The CFO AI Maturity Model: 5 Levels (Where are you today?)
he winners in finance won’t be the teams with the fanciest AI. They’ll be the teams that turn AI into a repeatable operating capability — secure, controlled, and measurable.
At GrowCFO webinars, finance leaders are regularly asked where they are on their AI journey. The results are consistent: many teams are experimenting. Far fewer are building real capability.
That pattern is likely to continue as finance leaders examine what’s truly limiting AI implementation in FP&A. One common issue? Almost nobody has a shared language for what “good” looks like across copilots, embedded AI, and agentic systems.
To address this, GrowCFO has developed a simple CFO AI Maturity Model designed to help finance teams:
- Align leadership on ambition and pace
- Choose sensible pilots (and avoid random acts of AI)
- Build governance that enables speed rather than blocking it
A quick caveat: organisations often sit at different levels across different processes. It’s entirely normal to be Level 3 in Accounts Payable but Level 1 in forecasting.
The CFO AI Maturity Model (5 levels)
Level 1 — Ad hoc prompting (“AI as a helper”)
What it looks like
Individuals use ChatGPT / Copilot-style tools for drafting, summarising, Excel help, meeting notes.
Where the value comes from
Fast productivity gains on small tasks: commentary, emails, first-pass analysis, slide drafting.
Where it usually breaks down
inconsistent outputs (depends who’s using it and how)
data risk (“what can I paste into this?”)
no standard approach to quality or review
What to do next
Create the basics: a short AI policy, a “safe use” guide, and 10 proven prompt patterns for recurring finance outputs.
Level 2 — Standardised copilots (“AI with guardrails”)
What it looks like
The team has approved tools, shared templates, prompt libraries, and basic training. Usage becomes consistent and safer.
Where the value comes from
Broad-based productivity uplift across reporting, analysis, narrative, and admin.
Where it usually breaks down
At first sight, things look easy—but benefits plateau because the workflow is still manual. People are faster, but they’re still doing the same process.
What to do next
Pick one workflow and design for measurable change end-to-end (not just better writing).
Level 3 — Embedded automation (“AI in the workflow”)
What it looks like
AI becomes part of finance processes themselves: matching, coding suggestions, anomaly detection, variance explanations, exception queues.
Where the value comes from
This is where measurable operational impact starts to show:
- lower touch time
- fewer errors
- faster cycle times
Where it usually breaks down
“Black box” decisions with weak evidence trails, or teams over-trusting output without proper review.
On paper, this is where it should be easy: repetitive steps, clear rules, classic “low-hanging fruit” for Power Automate / Copilot Studio-style tooling.
In reality, the “simple automation” story often collapses fast. Every customer portal is its own universe. Permissions vary by user and entity. APIs are patchy, inconsistent, or simply not there. And the data needed isn’t neatly structured—it’s trapped in PDFs, email chains, and on-screen views.
What looked straightforward can turn into weeks of integration grind: access requests, connector workarounds, brittle scripts, endless edge cases. At that point, teams often conclude it’s easier—and more defensible—to build something purpose-built rather than forcing generic automation tools to behave like an integration layer.
What to do next
Define decision rights and controls:
- what AI can recommend vs do
- when a human must approve
- how exceptions are escalated
- what must be logged
Level 4 — Agent-assisted operations (“AI does the work, humans own outcomes”)
What it looks like
Agents execute multi-step tasks across systems—within boundaries:
- chase missing invoices or approvals
- reconcile items and propose resolutions
- assemble evidence packs for audit
- draft board narratives with linked support
- escalate exceptions to named owners
Where the value comes from
This is the step-change level. It’s not “faster analysis”—it’s less work to do.
Where it usually breaks down
Control design becomes the bottleneck:
- segregation of duties (SoD)
- identity and permissions
- audit trails and logging
- accountability when an agent acts
What to do next
Introduce an “agent control framework” (risk tiers, human-in-the-loop rules, and a clear audit trail).
Level 5 — AI-native finance operating model (“continuous, predictive, proactive”)
What it looks like
Near real-time close, predictive signals, proactive interventions. Finance shifts from production to stewardship and decision support.
Where the value comes from
This goes beyond process metrics—it changes the capacity and role of finance.
Where it usually breaks down
Less about technology, more about operating model:
- roles, responsibilities, escalation paths
- change management and adoption
- measurement and continuous improvement
- governance as a core capability
What to do next
Treat AI as an operating capability—like FP&A or controllership—owned, measured, and continuously improved.
A 2-minute self-assessment (copy/paste)
Score each statement 0–2
(0 = not in place, 1 = partial, 2 = consistent)
- We have a clear AI policy for finance (data, tools, do’s/don’ts)
- We have approved tools and guidance (not shadow AI)
- We have standard templates/prompt patterns for recurring finance outputs
- AI is embedded in at least one workflow (not just drafting)
- We have human-in-the-loop rules and escalation for exceptions
- We have logging/audit trail for AI-assisted decisions
- We can measure impact (cycle time, touch time, error rates, DSO, forecast accuracy)
- We have an owner for AI capability in finance (not “everyone and no one”)
Scoring guide
- 0–5: Level 1–2 (individual productivity)
- 6–11: Level 2–3 (team adoption + early workflow value)
- 12–14: Level 3–4 (workflow transformation underway)
- 15–16: Level 4–5 (operating model shift in progress)
GrowCFO welcomes insights from teams who try this exercise and reflect on where they land.
A practical example: variance commentary (how maturity shows up)
- Level 1: an analyst drafts commentary with AI help
- Level 3: AI generates variance drivers and a draft narrative linked to evidence
- Level 4: an agent gathers supporting schedules, flags anomalies, routes exceptions to owners
- Level 5: continuous monitoring reduces “close surprises” before month-end
That’s the shift: from “better words” → to “less work” → to “different operating model”.
One practical next step you can take this week
Choose one finance workflow and map it as:
Intake → Decision → Exception → Evidence → Output
Then ask two questions:
- Where is the most touch time spent?
- Where could AI reduce touch time without increasing risk (and with a clear evidence trail)?
If touch-time hotspots aren’t visible, it’s hard to pick the right pilot.
What’s coming next (and how this connects to GrowCFO)
Over the next few weeks, GrowCFO will break down the highest-value workflows to pilot (close, cash, spend, forecasting) and the governance patterns that enable speed without losing control.
And for those wanting to see this in practice:
- The GrowCFO Innovation Report (AI-native and agentic AI) publishes on 5th March (a structured view of the landscape + practical takeaways)
- The GrowCFO Tech Showcase is on 25th March (AI-native finance solutions demoed with CFO-first discussion)
If you’re navigating this journey now, these dates will be useful markers in your calendar.
Quick community question
Which level best describes your team today—1 to 5—and what’s the biggest blocker: data, controls, skills, budget, or leadership alignment?
