Reference
Worked example: Northgate Industrial Supply, gross margin decline
Part of the Hypothesis Stress Testing skill · loaded on demand from SKILL.md
This is one complete run of the skill, from problem statement to surviving hypothesis, with every matrix cell shown. All figures are rounded to the nearest 0.1 point of margin or nearest basis point; reconciliations are within 5 bps of rounding.
Situation and inputs (from problem-framing)
Client: Northgate Industrial Supply, a mid-market distributor of maintenance, repair and operating supplies. Revenue $640M trailing twelve months, six regions, 41 branches, about 38,000 active customer accounts. Two customer segments in the management accounts: national accounts (14 contracts with multi-site customers, priced cost-plus with quarterly indexation to vendor cost) and core accounts (everything else, priced from the list price book less branch-level discount).
Problem statement: "Why has Northgate's gross margin fallen from 27.4% to 25.1% over the last eight quarters (Q1 Y1 to Q4 Y2), and which causes can be reversed within twelve months?" Decision-maker: CEO, reporting to a board that has asked for an answer before the Q1 Y3 meeting in five weeks. Required accuracy: causes sized to within 20 bps. Scope: gross margin only; operating cost is a separate workstream.
Day-one answer, held strongly by the CEO and the CFO: "National accounts are killing the margin. They have grown from 31% to 39% of revenue and they run ten points below core. Fix: slow down national-account growth or renegotiate."
Analysis budget: an engagement manager and two analysts for two weeks, with data access to the segment P&L, order-level pricing data, vendor cost letters, the price-book change log, product P&L, CRM and the controller.
Team's own prior (recorded because it differs from the CEO's): the engagement manager's experience in distribution is that eight-quarter margin declines during an inflationary period are usually pass-through lag, not mix.
Stage 1: question and day-one answer
Question: what explains the 230 bps decline in reported gross margin between Q1 Y1 and Q4 Y2, and how much of it is reversible within twelve months?
Day-one answer as written for attack: "Customer-mix shift toward national accounts explains most (more than half) of the 230 bps decline. If this is true, margin within core accounts should be roughly flat over the period." The second sentence is the killing observation; it was written on day one and turned out to be the sentence that mattered.
Stage 2: hypothesis set
| ID | Hypothesis | Killing observation | Champion |
|---|---|---|---|
| H1 | Customer mix: national accounts rising from 31% to 39% of revenue, at about ten points lower margin than core, explains more than half of the decline; mechanism is arithmetic blend, so core-account margin should be near flat | Dies if like-for-like core-account margin has fallen more than 50 bps | CEO, CFO |
| H2 | Discount creep: branch discretionary discounting in core accounts has widened by more than 100 bps of realised price across most regions since Q1 Y1, explaining at least 80 bps of the decline | Dies if realised discount off list has moved less than 100 bps, or the movement is confined to one or two regions | Head of sales (who believes branches are "giving it away") |
| H3 | Pass-through lag: vendor cost increases on the top-20 vendor lines have reached COGS faster than list prices were updated, compressing margin in list-priced core accounts by at least 50 bps blended while cost-plus national accounts were unaffected | Dies if list-price increases on top-20 lines kept pace with vendor cost increases within 30 days, or if national-account margin also fell | Engagement manager |
| H4 | Product mix: growth of the e-commerce channel has shifted sales from private label (about 38% margin) to branded commodity lines, cutting private-label share by three points or more and explaining at least 60 bps | Dies if private-label revenue share is flat within one point | COO (owns e-commerce) |
| H5 | Accounting artefact: the Q3 Y1 reclassification of outbound freight from operating expense to COGS explains most of the reported decline; the like-for-like decline is under 100 bps | Dies if the decline restated for the reclassification still exceeds 100 bps | Audit committee chair |
The null hypothesis ("no structural change; normal quarterly noise") was considered and excluded because a 230 bps move against a prior five-year range of plus or minus 40 bps is outside noise. H5 is the mandatory artefact hypothesis. All five have a plausible champion.
