Packaged ERP was sold on one promise: you shouldn't build what everyone else already has. That's still true for much of the back office. But AI has shifted two things, the cost of building software and the value of owning your data and process logic, so the line between buy and build has moved. This post gives you a vendor-neutral way to decide where it sits for each workflow, and a scoring sheet you can fill in this week.
The short answer
- Keep commodity processes like the general ledger, payroll, and statutory reporting on packaged software. Consider building where you compete.
- Score each workflow on six factors: differentiation, change frequency, customization already paid, integration, data and AI value, and risk.
- AI can lower the cost of building, but the evidence is mixed, and it doesn't remove the cost of running and supporting what you build.
Decide by workflow, not by system
"Should we replace our ERP?" is the wrong unit of decision. Whether you run SAP, Oracle, Dynamics 365, or something else, your ERP is dozens of workflows, and they don't all deserve the same answer.
Gartner's pace-layered application model is a useful frame. As CIO Wiki summarizes it, applications fall into systems of record (core transaction processing, with a low rate of change), systems of differentiation (unique company or industry processes), and systems of innovation (new applications built for new opportunities). Packaged ERP is strongest in the first layer. Most of the pain, and most of the custom code, lives in the second.
What AI changes, and what it doesn't
Building may be cheaper, but don't bank on a number. In a controlled experiment by researchers from GitHub and Microsoft, developers with an AI pair programmer finished a defined coding task 55.8% faster. A 2025 randomized trial by METR found the opposite for experienced developers working in their own large codebases: they took 19% longer with AI tools. METR has since said those results reflect early-2025 tools. Treat any claim of dramatic savings with care, ours included.
Owning your data and logic is worth more. Automation and AI agents work best on clean, well-structured data with clear interfaces. When a workflow's logic lives inside a vendor's configuration, you add AI where and when the vendor's roadmap allows.
Buying carries pricing risk you don't control. When SAP replaced its RISE package with SAP Cloud ERP Private in 2025, The Register reported Gartner's warning that some customers could see higher list prices per user equivalent.
What doesn't change: software you own still needs hosting, security, support, and people who understand it. IT leaders surveyed by TechTarget list ongoing maintenance of custom code among ERP's hidden costs, and that applies to anything you build.
The six factors
1. Differentiation
Does this workflow win or keep customers? Quoting and product configuration, production scheduling, supplier tracking, and project controls often do. Accounts payable rarely does.
2. Change frequency
How often do you change the process, its rules, or its reports? Frequent change on a packaged system means repeated configuration work, consultant time, and regression testing.
3. Customization already paid
Where you've written custom code or bought bolt-ons to make the package fit, you've already voted with money. Check usage before you count it: VendorBenchmark, which sells SAP benchmarks, estimates 15–40% of SAP custom code goes unused for 24 months or more.
4. Integration
Owning a workflow can simplify a tangle of interfaces, or add one more system to keep in sync. Score 5 if owning it would remove integrations and 1 if it would add them.
5. Data and AI value
Would better data or AI on this workflow move a number you track, like first-time fix rate, days sales outstanding, or forecast accuracy? If yes, owning the data model is worth more.
6. Risk
What happens if it breaks or gets something wrong? Statutory, tax, and audit-critical processes benefit from a vendor that ships legal updates. Score this one in reverse: 5 for low risk, 1 for high.
A simple scoring approach
Score each workflow from 1 to 5 on each factor. Weight differentiation and data and AI value double, because they drive the upside:
Build score = 2 × Differentiation + Change + Customization + Integration + 2 × Data and AI + Risk (reversed)
Scores range from 8 to 40. Use three bands:
- 30 or more: strong build candidate.
- 20 to 29: hybrid. Keep the record in the package and build the workflow layer on top.
- Under 20: keep it on packaged ERP.
| Workflow | Diff. | Change | Custom. | Integ. | Data/AI | Risk | Score |
|---|---|---|---|---|---|---|---|
| Quoting and product configuration | 5 | 5 | 4 | 3 | 5 | 4 | 36: build |
| Production scheduling | 4 | 4 | 3 | 3 | 4 | 4 | 30: build |
| Inventory and procurement | 3 | 3 | 3 | 2 | 3 | 3 | 23: hybrid |
| General ledger and close | 1 | 2 | 2 | 2 | 2 | 1 | 13: keep |
| Payroll | 1 | 2 | 1 | 2 | 1 | 1 | 10: keep |
Illustrative scores for a hypothetical electrical equipment manufacturer. Example: quoting = (2 × 5) + 5 + 4 + 3 + (2 × 5) + 4 = 36. Your scores will differ; the method matters more than the numbers.
Illustrative scores, using the weighting described in this guide.
View the data
| Workflow | Score | Band |
|---|---|---|
| Quoting and product configuration | 36 | Build |
| Production scheduling | 30 | Build |
| Inventory and procurement | 23 | Hybrid |
| General ledger and close | 13 | Keep |
| Payroll | 10 | Keep |
Turn scores into a plan
- Start with the highest scorer that has a clear business owner. A strong score with no owner stalls.
- Build it alongside the packaged system, which stays the system of record for finance.
- Retire what it replaces: the bolt-ons, the custom code, and the licenses your contract lets you drop.
- Measure against the number you picked in factor 5, then move to the next workflow. Never big-bang.
Then compare five-year costs for each build candidate: one-time build plus five years of hosting and support, against five years of license or subscription, consultants, and upgrade work tied to that workflow. We build owned software for a living, so weigh our view accordingly. In our worked example for a $200M manufacturer on SAP, rebuilding every operational workflow, moving finance and payroll to cloud products, and switching SAP off comes to about $5.8M over five years, including our fees, against $7.5M–$11M to migrate. That's an illustration, not a quote.
Questions to ask before you build anything
- Which workflows would customers notice if we did them better?
- Where have we spent the most on custom code, and is it still used?
- Which processes change every quarter, and what does each change cost us today?
- What would we do with clean data from this workflow that we can't do now?
- Who will own, host, and support it in year three?
- What does our ERP contract let us retire as we replace?
Where to go from here
Score five workflows this week and see where they land. If you'd like help checking the scores against your contract, tell us what you run and what you pay. To see how a staged replacement works across SAP, Oracle, and Dynamics 365, visit our Escape SaaS page.
Sources
- CIO Wiki: Gartner's Pace-Layered Application Strategy
- arXiv (Peng et al.): The Impact of AI on Developer Productivity: Evidence from GitHub Copilot
- METR: Measuring the impact of early-2025 AI on experienced open-source developer productivity
- The Register: More license upheaval to come after SAP kills RISE with SAP products
- TechTarget: Explaining the hidden costs of ERP implementations
- VendorBenchmark: SAP S/4HANA migration cost benchmark