Most failed Salesforce implementations can be rescued without starting over. The first step is a short assessment that separates what works from what does not; the second is stabilizing whatever the business depends on today; the third is a keep, fix or rebuild decision for each part of the build. A full restart is only justified when the data model itself does not fit how the business works.
Signs a project needs rescuing
- The go-live date has slipped more than once, with no firm new date.
- The system is live but teams have gone back to spreadsheets or the old tool.
- The previous partner or admin has moved on and nobody understands the build.
- You are paying for licenses or features nobody can use.
- Every change breaks something else.
The first two weeks
- Get admin access and a copy of the original scope, designs and any documentation.
- Interview the people who were supposed to use the system: what they need, what they avoid, what breaks.
- Inventory the org: objects, automation, custom code, integrations, managed packages and licenses.
- List what the business depends on right now, and stabilize that first.
- Compare what was promised with what was built, and note the gaps.
Keep, fix or rebuild
| Decision | When it applies | Example |
|---|---|---|
| Keep | It works, users rely on it, and it is maintainable | Standard objects and page layouts that match the process |
| Fix | The idea is right but the execution is not | Automation that fires in the wrong order, or a sync that fails silently |
| Rebuild | The design cannot support the process at any reasonable cost | A data model that forces workarounds for every record |
| Retire | Nobody needs it | Unused fields, reports, packages or licenses |
What rescues look like in practice
An industrial-services firm had paid a previous partner more than $100,000 for a Salesforce Field Service build it could not use. Within a 20-hour engagement, the scheduling errors were fixed and a custom Labor Resource object removed the need for Field Service licenses for 75–100 field workers. A medical-device company regained a broken Pardot instance by removing 970+ invalid records and restoring its automation. A political-advocacy agency moved a failed two-year Tableau investment to a working roadmap and active progress.
