Explainable decision intelligence
Decision transparency
How AIM makes recommendations reproducible, explainable, auditable, and reviewable by the people accountable for the outcome.

AIM is designed to withstand skeptical oversight. Every recommendation can be explained, every score can be traced, and every decision can be reproduced. This page documents how AIM achieves hostile auditor defensibility.
AIM operates upstream of execution. That means the audit-ready record is built at the point the modernization decision is being made — not retrofitted after a platform has already been chosen. The traceability, defensibility, and reproducibility described below are properties of the decision itself, not of the runtime that executes it.
1. How AIM Ranks Options
AIM uses RAO (Ranking and Optimization) to score and rank modernization candidates. Scores are computed deterministically using:
- Security – Lifecycle status, exposure, data sensitivity
- Compliance – Regulatory fit (HIPAA, CJIS, PCI, etc.)
- Cost – Total cost of ownership and budget alignment
- Complexity – How easy the technology is to operate. A higher score needs less specialized effort.
- Lock-in – How easy it is to leave the vendor. A higher score is more portable.
- Maturity – Market adoption and stability
📖 Detailed Methodology: See Scoring Methodology for a complete explanation of each dimension, what high/low scores mean, and how scores are curated for audit defensibility.
The recommendation list assigns one of four tiers from the RAO score:
- Must Have Score ≥ 80
- Strong Candidate Score 65–79
- Consider Score 50–64
- Avoid Score < 50
The explainability scorecard uses a shorter label for the same score: Recommended at 80 and above, Acceptable from 65 to 79, and Avoid below 65.
Learn more: Constraints & RAO Scoring →
2. Where AIM sits in the stack
The decision layer, not the work surface
AIM is the decision and sustainment record. ServiceNow and Jira remain the work surface and configuration database. PSM can read one ServiceNow business application, service, or application and, when ticket sync is connected, queue a change request to Jira or ServiceNow. Named humans keep those systems as the operational source.
Named humans stay accountable
AIM provides decision support. All outputs are designed to be reviewed, edited, and approved by human architects and stakeholders. Named humans remain accountable for all implementation decisions.
Expertise stays with the organization
AIM accelerates planning by structuring data and surfacing options. Subject matter expertise, procurement processes, and vendor negotiations stay with the organization.
3. Vendor Neutrality Statement
Platform Independence
- No paid placements – AIM does not accept vendor sponsorships, referral fees, or kickbacks
- Consistent methodology – All candidates are scored using the same RAO dimensions
- Open competition – Any qualified vendor can bid on work derived from AIM assessments
- Catalog-backed – Product recommendations come from a curated, version-controlled catalog
When AIM names specific products, these represent a solution baseline for planning and vendor comparison — not vendor preference.
Learn more: How AIM Works: Platform Independence →
4. Reproducibility & Audit Logs
Every recommendation and report generation is logged with:
| Field | Description |
|---|---|
| input_hash | SHA-384 hash of normalized assessment inputs |
| output_hash | SHA-384 hash of ranked outputs and scores |
| input_hash_algorithm / output_hash_algorithm | Algorithm that produced each digest. Current seals record SHA-384. |
| methodology_version | Version of scoring methodology used |
| rao_version | Version of RAO algorithm used |
| created_by | User who triggered the generation |
| created_at | Timestamp of generation |
Quantum-resistant cryptography and hash stability
Input hash: Stable as long as the assessment configuration is unchanged. An auditor can confirm exactly what inputs were used at the time a recommendation was generated.
Output hash: Captures the ranked recommendations, RAO scores, and tier assignments at the moment of generation. Re-running the same assessment at a later date will produce a different output hash if pricing data, labor rates, catalog entries, or compliance flags have been updated since — this is expected and correct behavior. The stored output hash is the auditable record of what was recommended and why at a specific point in time.
Algorithm: AIM-signed operational artifacts across the platform use hybrid NIST FIPS 204 signatures; that posture is published on Compliance and assurance, not as a scoring-methodology claim. Decision seals remain SHA-384. See Decision Provenance for the hash algorithm note.
5. Data Freshness & Updates
AIM uses multiple data sources with defined update cadences:
| Source | Update Cadence | Notes |
|---|---|---|
| Cloud Provider APIs | Monthly | AWS, Azure, GCP pricing (refreshed 1st of month) |
| Software/SaaS Pricing | Monthly | Vendor public pricing pages |
| Hardware MSRP | Monthly | Dell, HP, Cisco, Palo Alto, etc. |
| BLS Labor Rates | Annual | Bureau of Labor Statistics OES data |
| Technology Catalog | As needed | Product lifecycle, scoring, metadata |
Learn more: Cost Estimate Confidence Methodology →
6. Uncertainty & Missing Prices
AIM explicitly flags uncertainty rather than hiding it:
- “Requires Quote” – Products with enterprise custom pricing or population-based licensing
- Confidence Levels – HIGH/MED/LOW labels. Line items follow the price source. The overall label is the more cautious of a dollar-weighted unit price and a separate quantity rating. These labels are not a measured error percentage.
- Stale Data Flags – Warnings when data sources exceed expected update cadence
- Missing Input Warnings – Explicit flags when assessment inputs are incomplete
Honest Uncertainty
AIM prefers to say “This information was not provided” rather than invent data. All estimates are planning-level and should be validated with vendor quotes before procurement.
7. AI Role Statement
How AI Is Used
- Narrative assistance – AI improves readability of report text
- Context normalization – AI helps structure free-text inputs into normalized constraints
- Pattern detection – AI assists in identifying architecture patterns from system descriptions
- Conversational knowledge access (AIM Eagle) – AI powers the AIM Eagle chat assistant, which answers questions about the platform grounded in verified knowledge content. Eagle operates in a fully separate pipeline from the assessment and scoring engines — it has no ability to read, influence, or modify RAO scores, recommendations, or generated reports. Eagle responses are informational only and are not part of the auditable decision record.
Deterministic Scoring
AI does not change scores. All RAO scores, tier assignments, and rankings are computed using deterministic, rule-based logic that can be reproduced and audited.
AI narrative assistance is clearly disclosed in reports. The underlying scoring methodology is fully transparent and documented.
8. For Auditors
If you are auditing an AIM-generated report or recommendation:
Check the Explainability Appendix
Reports include an appendix with methodology version, data freshness, scorecards, and reproducibility hashes.
Verify Hashes
Each decision run stores an input hash (stable for unchanged assessments) and an output hash (captures recommendations at that moment). Compare the stored output hash against the report generated at the same time. Output hashes will differ if re-run later as pricing, labor rates, or catalog data has since been updated — this is expected, not an integrity failure.
Review Decision Runs
Decision runs are immutable audit records that cannot be modified after creation.
Request Data Exports
Structured data exports are available upon request via privileged access.
Learn More About AIM
Explore how AIM works and understand the methodology behind recommendations.