Reference
Frameworks behind problem framing
Part of the Problem Framing skill · loaded on demand from SKILL.md
This file holds the intellectual foundation in depth: the published methods, step by step, with the reasoning that makes each step work. SKILL.md carries the operating procedure; come here when a step feels arbitrary and you need to know why it exists, or when the user challenges the method and you need to defend it with attribution.
Contents
- Sources and attribution
- Conn and McLean's seven-step problem-solving cycle (2018)
- The McKinsey issue tree and MECE (Rasiel 1999; Rasiel and Friga 2001)
- Issue trees versus hypothesis trees
- Minto's introduction: Situation, Complication, Question (1987, 2009) and the SCR variant
- The so-what test and the one-day answer
- Prioritisation: impact against ability to influence
- Where the methods disagree and how to reconcile them
1. Sources and attribution
| Method | Source | What it contributes here |
|---|---|---|
| Seven-step problem-solving cycle | Charles Conn and Robert McLean, Bulletproof Problem Solving, Wiley, 2018 | The end-to-end procedure; the problem-definition criteria; logic-tree taxonomy; the prioritisation matrix; the one-day answer |
| MECE, initial hypothesis, issue tree, "don't boil the ocean", 80/20 | Ethan Rasiel, The McKinsey Way, McGraw-Hill, 1999; Rasiel and Paul Friga, The McKinsey Mind, 2001. Minto coined "MECE" at McKinsey in the 1960s; Rasiel's books are the public account of how the firm uses it | The decomposition discipline and the practice of forming a hypothesis before gathering data |
| Pyramid Principle; Situation, Complication, Question | Barbara Minto, The Pyramid Principle, first published 1987; revised edition 2009 | How to find the question the decision-maker is actually asking, and how to state an interim answer |
| SCR (Situation, Complication, Resolution) | Consulting practice derived from Minto's SCQ; the Resolution replaces the Question when you already have an answer or a hypothesis | The form of the day-one answer |
Do not cite internal McKinsey training documents. Everything below is grounded in the public books above. Where a practice is described as "how firms do it", it is a paraphrase of these authors, not a quotation from a firm.
2. Conn and McLean's seven-step cycle (2018)
Conn and McLean, both former McKinsey partners, published the seven steps as a general-purpose method. The steps, in their order:
- Define the problem
- Disaggregate the issues
- Prioritise the issues (prune the tree)
- Build a workplan and timetable
- Conduct critical analyses
- Synthesise findings from the analysis
- Prepare a powerful communication
The problem-framing skill runs steps 1 to 4 and hands steps 5 to 7 to the evidence skills, hypothesis-stress-testing, and executive-communication.
Why the cycle exists
The failure the cycle prevents is well known to anyone who has watched a team spend three weeks "gathering data" and then discover the client wanted an answer to a different question. Each early step is cheap and each late step is expensive; the method front-loads the cheap steps so that expensive analysis is only done on issues that are (a) actually part of the question, (b) large enough to matter, and (c) within the client's power to change.
The authors insist the cycle is iterative. A first pass through steps 1 to 4 should take hours, not weeks, and produce a one-day answer. Later passes refine the tree and the hypotheses as analysis returns. Treating the cycle as a linear waterfall is the most common misreading.
Step 1: Define the problem
Conn and McLean list the qualities of a good problem definition. Paraphrased:
- Expressed as an outcome, not an activity ("restore margin to 8 percent" rather than "review pricing").
- Specific and measurable wherever possible.
- Time-bound.
- Explicitly bounded by the decision-maker's values and constraints: what they will and will not do.
- Clear about the accuracy needed and the scale of ambition.
- Wide enough to allow for creative and unexpected answers, not pre-loaded with a solution.
- Solved at the highest level possible (for the organisation as a whole, not for the department that happened to raise it).
They operationalise this through a problem-definition worksheet with fields for the decision-maker, the success criteria, the constraints, the scope (in and out), the time frame, and the required accuracy. The skill's worksheet in templates.md follows this structure.
Mechanism: every field in the worksheet closes off a specific way the engagement can fail. "Decision-maker" prevents solving for the person who briefed you rather than the person who will decide. "Success criteria" forces the mandate into a measurable outcome. "Constraints" and "scope out" stop the team from spending effort on options the client will never take. "Required accuracy" stops over-engineering: a board go/no-go rarely needs sizing tighter than plus or minus 20 percent.
Step 2: Disaggregate the issues
The tool is the logic tree. Conn and McLean describe several types:
- Factor or component trees: break a quantity into its arithmetic parts (profit = revenue minus cost).
- Deductive logic trees: a chain of reasoning where each level follows necessarily from the one above.
