Where rescue begins
The first task is not to “fix the model.” It is to reconstruct the intended business outcome, current process, approved architecture, data rights, configuration, acceptance criteria, actual results, incident history, cost, and responsibilities.
Common root-cause categories
Process
Unstable workflow, missing ownership, hidden exceptions, or downstream bottlenecks.
Data and knowledge
Unavailable, unlawful, poor-quality, stale, contradictory, or ungoverned sources.
System design
Wrong model, weak retrieval, brittle integrations, excessive privilege, poor observability, or no fallback.
Evaluation
Demo criteria, contaminated tests, no baseline, missing critical cases, or unsupported vendor claims.
Adoption and operations
Review burden, unclear procedures, inadequate staffing, no incident ownership, or uncontrolled change.
Commercial model
Consumption surprises, custom-work dependency, lock-in, vague acceptance, or underestimated client effort.
Recovery decision
The recommendation separates sunk cost from future value. It identifies what must stop immediately, what evidence can be generated quickly, which assets are reusable, what alternatives exist, and the conditions under which further spending is justified.
Stop conditions
- No accountable executive or process owner.
- Data cannot be lawfully or contractually used.
- The system cannot be observed, stopped, or rolled back.
- Critical errors or security failures remain unresolved.
- The economic case fails after realistic review and operating costs.
- The client requires certification of a predetermined continuation decision.
