Readiness Over Models: Why F&A AI Programmes Fail Before They Start

Most enterprise Finance AI programmes are solving the wrong problem. They’re asking “which model?” before they’ve answered “are we ready?” — introducing FARA, the Finance AI Readiness Assessment, and why structured readiness evaluation changes everything.

There’s a pattern I keep seeing across enterprise Finance & Accounting AI programmes, and it’s costing organisations far more than they realise.

A CFO sees a compelling demo. A transformation lead runs a pilot on invoice processing. The AI team ships something in eight weeks. Everyone calls it a success. Then, six months later, the same organisation is back at square one — wondering why nothing is scaling, why the model “keeps hallucinating,” why the finance team still doesn’t trust the outputs.

The problem is never the model.

It’s almost never the model.

The problem is that these organisations deployed AI before they understood whether they were ready for it. They skipped the uncomfortable diagnostic and went straight to the exciting part. And now they’re stuck — not because the technology failed them, but because they had no framework to understand what “ready” even meant.

That’s exactly why I built FARA — the Finance AI Readiness Assessment.

FARA — Financial AI Readiness Assessment

The Question Nobody Wants to Ask

When a client asks me “which LLM should we use for our accounts payable automation?”, my first response is almost always another question: “Tell me about your data quality in SAP.”

The silence that follows is diagnostic in itself.

Most enterprise F&A functions have been running on a combination of legacy ERP configurations, manual reconciliation workarounds, and tribal knowledge that lives entirely in the heads of people who’ve been there fifteen years. You cannot drop an agentic AI system on top of that and expect it to perform.

And yet, this is exactly what happens — repeatedly, across organisations of every size, in every geography. The AI gets blamed for problems that were baked into the foundation long before the first API call was made.

FARA exists to surface those problems before they become expensive failures. It’s a structured assessment instrument that gives you a weighted, scored view of your organisation’s actual readiness to deploy, scale, and operate agentic AI within F&A — across eight dimensions that matter.

What FARA Measures

The Finance AI Readiness Assessment scores your organisation across eight dimensions, each weighted by its criticality to agentic F&A deployment. The scoring runs from L0 (no activity) to L5 (fully autonomous and self-optimising), with a weighted composite score that maps to one of three readiness zones.

Here’s what each dimension is actually assessing — and why it’s in the framework.

1. Data Readiness (18% weight)

This is the highest-weighted dimension because it’s the single most common failure point. Can your agents actually access clean, structured, timely financial data? Is your ERP data in a state where it can be queried programmatically without a human cleaning it first? L0 here means no usable data layer. L5 means real-time, self-healing data pipelines that agents can trust without validation loops.

2. Agent Architecture (16% weight)

Are you still thinking in terms of single-task bots, or have you architected for multi-agent orchestration? This dimension assesses whether your technical infrastructure supports the kind of memory, tool-calling, and inter-agent communication that agentic F&A requires at scale. Most organisations I work with are sitting at L1 or L2 here — a single agent doing one thing reasonably well, nothing more.

3. Process Design (15% weight)

AI cannot standardise what hasn’t been standardised. If your month-end close process has seventeen variations across business units, an agent will faithfully replicate the chaos. This dimension looks at whether your F&A processes are documented, consistent, and designed in a way that agents can actually operate within — not just augment.

4. AI Governance (14% weight)

Who owns the outputs when an agent makes a wrong journal entry? What’s your escalation path when an anomaly detection agent flags something it doesn’t understand? This dimension is about whether you’ve built the policy infrastructure — CoEs, control frameworks, accountability structures — that lets AI operate in a regulated financial environment without creating audit nightmares.

5. Talent & Change Readiness (12% weight)

The most underrated dimension in every AI programme. Technology deployment is ten percent of the problem. The other ninety percent is whether your finance team understands what agents are doing, trusts the outputs enough to act on them, and has the capability to maintain and improve the system over time. L5 here means AI-native roles embedded into the F&A org structure itself.

6. Use Case Clarity (10% weight)

This one seems obvious and yet it’s where most programmes are weakest. “We want AI in finance” is not a use case. This dimension assesses whether you have a prioritised, ROI-mapped, sequenced backlog of specific F&A use cases — from reconciliation automation to cash flow forecasting to anomaly detection in GL postings. Without this, your AI programme is an answer in search of a question.

