Pattern: Use-Case Intake & Risk Tiering

Layer 01 · Govern-as-Code Layer 02 · Inventory & Transparency In the chapter 05 catalogue

One working control on the engineering side of AI governance.

Summary: Route every proposed AI use case, built or bought, through one intake that writes a structured use-case record, screens it against prohibited practices and the EU AI Act risk ladder, and computes an internal risk tier from declared profile fields. The tier, not a meeting, decides which assessments, evals and approvals the system must clear before it ships, and the record becomes the registry entry every later gate reads.

Objectives

Make the first decision about an AI system a recorded, reproducible one: what it is for, what it must not be used for, how risky it is and, from that, how much governance it gets. Spend review effort where the risk is and let low-risk use cases through fast on a paved path.

Target users

AI governance engineer, product owner, platform team, legal and privacy reviewers.

Impacted stakeholders

People affected by the system’s outputs, deployers and operators, the AI governance committee, auditors, market surveillance authorities.

Relevant principles

Build the control at the earliest point it can block; make the governed path the easiest path; register and bound every actor before it acts.

Context

An organisation where many teams propose AI features, and most of them buy or call models rather than train them. Requests arrive by email, in slide decks and in procurement tickets, and each is reviewed with whatever questions the reviewer remembers. The EU AI Act measures most duties against the intended purpose, which it defines to include the context and conditions of use stated in the instructions for use, “promotional or sales materials and statements” and the technical documentation (Art. 3(12))¹. The NIST AI RMF asks that intended purposes and settings are “understood and documented” (MAP 1.1) and that risk tolerances are “determined and documented” (MAP 1.5)².

Problem

Without one intake, governance starts too late and scales badly.

• Forces. Reviewers want every use case assessed in depth; teams want an answer in days. Classification depends on facts only the team knows: the purpose, the affected persons, whether the system profiles people. A tier negotiated in a meeting drifts with whoever attends. A tier that is not machine-readable cannot switch a gate on.
• Failure mode. A high-risk use slips through as “just a pilot” while low-risk requests queue behind it. Nobody can show which systems were classified, by whom, on which facts, or why an Annex III system was treated as not high-risk.

Solution

Build intake as a form-plus-code path that ends in a record and a tier, not in minutes.

1. Capture the use-case record. A short structured form (intended purpose, out-of-scope uses, users and affected persons, decision authority, success metrics, error appetite, data sources, jurisdictions) writes a registry stub keyed to one id. Reuse the published use-case record schema (use-case-record.v1) so the form, the registry and the gates share one shape.
2. Screen before you score. Run the prohibited-practice screen (Art. 5) first: a hit is blocked at intake and never tiered. Then place the system on the Act’s ladder: an Annex III use, with the Art. 6(3) filter and its override (an Annex III system that profiles natural persons is always high-risk), transparency duties, or a general-purpose model. A provider that relies on the filter must document its assessment before placing the system on the market and register it (Art. 6(4), Art. 49(2))¹; the intake record is that documentation. The Digital Omnibus moved the Annex III high-risk obligations to 2 Dec 2027³: that changes when the duties bite, not whether the classification is recorded now.
3. Compute the internal tier. Declare profile fields (autonomy, decision impact, exposure, reversibility of the worst outcome, vulnerable groups, data class, third-party dependency) and let a versioned policy compute the tier. The AI RMF sets the level of risk-management activity by risk tolerance (GOVERN 1.3) and asks for the likelihood and magnitude of each identified impact (MAP 5.1)². Canada’s Directive on Automated Decision-Making applies the same idea in public administration, with four impact levels defined partly by reversibility and duration⁴. A team that disagrees with its tier changes a factor, with evidence, in a reviewed change; the tier follows.
4. Bind the tier to gates. The tier selects the required assessments (DPIA, FRIA, vendor due diligence), the eval categories and thresholds, the approvers and the review cadence, so the pipeline reads what it must enforce. Deployment refuses any system without an intake record (“no stub, no deploy”), which keeps the inventory complete by construction (GOVERN 1.6)².
5. Re-open on change. A new purpose, population, jurisdiction or data source, or a consumer that declares a use on the out-of-scope list, re-runs intake and may move the tier.

