SPARC is our structured approach to AI requirements specification. By defining Situation, Problem, Action, Result, and Constraints upfront, we eliminate the ambiguity that causes most AI projects to fail.
Most AI projects fail because requirements are vague. SPARC forces precision at the start, so everyone - engineers, stakeholders, and governance teams - has the same understanding.
What is the current state? Describe the business context, the people involved, the systems that exist, and the conditions that make this problem relevant. A good Situation statement makes the stakes clear without prescribing a solution.
What specifically is not working? The Problem statement must be precise, measurable, and scoped. Avoid vague language like "inefficiency" - instead specify: "Manual invoice processing takes 4 hours per batch and produces 8% error rate."
What should the AI system do? Describe the behaviour, not the implementation. "The system should extract line items, match against PO database, flag discrepancies above £500, and route for approval within 30 seconds."
What does success look like? Define measurable outcomes: "Processing time reduced to under 2 minutes per batch. Error rate below 1%. 99.5% uptime. ROI positive within 6 months." These become the evaluation criteria for the AI system.
What must the AI NOT do, or what limits apply? Data privacy rules, regulatory requirements, latency budgets, cost ceilings, explainability requirements, and integration limitations. Constraints prevent scope creep and governance failures.
We run SPARC specification workshops that produce a complete, signed-off requirements document in one day.
Book a SPARC Workshop →Monthly insights on agentic AI, ADLC methodology, and enterprise AI governance. No spam — unsubscribe any time.
The data is clear: organisations that invest in rigorous AI specification before building achieve dramatically better outcomes. Here is the evidence.
| McKinsey Dimension | Description | SPARC Phase |
|---|---|---|
| Strategy | AI aligned to business objectives with clear ROI targets | Specification |
| Talent | Right skills, roles and incentives for AI delivery | Specification |
| Operating Model | Agile delivery org with cross-functional AI teams | Pseudocode |
| Technology | Modern AI stack, MLOps, infrastructure at scale | Architecture |
| Data | High-quality governed data pipelines and lineage | Refinement |
| Adoption & Scaling | Change management, KPIs, continuous improvement | Completion |