Move to another large ERP, knowing exactly what changes
Oracle Fusion Cloud ERP, Workday or Microsoft Dynamics 365. Agents extract how your business actually runs and map it to the target's standard processes, so you adopt the standard with your eyes open.
Who this is for
The ECC deadline forces a decision, and some companies conclude that the next decade should not be spent on the same vendor. Common triggers:
- Vendor leverage. Credible alternatives change the negotiation, even if you end up staying.
- Fit. Your business has moved on from the processes SAP was configured for a decade ago, and the target's standard looks closer to what you do now.
- Consolidation after M&A. Several ERPs, several charts of accounts, one finance team. A deadline is a good moment to collapse the estate onto one platform rather than convert each system separately.
- A cloud-first mandate from the board or the parent company.
We use Oracle Fusion Cloud ERP as the worked example below. The method is the same for Workday and Dynamics 365. We have no stake in which one you pick, and we say so in the assessment if staying on SAP is the better answer.
Adopt, not adapt, knowingly
Large ERP switches usually go wrong in the same way. The target is configured to look like the old system because nobody can state precisely what the old system does. Customization piles up, costs grow, and the benefit of a standard platform disappears.
Our approach reverses the order. Agents read your configuration, custom code, document flows, usage and interfaces, and extract your business rules into a semantic model: entities, rules, mappings. A patch panel between what you run now and any target. We then map each rule to the target's standard process and sort them into three bins:
- Maps cleanly. The standard does the same thing.
- Changes. The standard does it differently. You see the new way, and who is affected.
- You would lose it. A rule or control that has no standard equivalent. You decide to adapt the process, extend the target, or keep it.
That list is what executives need before committing, and it exists before a single line is configured.
A cautionary tale. Birmingham City Council replaced SAP with Oracle Fusion. The initial estimate was £19M plus £1M contingency; the forecast is now £144.4M, with go-live slipping and the system being reimplemented.1 It is a public-sector outlier, and we do not know all the reasons. What it shows is how an army-run switch can drift when requirements are not pinned down and no one can say what the old system did. Pinning that down is what the assessment is for.
What agents do, and what people do
Agent shares are model assumptions, by hours of work. assumption
| Workstream | Agents | People | Agent share of hours |
|---|---|---|---|
| Business-rule extraction | Read configuration, code, documents and logs; draft rules with evidence links | Process owners confirm, correct and rank rules | high |
| Fit-gap against the target | Map each rule to standard processes; produce the change and loss list | Target-platform experts and your executives decide adopt, extend or keep | high |
| Configuration and integration | Build configuration, interface mappings and conversion programs | Review; own design decisions | high |
| Data migration | Mapping, cleansing, trial loads, reconciliation | Finance signs off balances | 80% |
| Testing | Generate and run scenarios from observed behavior | Design critical cases; user acceptance cycles | 85% |
| Change management | Role guides, how-do-I support | Lead the organization; train | 55% |
| Cutover | Runbooks, rehearsal tracking | Run the cutover; make go/no-go calls | 45% |
Only the workstreams with a stated percentage come from the model's per-workstream settings; the rest vary by project. Overall, agents absorb about 74% of hours.
How it runs
Assessment (8 weeks for a large estate)
Build the semantic model. Draft the fit-gap against your shortlisted target. Deliver a recommendation, including "stay on SAP" if that is what the evidence says, and a fixed-fee quote.
Decisions on the change and loss list
Your executives rule on each material difference. This is the main calendar floor on this path.
Build and migrate
Agents configure, map and convert; people review. Trial migrations repeat until reconciliations hold.
Test and rehearse
Regression generated from observed process behavior, user acceptance cycles, mock cutovers.
Go-live, parallel run and hypercare
Run old and new side by side through at least one period-end close where scope allows. Hypercare ends on agreed exit criteria.
Typical published ranges for Oracle Fusion Cloud ERP are 6–9 months for core finance at a single mid-market entity and 18–36 months for a global multi-country rollout (low-confidence source).2
What it looks like at scale
Example: a company with $5–20B in revenue, moderate customization, switching to a new ERP. Planning estimates, not quotes. assumption
| Delivery model | Delivery cost | Calendar | Team |
|---|---|---|---|
| Traditional SI, hourly | $36M | 33 mo | large |
| SI with AI copilots | $30.6M | 30 mo | large |
| Our small team, without agents | $33.2M | 35 mo | 18 |
| Our small team, with agents | $17.7M | 14 mo | 18 |
The same 18 people without agents would need 35 mo, longer than the traditional SI's 33 mo. Agents are what make the small team work.
What you pay
Be clear about this. A switch costs more one-time than staying on SAP. The “current path” here is an SI-led SAP-to-SAP program at $24M; even with agents, a switch ends up only $1.9M cheaper than that, because the work is bigger. Against a traditional SI doing the same switch ($36M) the gap is wide, but that is not the comparison that decides it. Do not choose Switch to save money. Choose it for leverage, fit, or consolidation. The 5-year view, which counts the run cost of the new platform, is about $1.9M better than the current path, essentially the same. We model run cost on the new platform as unchanged (about $10M a year, an assumption).
Risks on this path, and how we handle them
| Risk | What we do |
|---|---|
| Requirements are fuzzy and the project drifts | Rules are extracted from evidence before configuration starts. Every difference is listed and decided. |
| Controls and audit trails are lost in the move | The loss list includes controls. Finance and audit review it before build. |
| Target-platform detail is wrong | Experts on the target platform review all fit-gap output. We bring specialists from our network, and we work with the target vendor's channel where it helps. |
| The organization rejects the standard | People see the change list early, with the affected roles. Change management is staffed as its own workstream. |
| Overrun, as in the Birmingham case | Fixed fee set after assessment, 15% contingency inside it, success fee only on acceptance. Parallel run before decommissioning. |
Sources
- The Register, 29 Jan 2026. Birmingham Oracle ERP fiasco now £144M and still not working.
- ERP Research. Oracle ERP Cloud implementation. Low confidence; treat as typical published ranges.