This piece builds onto the reality that most CFOs aren’t held back by a lack of “AI” — they’re held back by the cost, risk, and disruption of replacing the core systems that run the business. The market is finally saying the quiet part out loud: you can’t bolt “agents” onto an organisation if the underlying ERP and process layer can’t reliably supply the data, controls, and execution paths that agents need. McKinsey recently framed this as an “AI agent vs ERP” divide — where investment pours into shiny AI use cases while ERP foundations and workflow clarity are neglected, limiting value at scale. (McKinsey & Company)
So the real question for legacy CFOs (Sage / Dynamics GP / older Oracle / QuickBooks + Excel, homegrown databases, shared inboxes) is not:
“When do we replace the ERP?”
It’s:
“How do we modernise outcomes now, without turning the next year into a transformation hostage situation?”
Why rip-and-replace fails (especially in finance)
The old modernisation playbook was simple: replace the system, then fix the process.
In reality, large ERP projects often fail CFO expectations because:
- they demand executive attention for quarters
- they introduce change fatigue across finance + ops
- they postpone outcomes until “go-live”
- they create new risk during migration, cutover, and stabilisation
Even mainstream IT leadership commentary has moved away from “rip it out” thinking toward incremental modernisation — making what you have work better while steadily reducing legacy constraints. (CIO)
And as AI enters the picture, this becomes more urgent: modernisation has to be workflow-first, not “platform-first.” Deloitte’s research on legacy modernisation with AI frames it as a staged approach: rethink processes, reengineer the digital core over time, and reimagine capabilities with AI — not one big-bang rewrite. (Deloitte Italia)
The better model: Edge Modernisation (legacy core + modern execution)
Think of your legacy ERP as the system of record. Keep it stable.
Then build a modern layer around it that does three jobs:
- Connect: pull the minimum viable data from the systems where work lives
- Orchestrate: route tasks through controlled workflows (exceptions, approvals, escalations)
- Prove: generate an audit trail + evidence links by default
This aligns with Gartner’s push toward a more composable approach in ERP initiatives — consolidating and modernising in parts, creating reusable capabilities across the enterprise rather than betting everything on a monolith replacement. (Gartner)
And it’s consistent with how current market analysis describes modernising legacy infrastructure: upgrading with AI, cloud, and analytics without starting from scratch. (CIO)
Why “dashboards” and “chat” plateau on legacy stacks
Most AI products aimed at finance start with: “ask questions over your data.”
That’s useful, but it hits a ceiling fast on legacy stacks because:
- Data is incomplete (critical info sits in emails, PDFs, spreadsheets, ticketing, shared drives)
- Context is missing (why something happened, what exceptions occurred, what approvals were given)
- There’s no workflow state (who owns this, where it’s stuck, what should happen next)
- There’s no execution path (the AI can tell you; it can’t safely do the next step)
- Auditability breaks (answers without evidence links don’t survive scrutiny)
That’s why many organisations adopt AI in phases: integration complexity, data quality, and governance are the real blockers — not model capability. (PYMNTS.com)
So, if you’re on legacy systems, the winning move is to shift from: Reports → Dashboards → Chat to: Exceptions → Actions → Evidence
The CFO’s “3-workflow wedge” (where legacy teams win fast)
If you want traction without ERP replacement, don’t modernise everything. Modernise three workflows that create visible value and reduce chaos.
1) Order-to-Cash exceptions (cash impact)
Goal: reduce disputes, accelerate collections, stop leakage.
- dispute intake (email/portal) → categorise → assign owner
- link to invoice / delivery evidence / contract terms
- route for approval (credit, rebill, write-off thresholds)
- push back actions into the core system (or produce a controlled task list)
Why it works on legacy: you don’t need a new ERP to stop exceptions. You need routing, evidence, and control points.
2) Procure-to-Pay leakage and compliance (margin + control)
Goal: catch duplicates, enforce approvals, reduce “maverick spend.”
- vendor onboarding checks (bank details, duplicates, approvals)
- invoice ingestion → match signals (PO/GRN/reference)
- exceptions queue with reason codes
- approval matrix enforcement and audit log capture
Why it works on legacy: AP is where “work-between-work” explodes; orchestration removes the rework loop.
3) Close-week exception management (close speed + confidence)
Goal: make close predictable by killing upstream chaos.
- track reconciliations and missing support as exceptions
- auto-chase missing evidence via workflow nudges
- standardise journals with approval/evidence pack generation
- produce a “close control dashboard” (not a reporting dashboard)
Why it works on legacy: close pain is mostly exception pain. Handle exceptions systematically and close time falls without swapping the ledger.
What “finance-grade execution” actually means
This is the bar CFOs should demand from any modern layer:
- Controlled actions: the system can propose actions, but execution follows policy gates
- Clear ownership: every exception has an owner, SLA, and escalation
- Evidence-first: every action links back to source documents/approvals
- Permissions-aware: users only see what they’re allowed to see (especially across shared drives and inboxes)
- Observable operations: you can see what’s stuck, what failed, and why
McKinsey’s AI–ERP framing is blunt: value comes when you clarify value at the workflow level and make the ERP/process foundations capable of supporting AI at scale — not by running disconnected “AI pilots.” (McKinsey & Company)
The 100-day playbook (legacy-friendly and board-safe)
Days 0–15: Baseline the exception economy
- Pick 1 workflow (start with O2C disputes or AP invoice exceptions)
- Measure: exception volume, cycle time, root causes, rework loops
- Define 3 control points (where approvals/evidence must exist)
Days 16–45: Build the execution layer around exceptions
- Standardise intake (one queue, tagged reasons)
- Route to owners with escalation
- Generate evidence pack links automatically
- Create “action outputs” back to the core system (or controlled task list)
Days 46–75: Add governance and resilience
- enforce approval matrix + thresholds
- instrument audit logs
- add monitoring (failure modes, queue backlog alerts)
- tighten permissions boundaries
Days 76–100: Scale the pattern to the second workflow
- replicate the playbook to AP or close exceptions
- publish KPIs weekly: exceptions down, cycle time down, cash/close confidence up
This is incremental modernisation in practice — consistent with mainstream CIO guidance that modernisation can be phased and need not be rip-and-replace. (CIO)
If you’re on legacy systems, you don’t need a miracle ERP project to modernise finance.
You need an operating model that:
- surfaces exceptions,
- routes actions with controls, and
- produces audit-ready evidence by default.
That is how you earn outcomes now — and still keep the option of an ERP upgrade later, on your timeline.
At Farben, we’re building an agent layer designed for this exact reality: legacy cores + modern execution. The goal isn’t to replace your system of record — it’s to reduce exceptions, enforce controls, and drive outcomes (cash, close, leakage) while keeping humans in the right approval moments.
If you like our articles, consider subscribing!