Skip to content

Toolkit — Issue tree

Gate: G2 Decompose. Category: structure substrate.

What problem it solves

The pyramid organises answers; many engagements arrive with only a question and need a tree whose top is the question being pursued. An issue tree is the pyramid's question-shaped cousin: it decomposes a top-level question into sub-questions, each of which is answered by branches of data, analysis, or judgement. Without the tree, the engagement tries to answer the top-level question directly and runs out of fidelity; with it, each branch is tractable on its own.

How it is used

A 90-minute G2 workshop, usually after the pyramid's top sentence has been provisionally written. The chair draws the top-level question and asks: to answer this, what do we have to answer first? Each child is a sub-question; the tree stops recursing when a question is answerable by the existing substrate (data, interview, document) or by one routable leaf. The tree is drawn on one sheet and attached to the G2 output.

Inputs

  • Framed problem paragraph.
  • The top-level question the engagement exists to answer — if the pyramid is already drafted, the question form of its top sentence.
  • The team's initial list of sub-questions.

Outputs

  • A question tree: top question, sub-questions, sub-sub-questions, each recursively until answerable.
  • A source column per leaf: where the answer will come from (data pull, interview, literature, analysis, decision).
  • A flagged list of unanswerable questions — leaves whose sources are not available within the engagement. These get a what would we need to answer this note and a decision on whether to pursue the source or close the leaf as unknown.

Visualisation

Issue tree — top-level question decomposed into sub-questions down to answerable leaves, each marked with a source Top question? what the engagement asks Sub-question 1? Sub-question 2? Sub-question 3? Leaf 1.1 data pull Leaf 1.2 interview Leaf 2.1 analysis Leaf 3.1 decision Leaf 3.2 unanswerable Every leaf has a source (or is flagged unanswerable).

Each node is a question. Each leaf (dashed) has a source — the substrate the answer will come from.

Anatomy

Top-level question. The question the engagement exists to answer. Written in its full form, not a shorthand. "What is driving the 16-minute growth in unloading times?", not "unloading."

Sub-questions. The questions that have to be answered to answer the parent. Each sub-question must, when answered, contribute to the parent's answer — otherwise it belongs on a different tree.

Leaves. Questions that are answerable by one identifiable substrate — a data pull, a single interview, a piece of analysis, a policy decision. A leaf with no identified source is not a leaf; it is a branch waiting for its children.

Source column. Per leaf: where the answer comes from. The engagement team uses this to parallelise work (leaves by source type), to estimate effort, and — critically — to identify which leaves are blocked on unobtainable sources.

Unanswerable leaves. Sometimes the answer is not within reach. Flag, name the unobtainable source, decide whether to pursue it (expand engagement scope), proxy it (use a weaker substrate), or close the leaf as unknown (and carry the unknown into G3 as a routing constraint).

Example

Paper trail — the freight-yard issue tree

G2 workshop following the pyramid, 80 minutes. Same team as the pyramid session.

T+0 — top question. From the pyramid's top sentence: "where did the 16 minutes go?" Ada draws the root.

T+10 — first-level sub-questions. The pyramid's three groups, reformulated as questions: How much time is in slot allocation? In dock-door coordination? In carrier-ETA misalignment? Three branches.

T+25 — second level. Under slot allocation, three sub-questions: What is the assignment-delay distribution? How often do reassignments happen? How often do dispatchers override the allocator's recommendation (when the allocator exists)? Each becomes a leaf.

T+45 — sources. Priya lists sources: assignment-delay is a data pull from the dispatcher's radio log; reassignment is a data pull; override is not measurable today because there is no allocator to override. Ada: "the override leaf is unanswerable ex-ante. Flag and carry forward — we learn it from the shadow stage of the rollout." Recorded.

T+60 — carrier-ETA branch. Leaves include ETA accuracy (data pull from carrier feed), paperwork-arrival delay (interview dispatchers), and priority-conflict rate (cross-cut with slot allocation; measured once, read twice).

T+75 — unanswerables. Three leaves are unanswerable within the engagement horizon: override rate (needs shadow stage), driver-side delays (driver data is not accessible, carriers will not share), third-party carrier rating of yard experience (no consent mechanism). Each flagged with what would we need and a routing note.

T+80 — close. Nine answerable leaves + three unanswerables, each with a source or a flagged-absence note. The tree and the sources feed G3 routing.

Pitfalls

Questions that are not questions. "Slot allocation" is a topic. The tree needs "how much time does slot allocation take?" — with a question mark and an answerable form.

Overlapping sub-questions. Two sub-questions whose answers would be the same data. The tree stops branching before that point.

Sources invented, not named. "We'll figure it out from ops" is not a source. "Radio log pull, dates W3–W5 2026, by shift" is.

Skipping the unanswerable flag. A leaf that isn't really answerable, presented as if it were, produces a confident wrong answer later. Flagging is more valuable than pretending.

Trees built by the team alone. A tree drawn only by the engagement team reproduces the team's blind spots. An outside reader (same as for the pyramid) is the trees' cheap insurance.

When not to use

  • Engagements where the pyramid form fits better (answer-shaped rather than question-shaped).
  • One-leaf engagements (single data pull, no decomposition).
  • Research-first engagements where the sub-questions are genuinely emergent and forcing a tree prematurely constrains what gets investigated.

Provenance

The issue-tree form is documented in Rasiel's The McKinsey Way [1] as the standard problem-decomposition substrate for consulting engagements. It is the question-shaped sibling of Minto's pyramid [2]. In the academic operations-research literature the same structure appears as goal decomposition in systems-engineering practice [3].

  • Minto pyramid. The answer-shaped cousin.
  • MECE. Same check; applied to sub-question groups.
  • Pair worksheet (G3). Unanswerable leaves with a candidate ML approach get recorded here.

Verification

[1] Rasiel EM. The McKinsey Way. McGraw-Hill; 1999. [verified] Documents the issue-tree as the standard consulting decomposition substrate.

[2] Minto B. The Pyramid Principle: Logic in Writing and Thinking. 3rd ed. Pearson; 2009. [verified] The pyramid's question-shaped sibling.

[3] NASA Systems Engineering Handbook. NASA SP-2007-6105 Rev 1. NASA; 2007. [verified] Goal decomposition in systems-engineering practice.