Stage 3: evidence list
| ID | Evidence | Source, date, reliability |
|---|---|---|
| E1 | Margin decline is present in all six regions, range 180 to 290 bps | Regional P&L, Q4 Y2, audited |
| E2 | National accounts grew from 31% to 39% of revenue; their margin was 20.6% at start and 20.8% at end; core accounts 30.5% at start and 28.9% at end (like-for-like, before reclassification) | Segment P&L, audited, relies on CRM account-type tags (see A1) |
| E3 | Core-account margin fell 160 bps like-for-like over the period; national-account margin rose 20 bps | Segment P&L, as E2 |
| E4 | Realised discount off list in core accounts rose from 8.1% to 8.6% of list; increase concentrated in Region 2 (+110 bps) and Region 5 (+90 bps); other four regions under 30 bps | Order-level pricing data, 2.1M order lines, high reliability |
| E5 | Top-20 vendor cost increases averaged 6.5% over the period; Northgate list-price increases on the same lines averaged 3.9%; these lines are 58% of core-account revenue and 55% of COGS | Vendor price letters (20 files) and price-book history, high reliability for the 20 vendors, no data for the remaining 45% of COGS |
| E6 | Q3 Y1 reclassification moved $4.1M annualised outbound freight from opex into COGS, equal to 64 bps of revenue; restated like-for-like decline is 166 bps | Controller memo dated Q3 Y1, single source, not yet checked against audit workpapers |
| E7 | Private-label share of revenue 14.2% at start, 14.6% at end; private-label margin 38.1% to 37.9% | Product P&L, audited |
| E8 | Median lag from vendor cost-letter effective date to list-price update rose from 21 days (Y0) to 74 days (Y2); the pricing team lost two of five staff in Q2 Y1 and did not backfill | Price-book change log; HR headcount records; high reliability |
| E9 | All 14 national-account contracts are cost-plus with quarterly indexation to vendor cost | Contract review, high reliability |
| E10 | Four of six regional VPs attribute the decline to "competitive pricing pressure from the nationals" | Interviews, testimony |
| E11 | Absence: no deterioration in core-account retention (94.1% to 94.3%) or quote win rate (flat at 61%) over the period | CRM, high reliability |
| A1 | Team assumption: CRM account-type tags (national versus core) are accurate at both dates | Untested; underpins E2 and E3 |
E11 is the dog that did not bark: if discounting had widened in response to competitive pressure, some pressure should be visible in retention or win rates. A1 is listed because, per Heuer, assumptions drive the judgment and must be exposed.
Stage 4: matrix
Cell question: if this hypothesis were true, how likely is it that I would see this evidence? CC strongly consistent, C consistent, N no bearing, I inconsistent, II strongly inconsistent.
| ID | Evidence (short) | H1 Mix | H2 Discount | H3 Pass-through | H4 Product mix | H5 Artefact | Diagnosticity |
|---|---|---|---|---|---|---|---|
| E1 | All regions down 180 to 290 bps | C | C | C | C | C | Low |
| E2 | National share 31% to 39%; national margin flat; core margin down | CC | N | C | N | N | Medium |
| E3 | Core margin down 160 bps like-for-like; national up 20 bps | II | C | CC | C | I | High |
| E4 | Discount 8.1% to 8.6%, concentrated in two regions | N | I | N | N | N | High |
| E5 | Vendor cost +6.5%, list +3.9%, on 58% of core revenue | N | N | CC | N | II | High |
| E6 | Reclassification worth 64 bps; restated decline 166 bps | N | N | N | N | I | High |
| E7 | Private-label share flat at 14.2% to 14.6% | N | N | N | II | N | High |
| E8 | Price-update lag 21 to 74 days after staff loss | N | N | CC | N | N | Medium |
| E9 | National contracts are cost-plus, quarterly indexed | C | N | CC | N | N | Medium |
| E10 | VPs blame competitive pricing pressure | C | C | C | N | N | Low |
| E11 | No churn or win-rate deterioration in core | N | I | C | N | N | Medium |
Notes on I and II cells, written at the time of rating:
- E3/H1 (II): H1 predicts core margin near flat. Core margin fell 160 bps, which on 61% of revenue is 98 bps blended, more than half of the 166 bps like-for-like decline on its own. H1 as "most of the decline" cannot survive this row. The arithmetic mix effect, holding segment margins at their start values and moving the share to 39%, is 27.43% minus 26.64%, which is 79 bps.
