it is not a story about which companies bought AI. It is a story about which companies built the foundation that lets AI work. The unpleasant truth for finance leaders is that most stalled AI initiatives did not fail because the models were weak. They failed because the models were bolted onto slow processes and incomplete, inconsistent data — and no model overcomes that.
This article explains why AI finance projects stall, what an "AI-ready" data foundation looks like, and how to revive an initiative that has gone quiet without ripping out your ERP.
Why projects stall: it is the data, not the model
When an AI finance pilot disappoints, the post-mortem usually surfaces the same root causes — none of which are about the AI itself.
- Inconsistent, incomplete data. The same customer, vendor, or account is represented three different ways across systems. The model learns the inconsistency and produces results no one trusts.
- A complicated chart of accounts. Years of additions and one-off accounts have turned the COA into something only a few people can navigate. AI mapping and automation inherit that complexity and amplify it.
- Disconnected systems. The ERP, the billing system, the CRM, and the spreadsheets in between do not agree, and reconciling them is a manual job in itself. AI cannot reason over data it cannot reach cleanly.
- AI layered onto a broken process. Automating a flawed close just produces flawed results faster. Speed without process discipline is not a win.
- No baseline and no instrumentation. Without a measured starting point, even a project that worked cannot prove it did — so it loses executive support.
In other words, AI is only as good as the data and process beneath it. Strengthening that foundation is what allows AI to deliver reliable, measurable results — and skipping it is the single most common reason initiatives stall.
What an AI-ready data foundation looks like
"Fix the data" is easy to say and vague enough to be useless. Concretely, an AI-ready foundation in a mid-market finance function has a few recognizable properties:
- Clean, governed master data. Customers, vendors, items, and accounts are deduplicated, standardized, and owned — with rules that keep them clean going forward.
- A rationalized chart of accounts. The COA reflects how the business is actually run and reported, without the archaeological layers.
- Integrated systems. The ERP and the surrounding systems share data via reliable integrations rather than re-keyed exports, ensuring a single version of the numbers.
- Documented, disciplined process. The close, order-to-cash, and procure-to-pay run as defined processes — the precondition for automating them safely.
- Measurable baselines. Close days, DSO, error rates, and hours are instrumented, so the impact of any AI initiative is provable.
This is unglamorous work, and that is exactly why it gets skipped in the rush to a flashy use case. It is also the work that determines whether the use case ever delivers.
The fix does not mean re-implementing your ERP
Finance leaders often assume that "fix the foundation" means a multi-year ERP re-implementation. Usually it does not. The more common — and far less disruptive — path is to clean and integrate your data, redesign the process, and optimize your existing ERP into an AI-ready state, whether that is NetSuite, Sage Intacct, SAP, or Oracle Cloud ERP. A full re-implementation is reserved for the cases that actually need one. The point is to make the foundation you already own trustworthy, not to start over.
How to revive a stalled initiative
- Run a data-quality and AI-readiness assessment. Audit your ERP, chart of accounts, integrations, and the specific data the stalled use case depends on. The output is a clear, prioritized remediation plan — not a vague "improve your data" recommendation.
- Remediate in priority order. Fix the master data and integrations that the highest-ROI use cases actually need first, rather than trying to boil the ocean.
- Re-sequence the use cases. With the foundation addressed, the previously stalled initiative — or a better-chosen one — is re-launched on data it can trust.
- Instrument and prove it. Measure against the baseline so the revived project earns back executive confidence with numbers, not anecdotes.
Where to start
If an AI initiative has gone quiet, the first move is not another tool or another model — it is a clear-eyed look at the foundation beneath it. A data-quality and AI-readiness assessment audits your ERP, chart of accounts, and integrations and hands you a prioritized remediation plan, so the path back to ROI is clear before any major commitment.
That is the core of our ERP + AI Data-Foundation Modernization service — the prerequisite most AI projects skip — and it begins with the same AI Readiness Assessment that grounds every Strategic Move engagement in a measurable business case.
The takeaway
The 44/21 gap is not a verdict on AI. It is a verdict on foundations. The companies seeing ROI are the ones that fixed their data, rationalized their chart of accounts, integrated their systems, and disciplined their processes before — or alongside — deploying AI. Stalled initiatives are rarely revived by a better model; they are revived by the unglamorous foundation work that should have come first. Fix the foundation, and the AI you already bought will finally pay off.