AI Won’t Fix Your Consolidation Problem — But Here’s What It Can Do for a Group CFO

Every Group CFO is receiving AI pitches right now. The use cases - automated variance commentary, cash flow forecasting, anomaly detection across entities - are genuinely compelling. What the pitches do not show is the data infrastructure those use cases require to work reliably in a multi-entity group. AI in finance is not a shortcut to a clean consolidation. It is a capability layer that sits above one. Deployed in the wrong sequence, it does not solve the underlying problem. It automates and amplifies it.

Why AI fails in groups that skip the foundation

The fundamental problem with AI in finance is simple and frequently ignored: AI applications inherit the data they are connected to. A group CFO who deploys a forecasting AI on top of unconsolidated entity data receives confident-looking forecasts built from incompatible inputs. A narrative AI connected to unadjusted intercompany positions writes plausible variance commentary explaining fluctuations that are definitional artefacts rather than operational signals. The AI is not producing incorrect outputs in the conventional sense. It is accurately describing a broken data foundation - and expressing that description in fluent, credible language that is harder to identify as wrong than a cell reference error in a spreadsheet.

Gartner's survey of AI adoption in enterprise finance functions identifies data quality as the primary barrier to value delivery in 68% of implementations that underperform against expectations. In multi-entity groups, the figure is likely higher because the data quality problem is structural - it cannot be resolved by cleaning a single dataset. It requires standardising definitions, automating elimination, and building a GL-anchored semantic layer that computes metrics deterministically before any AI application can reason about them reliably.

McKinsey Global Institute research on AI in finance estimates that finance functions with a standardised data foundation extract 4 to 6 times more value from AI investments than those without one. The difference is not in the AI models. It is in the quality and consistency of the data the models process.

Three AI applications that deliver real, measurable value - once the foundation is in place

Intercompany reconciliation and automated elimination. In a group with significant intragroup activity, the manual work of identifying, reconciling, and eliminating intercompany positions is one of the highest-cost, highest-risk steps in the monthly close. It is performed by knowledgeable individuals applying undocumented methodology, which makes it slow, error-prone, and not resilient to personnel change. AI-assisted intercompany matching - where the system automatically identifies matching payables and receivables across entities, flags discrepancies for human review, and proposes elimination entries - is the highest-ROI AI application in group finance. In groups with more than four entities and material intragroup flows, it routinely reduces IC reconciliation time by 60-75% and reduces error rates by eliminating the manual pattern-matching step that is the primary source of month-end adjustment errors.

Anomaly detection at entity level, before problems surface at group level. A Group CFO managing eight or twelve entities cannot monitor each one with the granularity required to catch problems early. AI-powered anomaly detection - watching for entities whose margins are drifting outside historical norms, whose receivables aging is accelerating relative to revenue, or whose cost ratios are diverging from group benchmarks without a clear operational explanation - provides an early warning layer across a group that is structurally too complex to monitor manually. The value is in the timing. Catching an entity-level margin deterioration in week two of the quarter rather than discovering it in the board pack changes the range of management responses available. Variance commentary at month-end explains what happened. Anomaly detection at week two creates the opportunity to intervene before the quarter is fully committed.

Scenario modelling across business units with different revenue dynamics. In a group with multiple business models - a subscription-based SaaS entity alongside a project-based professional services entity alongside a distribution business - understanding how an entity-level change flows through to group-level EBITDA requires modelling work that is time-consuming when done manually. A lost contract in the services entity, a currency movement in the Czech operation, a cost increase in shared services - each of these has a group P&L impact that depends on entity-level revenue dynamics and intercompany relationships that are difficult to model quickly. AI-assisted scenario modelling, connected to a clean data foundation, allows the Group CFO to run these analyses in minutes rather than days - and to answer the board's scenario questions in the meeting rather than in the follow-up pack.

Three AI pitfalls specific to multi-entity groups

Automating a broken process amplifies the error, not just the speed. The most common AI mistake in finance is automating an existing manual process without first verifying that the manual process is correct. If intercompany positions are not properly reconciled in the current manual process - which in most groups they are not - an AI system that automates the reconciliation step will produce incorrect eliminations faster and at greater scale. The speed advantage compounds the error rather than containing it.

Narrative AI on unconsolidated data is a governance risk, not just an accuracy risk. A variance commentary generated from unadjusted entity data may explain movements that are partly definitional, partly operational, and partly intercompany artefacts - all in the same paragraph, presented as if they were a single coherent story. A board or audit committee that reads this commentary as authoritative is making decisions on analysis that cannot be independently verified. The fluency of the narrative makes this harder to catch than a cell-reference error in a spreadsheet. In a regulated or PE-reporting context, this is not merely an accuracy problem. It is a governance problem.

AI forecasting without a deterministic data layer creates an audit-proof gap. When a board member or auditor asks how a consolidated forecast number was computed, the answer must be traceable to specific inputs and specific calculations. An AI model that generates forecasts from patterns in historical data cannot provide that traceability in a way that satisfies an audit or a fund's reporting requirements. The deterministic layer - which computes numbers from GL-level data using defined rules - must exist before AI is added on top. AI explains and reasons from that layer. It does not replace it.

The correct sequencing: what groups that get this right do differently

Finance functions that have deployed AI successfully in a multi-entity group context share one practice: they built the data foundation first, validated the reliability of BI outputs, and then identified narrow, high-ROI AI use cases that sit above a trustworthy data layer. They did not begin with the AI use case and build backwards.

The implication is direct: an AI Finance Readiness Assessment before any AI project is scoped is not a delay tactic. It is the step that determines whether the investment will produce a trustworthy, auditable output - or a well-funded repetition of the same data quality failure that has already consumed previous technology budgets in the group.

If you are evaluating AI for your group's finance function and want a grounded view of where the value is and what the prerequisites look like in your specific structure - talk to us about an AI Finance Readiness Assessment before the scope and budget are set.