- Inductive logic trees: groupings where the branches are related by a shared attribute rather than a formula, reasoning from particular observations up to a general conclusion; the grouping is only as sound as the attribute chosen, which is why these trees need the MECE tests more than factor trees do.
- Hypothesis trees: the branches are assertions to be proven or disproven.
- Decision trees: branches are choices and their consequences.
The first pass usually uses a factor or deductive tree, because arithmetic identities are the most reliable source of MECE. The mechanism: an identity (margin equals price minus cost, divided by price) is MECE by construction; a brainstormed list is not.
Step 3: Prioritise and prune
The authors use a two-by-two matrix: potential size of impact on one axis and the client's ability to influence on the other. Work first on branches that are high on both. Prune branches that are low on either, and record why they were pruned so they can be revived if the analysis surprises you.
Mechanism: this is Rasiel's 80/20 turned into a procedure. A handful of branches will contain most of the answer. Pruning early is what makes the workplan feasible inside the deadline.
Step 4: Workplan and timetable
For each surviving branch: the issue, the hypothesis, the analysis that would confirm or refute it, the data source, the owner, and the due date. Conn and McLean stress that workplans should be lean and staged: the first workplan covers the next one to two weeks in detail and the rest of the engagement only in outline, because the analysis will change the tree.
Steps 5 to 7
Analysis (with the discipline of designing analyses to disprove rather than confirm), synthesis (moving from the one-day answer to a pyramid), and communication (Minto's structure). These are the province of other skills in the engine.
3. The McKinsey issue tree and MECE
Rasiel (1999) described MECE as the organising principle of McKinsey's problem solving: a decomposition where no two branches overlap (mutually exclusive) and, taken together, the branches cover the whole of the parent (collectively exhaustive). The term itself is Barbara Minto's, from her time at McKinsey; Rasiel's book is the public account of the practice. He also described the initial hypothesis formed at the start of the engagement, before data is gathered, and the issue tree that maps the logical structure of the investigation.
Why MECE matters
Overlap causes double counting and turf disputes (two analysts sizing the same dollars under different headings). Gaps cause missed answers (the real driver was in no branch, so nobody looked). Both errors are invisible in a plausible-looking tree and become visible only when the synthesis fails to add up. MECE is the test that catches them before the analysis starts.
Practical MECE tests at a single level
- The addition test (for quantitative branches): do the branches sum to the parent, with no residual larger than rounding?
- The classification test (for categorical branches): could any single fact, customer, product or cost be filed under two branches? If so, the level is not mutually exclusive.
- The "what else" test: name three things the parent could contain and check each lands in exactly one branch. If any lands nowhere, the level is not exhaustive.
- The count test: two to five branches per level. One branch is not a decomposition; more than five usually means two levels have been collapsed into one.
- The same-cut test: every branch at a level must be a slice on the same dimension (all by customer segment, or all by cost line, not a mixture).
Sources of MECE that are reliable
- Arithmetic identities (profit, unit economics, capacity, conversion funnels).
- Process stages (order to cash, lead to close), provided the hand-offs are clean.
- Exhaustive categorical splits (internal or external; existing customers or new; fixed or variable).
- Time (before, during, after; current period, next period).
Frameworks such as the 4Ps, 3Cs or Five Forces are not automatically MECE for a given problem; they are lists of things worth considering. Use them to check exhaustiveness, not as the branches themselves.
The "and other" branch
A branch labelled "other" is sometimes necessary for exhaustiveness. It is acceptable at level 2 or below if it is small (under 10 percent of the parent) and explicitly sized. At level 1 it is a sign the decomposition dimension is wrong.
4. Issue trees versus hypothesis trees
The two look alike on a page and are often confused. The distinction matters because they are built at different moments and used for different things.
| Issue tree | Hypothesis tree | |
|---|---|---|
| Node content | Questions or issues ("Is the margin decline driven by price or by cost?") | Assertions ("The margin decline is driven by price realisation, not cost inflation") |
| Purpose | Map the full space of what could matter | State what you believe and what must be true for it to hold |
| Built when | You do not yet know where the answer lies | You have enough knowledge to place a bet |
| MECE requirement | Strict at every level | Strict for the supporting assertions under each hypothesis |
| Failure it prevents | Missing the real driver | Boiling the ocean; analysing everything equally |
| Feeds | Prioritisation, then hypotheses | Analysis design, then hypothesis-stress-testing |
Rasiel's advice to "figure out the solution before you gather data" describes the hypothesis tree. Conn and McLean's warning against jumping to solutions describes the issue tree. Both are right, in sequence: build the issue tree to make sure nothing is missed, prune it, then convert the surviving branches into hypotheses. A hypothesis tree built without an issue tree behind it is the source of the classic consulting failure: an elegant, well-analysed answer to the wrong question.
