Decision delegation criteria

You are bottlenecking yourself on decisions that do not need you

Every decision you make yourself is a decision you are preventing someone — or something — else from making. That sounds like control. It feels like diligence. It is actually a failure of design.

The average knowledge worker makes an estimated 35,000 decisions per day, most of them trivial, many of them habitual, and a small number of them genuinely consequential. The problem is not the volume. The problem is that most people apply the same level of personal involvement to all of them. They review the expense report and the market entry strategy with the same sense of ownership, and then wonder why they are exhausted by Thursday and their strategic thinking has degraded to pattern-matching.

This lesson gives you explicit criteria for sorting decisions into three categories: those you must make yourself, those you should delegate to others, and those that should be automated or defaulted so that nobody has to make them at all. These criteria are not preferences. They are structural — grounded in the properties of the decision itself, not in your comfort level or your identity as a decision-maker.

The primitive is precise: know which decisions you must make yourself and which can be delegated. Not "try to delegate more." Not "learn to let go." Know. As in, have explicit criteria that produce a clear answer for any given decision. That is what we are building.

The irreversibility test: the single most important delegation criterion

Jeff Bezos articulated the clearest delegation heuristic in his 1997 letter to Amazon shareholders, and he has reiterated it in every annual letter since: decisions are either one-way doors or two-way doors. One-way doors are irreversible or nearly irreversible — once you walk through, you cannot walk back. Two-way doors are easily reversible — if the decision turns out wrong, you can return to where you started at low cost (Bezos, 1997).

One-way door decisions should be made carefully, slowly, with maximum input and your direct involvement. Two-way door decisions should be made quickly, by the person closest to the information, with your involvement limited to setting constraints.

The operational test is simple. For any decision, ask: If this decision turns out to be wrong, how difficult and expensive is it to reverse? If the answer is "very difficult" or "impossible," you are looking at a one-way door. Retain it. If the answer is "we can change course within a week at minimal cost," you are looking at a two-way door. Delegate it.

Most decisions are two-way doors. The vendor choice for a quarterly marketing campaign. The structure of the weekly team meeting. The prioritization of the next sprint. The hiring of a contractor for a three-month engagement. These are all reversible at reasonable cost. And yet leaders routinely treat them as one-way doors — deliberating for weeks, convening committees, requesting multiple rounds of analysis — because they have not applied the irreversibility test explicitly.

The consequence is organizational paralysis disguised as thoroughness. Decisions that should take hours take weeks. People who should be empowered to act are waiting for approval. And the genuinely irreversible decisions — entering a new market, choosing a technical architecture, making a key hire — get insufficient attention because the leader's cognitive budget was spent on two-way doors.

The Eisenhower matrix: urgency and importance as delegation filters

Dwight Eisenhower reportedly said, "What is important is seldom urgent, and what is urgent is seldom important." Whether or not the attribution is accurate, the principle is foundational, and the Eisenhower Matrix that emerged from it provides a second delegation criterion that complements the irreversibility test (Covey, 1989).

The matrix sorts decisions along two axes — urgency and importance — creating four quadrants:

Quadrant 1: Urgent and Important. These require your direct involvement now. A production system is down. A key employee just resigned. But if you live in Quadrant 1 permanently, that is a design failure — you have not built the systems that prevent recurring crises.

Quadrant 2: Important but Not Urgent. Strategic planning. Relationship building. System design. These determine your long-term trajectory and are easy to defer because no one is forcing them today. This is where your highest-value personal decision-making belongs. Protect this quadrant ruthlessly.

Quadrant 3: Urgent but Not Important. This is the delegation quadrant. The email needs a reply, the meeting needs scheduling, the report needs formatting. These feel pressing because someone is waiting, but they do not require your judgment. Delegate them to a person or to a process. Urgency creates a psychological pull that mimics importance. Resist it.

Quadrant 4: Neither Urgent nor Important. These are decisions to eliminate, not delegate. Automate them, default them, or stop doing the thing that generates the decision.

The matrix maps to delegation strategy: retain Quadrants 1 and 2, delegate Quadrant 3, eliminate Quadrant 4. Combined with the irreversibility test, you get a two-dimensional filter. An irreversible Quadrant 2 decision — choosing your technical architecture — demands your deepest attention. A reversible Quadrant 3 decision — selecting a catering vendor — should never reach your desk.

RACI: formalizing decision roles

The Eisenhower Matrix tells you which decisions to delegate. RACI tells you how to delegate them — specifically, how to assign clear roles so that delegation does not collapse into ambiguity.

