Build it by hand, build it with AI, add AI to your FMEA tool, hire consultants, do nothing, or deploy a system that compounds with every correction. Here’s how each option plays out.
Six paths. Five cost you time. One compounds in your favor.
 
|
By hand (your engineers) | Internal AI build | AI inside your FMEA tool | Consultants | Not done | |
|---|---|---|---|---|---|---|
| FMEA for 1 system | Typical: engineer review of AI-generated baseline | Typical: 100-300 engineer hours spread across months | Fast first draft once the pipeline exists | Faster entry: the assistant suggests rows, engineers still build the FMEA | Typical: dedicated project over several weeks plus workshops | Skipped |
| Calendar time | Typical: 1-3 weeks for scoped systems | Often 3-6 months part-time | Months to build the pipeline, then fast drafts | Weeks, mostly engineer time | Often 6-10 weeks | Never |
| How it stays current | ✓ Updates continuously as failure reports, events and source documents change | ✕ Usually updated manually, infrequently | ✕ Only if someone keeps maintaining the pipeline | ✕ Only when an engineer opens the FMEA and asks | ✕ Usually static unless re-engaged | Not maintained |
| After delivery | ✓ Searchable, reusable working model | Static spreadsheet/document | Depends on the builder staying | Static document in the tool | Static deliverable | Reactive firefighting |
| Your current FMEA tool | ✓ Stays the record: import from its exports, approved changes export back | Stays | Stays | Stays | Stays | Stays |
| Resource impact | Engineers review and approve | Engineers build it themselves | Engineers build and maintain the tool, plus review | Engineers drive every step, prompt by prompt | Engineers still support workshops/review | Hidden reactive cost |
| Methodology | Structured around IEC 60812 and common FMEA frameworks such as AIAG-VDA and SAE J1739 | Varies by team/site | Whatever the builder encodes. Rarely AIAG-VDA scoring tables | ✓ The tool’s AIAG-VDA structure | Usually strong methodology and facilitation, but output is project-based and update cycles are expensive | None |
| Personalization | ✓ Built from your data, your systems, your history | Best context. They know the equipment. Quality varies and it’s rarely written down well. | ✓ Your data, if it’s connected | What is already in the tool. 8Ds and field data stay outside | Fresh perspective, but less equipment-specific depth | N/A |
| Traceability | ✓ Every row linked to source paragraph. Change flagging on revision. | Spreadsheet-based. Weak audit trail. | Usually none at row level | Row history in the tool, no link to the 8D or failure report that should change it | Better documented, but not linked to source data | None |
| Knowledge retention | ✓ Stored, queryable, reusable | ✕ Mostly tribal knowledge | ✕ Prompts and scripts, not a governed record | Stored in the tool. The assistant does not learn from corrections | ✕ Locked in deliverables | Lost over time |
| Survives AI model changes | ✓ Knowledge stored as domain data, not model weights. Models upgrade, your corrections transfer forward. | Tied to whatever tool or model version was used | Rebuild prompts when the model changes | The vendor’s roadmap | Static deliverable. No AI layer to upgrade. | N/A |
| Scalability | Expandable across systems/sites. Supports multilingual teams. | Limited by bandwidth | Limited by the internal team’s roadmap | One FMEA at a time | Limited by budget | No scale |
| Commercial model | Predictable software/project spend | Internal opportunity cost | Hidden engineering cost, ongoing | Seat licences | Recurring engagement cost | Cost shows up as failures, downtime, and audit pain |
General-purpose AI drafts a good FMEA from a manual. Here is what it does not keep.
See comparison against ChatGPT.
Send us a sample document. In 30 minutes we run it through the pipeline and show you what we find versus a manual process. No commitment.