The Migration Readiness Score (MRS) is the primary output of every NextAI Foundry assessment. It is a number between 0 and 100 that represents how ready a given application is to absorb agentic AI capabilities in its current state. This post explains what the score measures, how it is calculated, and how to read the tier thresholds correctly.
Why a composite score rather than a checklist
Early prototypes of the assessment produced a checklist: 47 yes/no questions mapped to a pass/fail outcome. Enterprise architects hated it. A single no on a critical dimension could block an otherwise strong application, while an application could game the checklist by satisfying every item without actually being deployable.
A weighted composite score reflects the reality that readiness is a gradient, that organisations can sequence remediation work across dimensions, and that strength in one area can partially offset weakness in another — within limits. The MRS was designed to be directional and actionable, not to be a binary gate.
The five dimensions
Each MRS tier maps directly to a level in the Agentic AI Maturity Model — understanding the five levels helps contextualise what a score of 70 versus 85 actually unlocks in practice.
The Agentic AI Maturity Model: Five Levels from Automation to Autonomous Enterprise →- Architecture (25%) — Service boundaries, API surface, modularity, and the degree to which business logic is separated from presentation and infrastructure concerns.
- Data (25%) — Data ownership clarity, schema quality, access latency, and the ability of the application to serve as a reliable data source for an AI agent operating at runtime.
- Integration (20%) — Existing API contracts, event streams, and the maturity of the application's integration patterns. Applications with published, versioned APIs score higher.
- Team (15%) — Deployment frequency, test coverage, observability maturity, and the team's prior exposure to AI or automation projects.
- Process (15%) — Business workflow documentation, human escalation paths, exception handling maturity, and the degree to which the application's decision logic is explicit rather than embedded in institutional knowledge.
Architecture and Data carry the highest weights because they are the hardest to change quickly and have the most direct impact on whether an agent can function at all. Team and Process carry lower weights because, while they matter, they can be developed in parallel with technical remediation on shorter timescales.
How Claude Sonnet scores each dimension
The 25-question assessment intake produces a rich text description of each application. Claude Sonnet analyses these answers against a structured rubric for each dimension, producing a score from 0 to 100 with a confidence level and a structured rationale. The rubric was developed from patterns observed across application portfolios and is versioned — we update it when expert review surfaces systematic calibration errors.
The AI does not simply keyword-match. It reasons about combinations. An application that describes a microservices architecture but also describes a single shared transactional database and no API versioning will score lower on Architecture than the microservices label alone would suggest. This is why the intake questions probe beyond technology choices into operational practice.
The four tiers
- Not Ready (0–39) — Fundamental blockers exist. Agentic augmentation requires significant preparatory investment before any agent deployment is advisable.
- Emerging (40–69) — Partial readiness. Narrow, well-scoped agent use cases are viable with mitigations. Full agentic capability requires targeted remediation across one or more dimensions.
- Ready (70–84) — Agentic augmentation is viable with standard risk management. Most agent patterns can be applied directly. Minor remediation may accelerate outcomes.
- Accelerate (85–100) — Strong readiness across all dimensions. The application is a strong candidate for advanced agentic patterns including multi-agent coordination and autonomous decision-making.
A common misreading: the tier thresholds apply to the composite MRS, not to individual dimension scores. An application can score 90 overall while scoring 60 on Process — it would still be classified as Accelerate, but the Process score would appear as a risk flag in the detailed report.
Limitations of the score
The MRS is based entirely on information provided in the intake form. It reflects the assessor's accurate description of the application. If the intake answers are aspirational rather than factual — describing the roadmap state rather than current state — the score will overstate readiness. We strongly recommend completing the assessment based on what exists today, using a separate section of the report to capture planned improvements.
The score also does not account for regulatory constraints, budget availability, or vendor lock-in — factors that may prevent technical remediation regardless of the score. These should be applied as a filter after reviewing the MRS output, not embedded in the score itself.
Before completing the intake form, review the five structural signals that predict whether an application can absorb agentic capabilities — they map directly to the Architecture and Integration dimensions of the MRS.
Five Signs Your Legacy Application Is Ready for Agentic AI →