RACI is an acronym for four decision roles: Responsible (the person who does the work and makes the decision), Accountable (the person who is ultimately answerable for the outcome — there must be exactly one), Consulted (people whose input is sought before the decision is made), and Informed (people who need to know the decision was made but do not participate in making it). The framework was developed in the 1950s for project management and has been adapted across industries since (Smith & Erwin, 2005).

The power of RACI is not the acronym itself — it is the discipline of making the roles explicit. Most delegation fails not because the wrong person was chosen, but because multiple people believed they were the decision-maker, or no one did. When three people think they are Accountable for the same decision, you get conflict. When zero people think they are Accountable, you get drift. RACI eliminates both failure modes by forcing a single answer to the question: who owns this decision?

Here is how to apply RACI to delegation:

For decisions you retain: You are both Responsible and Accountable. Identify who should be Consulted (people with relevant expertise or who will be affected) and who should be Informed (people who need to know the outcome).

For decisions you delegate: The delegate is Responsible. You remain Accountable — delegation transfers the work, not the accountability. Specify who should be Consulted (this is where you provide input without holding the decision) and who should be Informed.

For decisions you eliminate: No RACI assignment is needed. Set a default and move on.

McKinsey's research on organizational decision-making found that one of the most common dysfunctions in large organizations is what they call "decision ambiguity" — situations where it is unclear who has the authority to make a given decision. Their finding: organizations that explicitly assign decision roles outperform those that leave roles implicit, not because the role assignments are themselves valuable, but because the process of making them forces clarity about decision boundaries, authority levels, and escalation paths (Rogers & Blenko, 2006).

Management by exception: defining the boundaries of delegated authority

RACI tells you who makes the decision. Management by exception tells you when a delegated decision should be escalated back to you.

The principle, articulated by Frederick Winslow Taylor and developed extensively in management science, is straightforward: subordinates handle all decisions that fall within defined parameters. Only deviations beyond those parameters — exceptions — are escalated to the manager. The manager's role shifts from making decisions to defining the parameters within which others make decisions (Taylor, 1911).

This is more than a time-management technique. It is a fundamentally different model of leadership. Instead of being the person who makes decisions, you become the person who designs decision-making systems. Instead of reviewing every output, you define the acceptable range and only engage when the output falls outside it.

Practical implementation requires three elements:

Clear standards. The delegate must know what "normal" looks like. If you delegate pricing decisions, define the acceptable range, the discount authority, and the margin floor. Without standards, the delegate either plays it too safe or has no reference point for what constitutes an exception.

Defined thresholds. "Use your judgment" is not a threshold. "Escalate any discount above fifteen percent or any deal with payment terms beyond net-60" is. Ambiguous thresholds produce two failure modes: under-escalation (they make a decision that should have been yours) and over-escalation (they bring you everything because they are unsure where the line is).

Trust in the system. If you audit every delegated decision, you have not delegated — you have created an approval process and called it delegation. You must tolerate decisions you would have made differently, as long as they fall within the acceptable range.

Appelo's seven levels: delegation is not binary

One of the most damaging assumptions about delegation is that it is all-or-nothing: either you make the decision yourself, or you hand it off completely. This false binary causes leaders to retain decisions they should delegate (because full delegation feels too risky) and to under-support decisions they do delegate (because they assume delegation means zero involvement).

Jurgen Appelo's Management 3.0 framework identifies seven levels of delegation, each representing a different balance of authority between the leader and the delegate (Appelo, 2011):

1. Tell. You make the decision and announce it. No input requested, no discussion.
2. Sell. You make the decision and explain your reasoning, seeking buy-in but not input.
3. Consult. You gather input from others, then make the decision yourself.
4. Agree. You and the delegate make the decision together, by consensus.
5. Advise. The delegate makes the decision, but seeks your input first.
6. Inquire. The delegate makes the decision and informs you afterward.
7. Delegate. The delegate makes the decision with full autonomy. You may not even be informed.

The seven levels give you precision in delegation design. A hiring decision for a senior role might be Level 3 — you consult the team but make the final call. A hiring decision for an intern might be Level 6 — the hiring manager decides and tells you afterward. The budget allocation for a new initiative might be Level 4 — you and the finance lead agree together. The choice of project management tool might be Level 7 — the team decides and you do not need to know.

The practical value is that you can calibrate delegation to the decision's stakes, the delegate's experience, and the organizational context — without defaulting to either "I decide everything" or "do whatever you want." Most delegation failures happen at the extremes. The seven levels give you five additional options between them.

