River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Hypothesis Driven Consulting Workplan

Three sheets and five documents that price every hypothesis in analyst-days, check the data behind it, and fix the number that kills it first.

Free download  ·  No account needed

Analysis Plan  ·  Marchwood Distribution  ·  margin down 4.2 points

The budget is a residual, not the headcount

Two analysts full time, manager at 40 percent, six weeks72 person-days
less interviews, scheduling and write-up−14
less synthesis and document production−11
less steering meetings and client management−6
Analyst-days available to test anything41

22 leaves cost 105 days to test in full

RefHypothesisPtsFullDataThreshold, declared firstBought
H-01Discount authority used beyond policy1.34NowUnder 15 of the 40 above the bandFull
H-06Competitors forced price concessions1.211NoneNone. No competitor prices existCannot test
H-08Supplier cost increases absorbed1.15NowTop 200 pass-through over 80 pctFull
H-04Rebate accruals understated0.663 daysYear-end variance under $250,000Proxy
H-12Import currency exposure unhedged0.46BuildNone. Needs a rebuilt ledgerDropped

105 analyst-days of tests against a budget of 41, a 2.6 times overrun. Five of the 22 leaves shown, and candidate sizes sum to 13.5 points against the 4.2 actually lost. Nine bought in full at 33 days, five proxied at 6, six dropped, two untestable.

Search this query and every result is the same artifact. An overriding question, a MECE issue tree, falsifiable hypotheses at the leaves, then a table mapping each one to the analyses that would test it, the data sources, an owner and a date, usually drawn as a Gantt. The best of them say outright that a workplan should detail the analyses, end products, sources, and the timing and responsibility per hypothesis. Not one of them prices the analysis.

Three columns are missing from all of them. What the full test costs in analyst-days, because a plan with owners and dates and no day count cannot decline a branch. Whether the data exists, because every template has a source column and treats it as an aspiration. And the number that kills the hypothesis, declared before anybody looks: the ICMJE requires the whole registration data set public before the first participant is enrolled precisely to stop outcomes being reframed later.

Marchwood Distribution, a $186m industrial distributor, lost 4.2 points of gross margin over seven quarters. Six weeks and three people is 72 person-days gross and 41 analyst-days once interviews, synthesis and steering come out. The 22 leaves cost 105 days to test in full, so nine were bought, five proxied, six dropped with written reasons and two recorded as untestable. Run the interviews first with the stakeholder interview programme pack, or read the client's own material with the document review and evidence log.

22 hypotheses, 41 days, and the 0.3 points nothing explains

The Hypothesis Register, the Workplan, the Data Requirements Log, and What the Tests Found.

Hypothesis Register

The columns the category leaves blank: full test cost, the threshold written down before anybody looked, and the verdict against it.

RefPtsFullBoughtThresholdResultVerdict
H-011.34415 of 40 above the band23 of 40Confirmed 1.1
H-020.933List-to-cost gap moved 1.5 pts3.1 ptsConfirmed 0.7
H-081.155Pass-through above 80 percent54 percentConfirmed 0.6
H-141.033Category share grew 3 pts5.8 ptsConfirmed 0.4
H-030.75210 of 60 carry an escalator4 of 60Killed day 7
H-040.666Accrual variance above $250,000$61,000Killed day 19
H-100.431Billed freight share moved 5 pts1 ptKilled day 8
H-090.6443 agreements within 5 pct of a tierNo splitAmbiguous, d22

Eight of the eleven resolved rows are shown, and three of the four kills were things somebody senior believed. A kill is worth as much as a confirmation and costs less, and it only reads as a kill because the number was fixed before the query ran.

Workplan

Three workstreams by branch, one owner each. The full column is what the leaves would cost tested in full; bought is what the budget allowed.

RefWorkstreamOwnerLeaves boughtFullBoughtUsedWindow
W-1Price realisationAnalyst AH-01, H-02, H-03, H-04, H-05, H-07241920Weeks 1 to 4
W-2Cost of goodsAnalyst BH-08, H-09, H-10, H-11141411Weeks 1 to 5
W-3Mix and channelManagerH-14, H-15, H-16, H-201268Weeks 2 to 5
Total503939

The reallocation, which is why two rows used more than they bought

DayWhat diedBudgetedUsedReturned
5H-11 write-downs into cost of sales211
7H-03 escalators never invoked523
8H-10 inbound freight shifted312
 Handed back to the pool6

Six days back, and six days is exactly what the two promotions cost. H-04 went from a 2 day proxy to a 6 day full test and killed cleanly at $61,000 against a $250,000 threshold. H-16 went from 1 day to 3 and confirmed a 0.2 point cause a proxy could have flagged but not reported. Neither was affordable at plan time.

Data Requirements Log

Twelve requirements serving 22 hypotheses. Only the first two states can be scheduled.