A useful practical rule: a diagnostic issue tree ("why is X happening?") comes before a solution tree ("how do we fix X?"). If the cause is unknown, the solution tree will be built on guesses. If the cause is already known and agreed, skip the diagnostic tree and state the cause in the Situation and Complication.
5. Minto's introduction: Situation, Complication, Question
Minto (1987; revised 2009) designed the pyramid for writing, but its introduction is the sharpest tool available for problem framing, because it forces you to locate the question the reader is actually asking before you answer anything.
- Situation: a statement about the subject that the reader already knows and will accept as true without argument. It establishes shared ground. If the reader could disagree with the Situation, it is not a Situation, it is a claim.
- Complication: what has changed in the Situation to create tension. Something happened, or is about to, or a proposed change threatens the stable state. The Complication is what raises the Question.
- Question: the question that arises naturally in the reader's mind from the Complication. It must be the reader's question, not the writer's.
- Answer: the top of the pyramid; the main point.
Minto notes the elements can be reordered for effect (leading with the Complication for a reader who is already worried; leading with the Answer for one who wants it immediately; leading with the Question for one who is sceptical). The content of the four elements does not change with the order.
Why SCQ is a framing tool, not only a writing tool
A mandate as delivered by a client is usually a Complication ("margins are down") or an Answer in disguise ("we need to cut costs"). It is rarely a Question. Rewriting the mandate as an explicit SCQ exposes which piece is missing. If you cannot write a Situation the client would accept, you do not understand the baseline. If you cannot write a Complication, there is no problem, only a wish. If the Question you write is not one the decision-maker would recognise as theirs, you are about to solve the wrong problem.
SCR: Situation, Complication, Resolution
In consulting practice the Question is often replaced by a Resolution: the current best answer, stated as a hypothesis. This is the shape of the day-one answer (section 6). SCQ is for framing; SCR is for stating a position. The framing skill uses SCQ to find the question and SCR to record the day-one hypothesis.
6. The so-what test and the one-day answer
The so-what test
Rasiel and Friga describe the "so what" as the question a McKinsey manager asks of every finding: what does this mean for the client, and what should they do about it? Applied to an issue tree it becomes a pruning rule. For each branch ask: if the analysis of this branch came back with the most extreme plausible answer, what would the client do differently? If the honest reply is "nothing", the branch fails the test. Delete it, or demote it to a footnote in the workplan.
Mechanism: the test converts an abstract judgement ("is this important?") into a concrete one ("would it change a decision?"). It is the single most effective way to keep a tree from sprawling.
The one-day answer
Conn and McLean ask their teams to be able to state, at any point in a project, a coherent summary of their best current understanding: the Situation, the Complication (what they have observed), and the Resolution (the current best thinking on what to do). This is the one-day answer. On day one it is mostly hypothesis; by the final week it is mostly evidence.
Mechanism: forcing an interim answer exposes what the team does not know, and it does so in the language of the eventual deliverable. A team that cannot state a one-day answer has not finished framing.
7. Prioritisation: impact against ability to influence
Conn and McLean's matrix:
| Low ability to influence | High ability to influence | |
|---|---|---|
| High potential impact | Watch: size it roughly, note it, do not spend team time yet | Work here first |
| Low potential impact | Prune | Prune unless trivially cheap to resolve |
Two refinements from practice:
- "Ability to influence" must be judged inside the constraints from the problem statement. A branch the client could act on in principle but has ruled out (say, closing owned distribution centres) belongs in the low-influence column for this engagement.
- "Impact" should be sized in the unit of the success criterion (dollars of EBITDA, points of share, days of cycle time), using a rough estimate. An order-of-magnitude number is enough; the point is to rank, not to model.
8. Where the methods disagree and how to reconcile them
Hypothesis-first (Rasiel) against structure-first (Conn and McLean). Reconcile by sequence: structure for an hour, then hypothesise. The issue tree is the guard against a wrong hypothesis; the hypothesis is the guard against a bloated workplan.
Minto's Question against Conn and McLean's problem statement. They are the same object at different resolutions. The Question is one sentence; the problem statement adds the worksheet fields (decision-maker, criteria, constraints, scope, time, accuracy). Write the Question first and expand it into the statement.
MECE strictness. Some practitioners argue for approximate MECE at lower levels. Hold strict MECE at levels 1 and 2 and accept approximation at level 3 and below, where the branches become individual analyses rather than categories.