Governance Mandate, Inventory and Decision Rights · Lesson 1
Define the governance mandate and risk appetite
10 min read
Lesson objectives
Analyze turning organizational principles into testable decision rights and operating boundaries and justify an evidence-proportionate decision.
Reading tools
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
Imagine an executive committee approves an AI assistant because it wants faster service. Operations later asks who may stop the assistant if harm rises. Security and customer service disagree; who resolves the disagreement? At what point does a change become a breach rather than ordinary variation? If the approval record cannot answer those questions, the organization does not merely lack another policy. It lacks an operable mandate. A governance mandate grants named people or bodies authority to make named decisions about an identified system and states what they may not exceed. Risk appetite describes the type, amount, and conditions of exposure the organization is willing to accept in pursuit of a declared benefit. Appetite does not mean risk is desirable, and it does not mean every harm can be traded for speed or revenue. Some conditions are eligibility gates: no legitimate basis for using the data, no challenge route for a consequential decision, or no person able to suspend service. A failed gate prevents the option from entering the trade-off at all. Draft the mandate around a decision, not a technology label. For example: may a reply-drafting feature be piloted for low-impact information requests for six weeks? That sentence fixes the function, population, and duration. Name four distinct authorities next. The service owner proposes operating scope. The risk owner accepts residual exposure within a stated boundary. An oversight function challenges the evidence. A pause authority suspends use when an agreed signal appears. One person may hold several roles in a small organization, but the record must expose the conflict and explain its treatment. Reserve material decisions for a higher body: expansion into a high-impact use, a new purpose, sensitive data, or a higher risk limit should not become a product-owner choice by default. State the authorization period, system version, controlling record, deputy when an authority is unavailable, and the rule for urgent action outside normal hours. NIST AI RMF is useful because governance reaches across context, measurement, and management rather than ending after an opening approval. The voluntary framework is not thereby converted into law and cannot replace applicable organizational or sector duties. Stress-test the mandate with a disagreement. An executive requests expansion, errors are rising, and two teams use different denominators. Can each participant identify the action available now, the matter that requires escalation, and the version of the limit that governs? If not, the mandate is descriptive rather than operational.
Translate appetite into inspectable boundaries through three layers. Layer one is a qualitative position connecting benefit and harm: the organization accepts a bounded trial to reduce response time, but does not accept an automated final decision that denies service or removes a challenge channel. Layer two is an operational measure with a numerator, denominator, and observation window. Replace “high accuracy” with “replies containing a severe error per 1,000 replies reviewed over fourteen days.” Layer three is a pre-agreed action rule: when the measure reaches its threshold, who stops which activity, how quickly, and what must be true before resumption? Selecting measures requires a model of the harm pathway. An average can improve while a rare category remains exposed to severe consequences, so one dashboard number cannot govern every case. Combine an outcome signal, such as misleading replies that reached customers, with an early warning, such as the share of cases not reviewed on time, and a response-capacity signal, such as time to detect deviation. Do not immediately collapse them into a composite score. A weighted total can conceal failure of an essential condition behind improvement in a less important indicator. Build a small four-zone matrix. Green permits operation within approved scope. Amber requires quicker review or freezes expansion. Red pauses the affected pathway and calls the risk owner. Black represents a non-waivable blocker, such as loss of the decision log or inability to identify affected cases. For each zone, specify the boundary, owner, response deadline, and data source. Test measurement quality as carefully as the threshold. Does the log cover every case or a sample? Do reviewers share a definition of severe error? Did the version or sampling method change? Can the team distinguish no event from no observation? Allow for uncertainty. If a small category has ten observations, zero recorded errors is not proof of safety; the defensible choice may be constrained use until a prespecified sample is complete. Treat an exception as its own decision containing a reason, duration, scope, compensating safeguard, approver, and exit. An open-ended exception destroys appetite because it turns a boundary into a negotiation after every breach. Examine incentives too. A team rewarded only for launch speed may delay reporting or reclassify error. Separate acceptance authority from delivery targets and provide an escalation route outside the product line. Finally, read the rule as an operator on duty. Can that person calculate it from available data? Is the first action clear without convening a meeting? Is a deputy available? Those tests distinguish polished language from a decision system that functions under pressure.
Lesson summary
The lesson connects turning organizational principles into testable decision rights and operating boundaries to a measure, owner and review condition, recognizing that the teaching threshold is not a legal or market standard and must be calibrated to harm and context.