- E3/H5 (I): an artefact applied uniformly cannot produce a like-for-like within-segment decline after restatement.
- E4/H2 (I): direction is right, magnitude is not. A 50 bps widening of discount on core revenue moves core margin from 30.5% to about 30.1%, a 38 bps effect within core and about 23 bps blended, against a claim of at least 80 bps. Concentration in two regions also contradicts "across most regions".
- E5/H5 (II): a real gap of 2.6 points between vendor cost and list-price movement on 55% of COGS is incompatible with the decline being mainly an accounting artefact.
- E6/H5 (I): the reclassification is real and worth 64 bps, but the restated decline of 166 bps is well above the 100 bps threshold in H5's killing observation.
- E7/H4 (II): private-label share rose 0.4 points; H4 required a fall of three points or more.
- E11/H2 (I): H2's mechanism (competitive pressure forcing branches to discount) predicts some visible pressure in retention or win rate; none is present. Weaker than E4 because absence is easier to explain away.
Stage 5: refinement
- E1 and E10 removed from the inference (Low diagnosticity) and kept in the appendix. E10 is retained in the narrative only because the VPs' "we cannot raise prices" is what a pass-through lag looks like from a branch.
- H1 reworded from "explains most of the decline" to two versions: H1a "explains most" (refuted by E3) and H1b "explains about 80 bps through pure mix arithmetic" (alive, consistent with E2, E3, E9). The rewording is recorded as a split, not a rescue: H1b keeps a killing observation (dies if the recomputed mix effect is under 50 bps).
- H5 split the same way: H5a "explains most" (refuted) and H5b "explains 64 bps" (alive, supported by E6).
- Candidate new hypothesis from the evidence: e-commerce small orders absorbing freight now sitting in COGS. Computed at 1.1 points lower order margin on 21% of orders, about 15 bps blended, which is inside the unexplained residual (stage 6). Parked, noted as a possible explanation of the residual, not added to the matrix.
- H2 and H3 are separated by E4, E5, E8 and E11, so no merge is needed.
Stage 6: tentative conclusions by elimination
Evidence against each hypothesis, with reliability:
| Hypothesis | Inconsistent rows | Reliability of those rows | Standing |
|---|---|---|---|
| H3 Pass-through lag | None | n/a | Leader: least evidence against, supported by E3, E5, E8, E9 |
| H1b Mix, about 80 bps | None | n/a | Alive as the largest single cause |
| H5b Artefact, 64 bps | None | n/a | Alive as a component |
| H2 Discount creep | E4 (I), E11 (I) | E4 high, E11 high | Refuted as a primary cause; contributor of about 23 bps |
| H1a Mix, most of decline | E3 (II) | High, subject to A1 | Refuted |
| H4 Product mix | E7 (II) | High | Refuted |
| H5a Artefact, most of decline | E3 (I), E5 (II), E6 (I) | High | Refuted |
Reconciled decomposition of the reported 230 bps decline:
| Component | Basis | Blended effect |
|---|---|---|
| Freight reclassification (H5b) | E6: $4.1M on $640M | 64 bps |
| Customer mix (H1b) | E2: share 31% to 39% at start margins 20.6% and 30.5% | 79 bps |
| Pass-through lag in core accounts (H3) | E5: lines at 29% margin, cost +6.5%, price +3.9%, gives 178 bps on 58% of core revenue, 103 bps within core, times 61% core share | 63 bps |
| Discount creep in core accounts (H2) | E4: 50 bps of price on core, 38 bps within core, times 61% | 23 bps |
| National-account margin improvement | E3: +20 bps on 39% share | (8 bps) favourable |
| Unexplained residual | Core decline 160 less 103 less 38 equals 19 bps within core, times 61% | 12 bps |
| Total | 233 bps against 230 reported |
Reconciliation is within rounding tolerance. The residual of 12 bps is smaller than the required accuracy of 20 bps and is plausibly the parked e-commerce freight effect.