Illustrative use-case record at intake, valid against use-case-record.v1 (the profile fields and the filter claim travel in extensions):

{
  "$schema": "https://aigovernanceengineer.com/schemas/use-case-record.v1.json",
  "record_id": "uc-2026-042",
  "title": "Payslip field extraction for mortgage applications",
  "business_owner": "head-of-mortgage-operations",
  "intended_purpose": "Extract income fields from uploaded payslips into the application form for an underwriter to confirm; the affordability assessment is made elsewhere.",
  "out_of_scope_uses": ["affordability scoring", "automatic decline", "employment verification"],
  "users": ["mortgage underwriters"],
  "affected_persons": ["mortgage applicants"],
  "decision_authority": "human_decides",
  "ai_justification": {
    "alternatives_considered": ["manual keying", "template-based OCR"],
    "why_ai": "Payslip layouts vary too much for templates; every extracted field is confirmed by an underwriter."
  },
  "success_metrics": [
    { "metric": "field-level exact match on a frozen sample", "target": ">= 0.98", "direction": "higher_is_better" }
  ],
  "error_appetite": "A wrong income figure can distort an affordability decision; low-confidence fields are routed to manual keying.",
  "data_sources": [{ "name": "applicant payslips", "personal_data": true, "special_category": false }],
  "jurisdictions": ["ES", "PT"],
  "preliminary_classification": {
    "eu_ai_act_category": "minimal",
    "internal_tier": "medium",
    "rationale": "Preparatory task to an Annex III 5(b) assessment (Art. 6(3) filter claimed, no profiling); assessment documented and registered under Art. 6(4) and Art. 49(2)."
  },
  "assessments_required": ["dpia"],
  "decision": {
    "outcome": "approved_with_conditions",
    "conditions": ["Art. 49(2) registration before go-live", "monthly 2% sample checked against source payslips"],
    "decided_by": "ai-governance-review",
    "decided_at": "2026-09-22"
  },
  "register_entry": "mortgage-extract-01",
  "extensions": {
    "risk_profile": { "autonomy": "suggests", "decision_impact": "informs", "exposure": "customers",
                      "reversibility": "reversible", "vulnerable_groups": [], "data_class": "personal",
                      "third_party": ["ocr-vendor-02"] },
    "tier_rule": "tiering-policy.v3",
    "annex_iii_point": "5(b)",
    "art_6_3_condition": "preparatory_task",
    "profiling": false
  }
}

Consequences

Every system has a purpose, a class and a tier on record before it costs compute; review effort follows risk; the inventory is complete because deployment depends on it; and each classification is auditable down to the facts it rested on. The costs: the form must stay short or teams route around it; the tier rule needs calibration and an appeal path; and self-declared facts can be wrong, so intake needs spot checks against discovery and procurement.

Related patterns

Agent Registry; Policy Card; FRIA-as-Code; Vendor / Model Due-Diligence Gate; Shadow-AI Discovery; AI Threat Model; Dataset Admission Gate.

Maps to: EU AI Act Art. 3(12), Art. 5, Art. 6(3)–(4), Art. 49(2), Annex III · ISO/IEC 42001 A.5.2, A.9.4 · NIST AI RMF (Govern 1.3, 1.6; Map 1.1, 1.5, 5.1) · Layer 01 Govern-as-Code / Layer 02 Inventory & Transparency.

Function and subcategory labels follow the NIST AI RMF²; ISO/IEC 42001 Annex A ids follow a published crosswalk, not the standard’s text⁵. Mappings are illustrative, not a claim of conformity.

Sources

1. [1] Regulation (EU) 2024/1689 (AI Act): Art. 3(12) intended purpose (incl. “promotional or sales materials and statements”); Art. 5 prohibited practices; Art. 6(3) filter and profiling override, Art. 6(4) documented assessment before placing on the market; Art. 49(2) registration of systems concluded not high-risk under Art. 6(3); Annex III (text read on the Commission’s AI Act Service Desk, 2026-09-24). Publications Office of the EU (EUR-Lex). 2024-07-12. https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng (verified: primary)
2. [2] Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (GOVERN 1.3 level of risk-management activity by risk tolerance; GOVERN 1.6 inventory of AI systems; MAP 1.1 intended purposes and settings “understood and documented”; MAP 1.5 risk tolerances “determined and documented”; MAP 5.1 likelihood and magnitude of each identified impact). NIST. 2023-01-26. https://doi.org/10.6028/NIST.AI.100-1 (verified: primary)
3. [3] Regulation (EU) 2026/1744 (Digital Omnibus on AI), of 8 July 2026, amending Reg. (EU) 2024/1689; OJ L, 24 July 2026; in force 27 Jul 2026; Annex III high-risk obligations moved to 2 Dec 2027. Publications Office of the EU (EUR-Lex). 2026-07-24. https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng (verified: primary)
4. [4] Directive on Automated Decision-Making (algorithmic impact assessment before production; Appendix B impact levels I to IV defined by reversibility and duration). Treasury Board of Canada Secretariat. 2025-06-24. https://www.tbs-sct.canada.ca/pol/doc-eng.aspx?id=32592 (verified: primary)
5. [5] NIST AI RMF to ISO/IEC FDIS 42001 crosswalk (provider: Microsoft; lists the Annex B implementation-guidance clauses, whose numbers mirror the Annex A control ids, e.g. B.5.2 AI system impact assessment process, B.9.4 intended use of the AI system; the ISO text was not opened). NIST AI Resource Center. 2023. https://airc.nist.gov/docs/NIST_AI_RMF_to_ISO_IEC_42001_Crosswalk.pdf (verified: secondary)