The Drucker filter: what only you can do

Peter Drucker, in The Effective Executive, offered perhaps the most ruthless delegation criterion: effective executives "get rid of everything that can be done by somebody else so that they can focus on the things that only they can do" (Drucker, 1967). This is not about efficiency. It is about irreplaceability.

The question is not "Can I do this well?" You can probably do most things well — that is likely how you got to your current position. The question is "Does this require something that only I can provide?" If the answer is no — if someone else could make this decision with acceptable quality — then making it yourself is a misallocation of your most constrained resource: your attention.

Drucker's filter produces a short list of decisions that only you can make:

• Decisions that require your unique knowledge. You have context that no one else has — knowledge of the full strategic picture, relationships with key stakeholders, understanding of organizational history that informs the present.
• Decisions that require your positional authority. Some decisions can only be made at your level because they cross organizational boundaries, commit resources above a certain threshold, or set precedent that affects the entire organization.
• Decisions that shape identity and direction. What does this team stand for? What do we refuse to do? What defines our culture? These are not delegatable because they are, in essence, decisions about who you are — and identity decisions must come from the identity-holder.

Everything else — and it is most things — can be made by someone closer to the information, with more time to deliberate, and with greater investment in the outcome. Your job is not to make those decisions. Your job is to build the system within which those decisions get made well.

The AI parallel: human-in-the-loop as computational delegation criteria

Every delegation criterion in this lesson maps directly to one of the most consequential design problems in modern AI: deciding which decisions an AI agent handles autonomously and which require human oversight. The field calls this "human-in-the-loop" design, and the parallels are exact (Permit.io, 2025).

The irreversibility test applies directly. AI systems that execute irreversible actions — transferring funds, deleting data, making medical recommendations — require human oversight at the point of commitment. Reversible actions — drafting an email, suggesting a schedule, proposing a code change — can be fully automated. One-way doors need a human. Two-way doors do not.

Management by exception translates to confidence thresholds. Modern agentic systems define boundaries: if the model's confidence exceeds 0.85, it acts autonomously. Below that, it escalates. The human reviews only the exceptions — Taylor's principle, implemented in code.

Appelo's seven levels map to what practitioners call "bounded autonomy architectures." A coding assistant operates at Level 5 — it writes code and requests review before committing. A spell-checker operates at Level 7 — fully autonomous. A medical diagnosis system operates at Level 3 — it presents analysis, but the physician decides. The level is calibrated to stakes, system reliability, and cost of error — the same variables you use when delegating to a human.

The deepest parallel: organizations that deploy AI agents effectively are the ones that have already solved the human delegation problem. If you cannot articulate which decisions require your judgment, you cannot design an AI system that escalates appropriately. The delegation criteria you build in this lesson are design specifications for any system — human, algorithmic, or hybrid — that acts on your behalf. And the failure modes are identical. A system that escalates everything adds latency without value. A system that never escalates will eventually make a consequential decision that should have been yours. The calibration problem is the same: find the boundary between autonomy and oversight that maximizes decision quality across the full portfolio.

Building your delegation criteria

Here is the operational framework. For any decision that reaches you, apply these five filters in sequence:

Filter 1: Irreversibility. Is this decision reversible at low cost? If yes, delegate it. If no, proceed to Filter 2.

Filter 2: Unique knowledge. Does this decision require knowledge or context that only you possess? If no, delegate it. If yes, proceed to Filter 3.

Filter 3: Authority level. Does this decision require your positional authority — does it cross boundaries, set precedent, or commit resources above the threshold you have defined? If no, delegate it. If yes, proceed to Filter 4.

Filter 4: Identity and direction. Does this decision define who your team is, what you stand for, or where you are going? If no, delegate it. If yes, retain it.

Filter 5: Development value. Even if you could make this decision, would delegating it develop someone else's judgment in a way that pays long-term dividends? If yes, delegate it at an appropriate level (Level 3 through 5) with mentoring support.

For every decision you delegate, specify three things: the owner (one person, not a committee), the constraints (the parameters within which they have full authority), and the escalation triggers (the specific conditions under which the decision comes back to you).

This is not a one-time exercise. It is a practice. As your team's capability grows, decisions that once required your involvement become safely delegatable. As circumstances change, decisions that were once routine may require your direct attention. The criteria are stable. The specific allocation is dynamic.

The goal is not to delegate everything. The goal is to delegate everything that does not require you — so that the decisions you do make receive the full weight of your attention, your experience, and your judgment. That is not less leadership. That is the only kind of leadership that scales.