The day-one answer named the largest real cause (mix, 79 bps) but not the largest addressable one (pass-through lag, 63 bps, reversible in two quarters by restaffing and automating the price-update cycle). The mix effect is a consequence of a deliberate national-account strategy and is a strategic choice, not a fix.
Stage 7: the refutation plan as it was written on day two
This is the plan that produced E3 through E9 and E11. It was agreed with thresholds before data was pulled.
| Hypothesis | Refutation condition (must hold for H to survive) | Known on day two? | Analysis and source | Result that kills H | Owner | Days | Also discriminates |
|---|---|---|---|---|---|---|---|
| H1 | Core-account margin near flat | No | Segment P&L like-for-like, restated for reclassification | Core margin down more than 50 bps | Analyst A | 2 | H5 |
| H1 | National margin about ten points below core | Yes (E2) | n/a | n/a | |||
| H2 | Realised discount widened over 100 bps | No | Order-level discount off list by region and quarter | Widening under 100 bps or confined to two regions | Analyst B | 3 | H3 (a lag shows as list not moving, not as discount moving) |
| H2 | Competitive pressure visible | No | Retention and win rate by quarter | Both flat | Analyst B | 1 | |
| H3 | List prices lagged vendor cost on top-20 lines | No | Vendor cost letters versus price-book history | Median lag under 30 days or price increases within 1 point of cost increases | Analyst A | 3 | H5 |
| H3 | Lag hits core, not national | No | Contract review of 14 national accounts | Any national contract priced from list | EM | 1 | H1 |
| H4 | Private-label share fell 3 points or more | No | Product P&L | Share flat within 1 point | Analyst B | 1 | |
| H5 | Restated decline under 100 bps | No | Controller memo, restated series | Restated decline over 100 bps | EM | 1 | H1, H3 |
Sequence: H5 and H4 first (one day each, cheap, and H5 changes the denominator for everything else), then H1's core-margin check (it is the day-one answer's killing observation), then H2 and H3 in parallel. Nine working days in total. The plan had seven analyses (one condition was already known from E2), of which four killed something; the remaining three confirmed conditions for the survivor.
Stage 8: sensitivity check
Linchpin rows: the ranking depends on E3, E5 and E6. E4 and E7 refute hypotheses no longer in contention for the lead.
| Row | Vulnerability | Confirmation check | Distance to flip | Status |
|---|---|---|---|---|
| E3 (and E2) | Depends on A1: CRM account-type tags. If national revenue is tagged as core, the core decline is overstated | Audit tags on the top 200 accounts (78% of revenue) against contract list | If 15% of national revenue is mis-tagged as core at both dates, core decline recomputes to about 142 bps, the true core share falls to about 54%, and the decomposition must be rebuilt (H3 blended becomes about 56 bps, and the within-core components, 103 plus 38, would then explain the whole core decline with no residual, so the E5 estimate would be at its ceiling); H1a stays refuted (core still down more than 50 bps); ranking unchanged. To flip H1a to alive, core decline must fall below 50 bps, which requires more than 35% of national revenue to be mis-tagged, implausible for a contract-based segment | Check done day 8: 3.4% of top-200 revenue mis-tagged; E3 restated to 155 bps; no change |
| E5 | Covers 20 vendors, 55% of COGS. The other 45% might have been priced differently. Price-book "list" might not be the price actually quoted | Sample 30 quotes per region against price book; pull cost letters for vendors 21 to 40 | If list-price increases on top-20 lines were 5.5% rather than 3.9%, H3's blended estimate falls to about 24 bps, level with discount creep; H3 would still have no evidence against it but would no longer be the largest addressable cause | Quote sample: 92% of quotes within 0.5 points of price book; vendors 21 to 40 show a similar 2.4-point gap. Held |
| E6 | Single controller memo; amount not tied to audit workpapers | Tie $4.1M to audited freight expense movement between opex and COGS | If the reclassification were $2.0M rather than $4.1M, the like-for-like decline rises to 199 bps and the residual to about 45 bps, which would exceed the accuracy requirement and force a new hypothesis for the residual | Tied to workpapers day 6: $4.08M. Held |
All three linchpins survived their checks. The unexplained residual moved from 12 to 9 bps after the E3 restatement (core decline 155 bps: 103 pass-through, 38 discount, 14 residual within core, 9 blended).
