Direct answer: repair when the platform still fits, ownership is clear, data is sound, and defects can be bounded. Rebuild when the inherited implementation cannot be evolved safely. Replatform only when the platform itself blocks the required operating model or total cost.
Do not begin with sunk cost
Money already spent does not prove that a system should be saved, and a failed delivery does not prove the platform is wrong. Separate four questions: does the platform fit the business, can the buyer prove ownership, can data and transactions be trusted, and can the architecture be changed without creating more risk?
Four decision paths
| Path | Evidence in favor | Evidence against | First acceptance gate |
|---|---|---|---|
| Stabilize and repair | Magento still fits; core data is reliable; failures are bounded; rollback is possible. | Unknown ownership, systemic data corruption, or recurring failures without a testable cause. | Checkout, orders, integration, and deployment meet a defined stable baseline. |
| Refactor in place | Business workflows are valuable, but custom modules, releases, or integrations create recurring risk. | Refactoring would recreate most of the platform while the business keeps paying for two systems. | One high-risk area is replaced behind tests and observable acceptance criteria. |
| Rebuild on Magento | The platform fits, but the inherited codebase, theme, or deployment model cannot be evolved safely. | Platform licensing, operating model, or capability limits are the real problem. | Destination architecture and migration scope pass an independent challenge review. |
| Replatform | A different platform better fits the required capabilities, skills, cost, and operating model. | The proposed destination has not been validated against data, SEO, integrations, and peak load. | A buyer-owned decision record proves why migration risk is lower than recovery risk. |
Ownership can decide before architecture
A technically repairable system may still be unsafe to rescue if the buyer cannot prove ownership of repositories, licenses, cloud accounts, domains, or critical custom code. Resolve contracts and access before asking a new agency to recreate credentials or copy assets. Use the takeover checklist.
Data and order integrity come before frontend quality
Prioritize payment state, order IDs, totals, stock, customer records, and ERP or PIM synchronization. A faster storefront is not a successful rescue if orders disappear, replay twice, or reach ERP with incorrect values. Require a documented reconciliation method and accepted exception list.
Test the platform-fit hypothesis
- List the business capabilities that are genuinely blocked by Magento or Adobe Commerce rather than by the implementation.
- Compare license, hosting, extension, integration, release, support, and specialist costs over the same time period.
- Model catalog, customer, order, content, URL, consent, search, promotion, and integration migration.
- Require the alternative platform to pass the hardest B2B, ERP, payment, and multi-market scenarios.
- Record who benefits commercially from the recommendation and keep final acceptance with the buyer.
Avoid the false 70% rule
No active evidence rule on this site says that rescue is correct below a fixed percentage of rebuild cost. Cost estimates are uncertain, and risk is not proportional to code volume. Use ranges, dependencies, confidence, and explicit stop conditions instead.
Minimum decision output
The assessment should end with an evidence pack, option table, dependencies, estimates, risk register, data plan, integration plan, SEO plan where relevant, acceptance tests, and one recommended path with a documented dissent or alternative. The buyer should be able to give that package to another provider.
A good rescue adviser can recommend repair, rebuild, or replatform. A delivery provider may still benefit from the chosen path, so compare the decision logic before comparing implementation proposals.
Continue the rescue review
Return to the eight rescue agencies, use the first 72 hours plan, or review the RESTORE-100 evidence method.