7. Risk & Compliance Controls (9% weight)

Particularly critical in regulated markets and GCC contexts. This dimension asks whether you have automated controls that can detect when an AI agent is about to do something it shouldn’t — and whether those controls are integrated into the workflow rather than bolted on after the fact.

8. Technology Infrastructure (6% weight)

Weighted lowest not because it’s unimportant, but because it’s the most addressable. Legacy ERP on-premises with no API layer is an L0. A cloud-native, API-enabled, Azure OpenAI-integrated stack with vector search is approaching L4. The reason this is last in the weighting is that organisations can move on infrastructure faster than they can move on culture, data, or process.

The Scoring Model

Each dimension is scored from 0 to 100. The composite score is calculated as:

Composite Score = Σ (Dimension Score × Weight)

Maximum weighted score: 500 points.

The three readiness zones map as follows:

Zone 1 — Foundational (0–150) You have pockets of AI curiosity but no systemic readiness. The right move here is not to deploy agents — it’s to invest in data infrastructure, process standardisation, and building a governance baseline. Deploying AI in this zone produces pilots that never scale.

Zone 2 — Operational (151–300) You have real AI deployments delivering real value, but they’re typically siloed, department-specific, and dependent on significant human oversight. This is where most large enterprises actually sit today. The work in this zone is integration — connecting your AI investments to each other and to the broader F&A operating model.

Zone 3 — Agentic (301–500) You’re running multi-agent systems with persistent memory, tool-calling capabilities, and meaningful automation of end-to-end F&A workflows. At this level, the conversation shifts from “how do we deploy AI” to “how do we govern a function that is increasingly AI-operated.”

How to Run the Assessment

FARA is designed to be conducted as a structured stakeholder workshop — not a survey that gets filled out in ten minutes and filed away.

The most useful inputs come from three groups:

The CFO and Finance Leadership team, who own the use case backlog and governance posture. The IT and Data Architecture team, who can speak honestly to the actual state of ERP data quality and infrastructure readiness. And the AI and Transformation team, who can assess the technical architecture and change management capability.

Running the assessment with only one of these groups gives you a partial picture — and it’s usually the most optimistic one.

The output is a scored radar view across the eight dimensions, with a composite score and a set of prioritised interventions mapped to the gap between current and target state. FARA intentionally produces a roadmap, not just a diagnosis.

What I’ve Learned Running This in the Field

A few patterns that have consistently shown up across the programmes I’ve delivered.

Most organisations overestimate their data readiness by two levels. There’s a consistent gap between what IT believes about data quality and what the finance team experiences every day in practice. FARA surfaces this gap explicitly, often for the first time.

Governance is the bottleneck nobody talks about. Companies invest heavily in technology and talent, then get stuck because nobody has made a decision about who’s accountable when an agent makes a wrong call. You need that decision before you deploy at scale — not after your first incident.

Use case prioritisation is driven by excitement rather than feasibility. The most commonly requested F&A AI use cases — cash flow forecasting, intelligent FP&A, autonomous close — are also among the hardest to deploy well. They require L4 or above on data readiness and process design. Starting there is almost always the wrong call. FARA helps organisations sequence their use case roadmap against their actual readiness profile.

The jump from Zone 2 to Zone 3 is qualitatively different from any prior jump. Moving from Foundational to Operational is largely a technology and process problem. Moving from Operational to Agentic requires a fundamental rethinking of how the F&A function is organised and governed. Organisations that don’t treat this as an operating model change — not just a technology upgrade — consistently struggle at this transition.

Why Readiness Matters More Than Your Model Choice

The LLM landscape is evolving fast. The model you choose today will be superseded within twelve to eighteen months. What won’t be superseded is your data layer. Your governance framework. Your process architecture. Your team’s ability to work alongside agents rather than just tolerate them.

The organisations that are going to win at Finance AI over the next five years are not the ones that picked the best model in 2024. They’re the ones that invested in the foundations that make AI actually work in a regulated, high-stakes financial environment.

FARA — the Finance AI Readiness Assessment — is a tool for building those foundations deliberately, and for knowing with precision where you stand and what needs to happen next.

If you’re leading an F&A AI programme and you haven’t done a structured readiness assessment, you’re flying blind. Not because the technology doesn’t work — but because you don’t yet know what it would take for it to work for you.

That’s the question worth answering first.