Stage 9: report
Question: what explains the 230 bps decline in reported gross margin, and how much is reversible within twelve months?
Day-one answer and what happened to it: "National-account mix explains most of the decline" was refuted on day four by the core-account like-for-like margin. Mix is the largest real cause at 79 bps but explains a third, not most.
Standing of all hypotheses:
- H3 Pass-through lag on core accounts: surviving leader, 63 bps, no evidence against; supported by E3, E5, E8, E9. Reversible in two quarters.
- H1 Customer mix: alive as a component, 79 bps; refuted as "most" by E3. Structural; reversing it is a strategic choice, not a fix.
- H5 Freight reclassification: alive as a component, 64 bps; refuted as "most" by E3, E5, E6. Not a real decline; a presentation issue for the board.
- H2 Discount creep: contributor, 23 bps, concentrated in Regions 2 and 5; refuted as primary by E4, E11.
- H4 Product mix: refuted by E7. Zero contribution.
- Unexplained: 9 to 12 bps, within the 20 bps accuracy requirement, plausibly e-commerce small-order freight.
Surviving hypothesis: "Of the reported 230 bps decline, 64 bps is the freight reclassification and 166 bps is real. Of the real decline, 79 bps is customer mix, 63 bps is pass-through lag on list-priced core accounts caused by the price-update cycle lengthening from 21 to 74 days, and 23 bps is discount creep in two regions. About 85 bps (pass-through and discount) is reversible within twelve months; the mix effect is a strategy question."
Confidence: moderate-high, roughly 75%. Reasons: three linchpin rows, all from audited or high-reliability sources and all confirmed by independent checks; residual inside tolerance; the one soft element is that E5 extrapolates from 55% of COGS, and the vendors 21 to 40 sample supports the extrapolation.
What would overturn it: a tag audit showing more than 60% mis-tagging (implausible after the 3.4% finding); quote evidence that realised prices rose faster than the price book (contradicted by the sample); or failure of the tripwire below.
Tripwires (owner, date, threshold):
- Core-account margin must recover at least 40 bps within two quarters of the price-update cycle reaching 30 days or less. Owner: CFO. Review: Q4 Y3. If not met, H3 is overstated and the residual analysis reopens.
- Discount off list in Regions 2 and 5, monthly. Owner: head of sales. If either exceeds 9.0%, H2 is growing and must be re-ranked.
- National-account share. Owner: CEO. If it passes 45%, the mix effect becomes the dominant question and goes to
strategic-choice. - Vendors 21 to 40 cost letters, quarterly. Owner: pricing lead. If the cost-to-list gap on those lines exceeds 3 points, H3's size is understated.
Hand-off: the pricing-cycle fix (restaff the pricing team, automate cost-letter ingestion, 30-day update target) and the discount-authority tightening in Regions 2 and 5 are now a recommendation and a plan. They go to executive-communication for the board storyline and, before implementation, to a pre-mortem for the pre-mortem: that skill, not this one, will list how the pricing-cycle fix could fail (customer reaction to faster increases, vendor letter formats, branch resistance), set kill criteria and design the early-warning plan. The national-account mix question goes to strategic-choice. Nothing in this report addresses any of those.
What made this run work
The killing observation for the day-one answer was written on day one, so when the core-account margin came back down 160 bps there was no argument about whether it mattered. The artefact hypothesis was forced in, and it turned out to be worth 64 bps that would otherwise have been attributed to a real cause. Rows were rated on magnitude, which is why H2 was demoted despite pointing in the right direction. The plan was seven analyses with thresholds, not a tour of the business, and it finished in nine days with three days to spare for the sensitivity checks that make the confidence statement honest.