RefWhat is neededWhere it livesReadinessLeadWho releases it
D-01Transaction sales against listCRMAvailable now0Commercial Director
D-07Cost change notices, top 200PurchasingAvailable now0Head of Purchasing
D-03The 60 largest contractsContract storeOn request4Legal
D-04Rebate accrual papersGroup FinanceOn request3Financial Controller
D-08Rebate tier schedulesThree buyersOn request6Head of Purchasing
D-10Ledger by settlement currencyNowhere yetMust be built9Finance Systems
D-06Competitor transaction pricesNo sourceDoes not existNobody
D-11Channel flag on the orderNot capturedDoes not existNobody

D-08 decided the running order of a whole workstream. The tier schedules sit with three separate buyers rather than in a contract store, and six working days came from the buyers themselves rather than their director, so the test it serves was requested on day one and run last.

By state: available now 9 hypotheses and 6.2 candidate points, on request 6 and 3.5, must be constructed 5 and 1.7, does not exist 2 and 2.1.

What the Tests Found

The residual arithmetic, written as arithmetic. 39 of 41 days spent, 11 of 22 hypotheses resolved.

Quantity in dispute4.2 points$7,812,000
Attributed to six confirmed causes3.3 points$6,138,000
Residual0.9 points 
  of which H-09 could account forup to 0.6 
  of which nothing on this tree explainsat least 0.3 

What was never bought, kept apart from what could not be bought

StateCountDays avoidedCandidate points
Dropped at plan, reason on the row6352.1
Untestable, no record exists at any price2202.1

22 analyst-days produced the 3.3 attributed points, 6.7 days per point. The six drops would have cost 35 days to chase 2.1 candidate points, 16.7 days per candidate point. That comparison defends every no on the sheet, and it can only be made because both sides were priced.

The 0.3 points nothing explains stayed on the page. Spreading it across the six proven causes would have made the document total and destroyed the one property that makes those six defensible in the room, which is that each is a measured number tied to a named population.

What's in the pack

01

Hypothesis Register

One row per leaf carrying its candidate size, full test cost in analyst-days, data state, declared threshold, what was bought, and the verdict against the number.

02

Workplan

One row per workstream with an owner, showing the full cost of what it holds against what the budget bought and what it actually used after reallocation.

03

Data Requirements Log

One row per requirement with its readiness state, its lead time in working days and the person who has to release it. The six day item reorders a whole workstream.

04

Hypothesis Tree

The question, three branches that exhaust the arithmetic, and every leaf sized before it is priced, including the two struck out with the reason they failed.

05

Analysis Plan

What was bought, what was declined and in what order it runs, with the sequencing note per workstream and the reallocation rule agreed before the work starts.

06

What the Tests Found

Confirmed, killed, ambiguous, never bought and could not be bought, kept in five sections, closing on the residual arithmetic rather than on a total.

07

How the Test Budget Is Allocated

The full worked example: 72 person-days down to 41, 105 days of tests refused to 39, and the six returned days traced to the two tests they bought.

08

A hypothesis you cannot afford to test is not a hypothesis

The standing rule. Four fields on every row, four states per leaf, and three of the four are a no. Turn the answer into a report with the findings pack.

How to use it

  1. 1

    Open in River, or take it blank

    Open the pack in River and send the question with your intake material and the team, or download the Word documents and CSV sheets and work through them yourself.

  2. 2

    Compute the budget before the tree

    Gross capacity minus interviews, synthesis, document production and steering. That residual is the only day count you can spend on testing, and it is usually a third smaller than the team looks.

  3. 3

    Price every leaf, then check its data

    Cost each test to a defensible verdict rather than to a first look, then classify its data as available now, available on request with a lead time, needs building, or does not exist.

  4. 4

    Declare the number, then buy

    Write the kill threshold and the resolve-by day on every row before any query runs. PCAOB AS 2105 requires materiality to be expressed as a specified amount during planning, not after the evidence.

Frequently asked questions

Is this template free?

Yes. The zip is Word documents and CSV sheets, no account and no card. Edit with AI is the optional half: the agent builds the tree, prices every test and puts a threshold on each row. The rest sit in the template library.

What format are the downloaded files?

Five Word documents and three CSV sheets in one zip, no conversion needed. The sheets carry the columns that do the work: full test days, data readiness, threshold declared before testing, and days used. Open them in Excel, Numbers or Google Sheets.

What does Edit with AI actually do?

It reads the engagement question and your intake material, then computes the testing budget as a residual. From there it builds and sizes the tree, prices every leaf, and checks each data source into one of four readiness states. The allocation comes back stated as what is not being bought.

How do I price a test I have never run?

Estimate days to a defensible verdict, which means the extract, the reconciliation to something the client recognises, the analysis and the write-up. Pricing the first day of an analysis as the analysis is how a six day test gets scheduled as a two day one and takes six anyway.

Can a kill threshold be changed once it is set?

Yes, and the revision is logged with the old number, the new one, the date and the reason. AS 2105 requires the auditor to reevaluate materiality as the audit progresses. What is not allowed is a silent revision after the result is known.

What if the data for my biggest hypothesis does not exist?

Then it is a different kind of item, not a low priority. Both untestable hypotheses in the worked example ranked in the top six by size, so they ship in the executive summary as a stated limit. Evidence has to be relevant and reliable to support anything.

Does this replace the workplan I report against?

No. This decides what gets bought and what the budget refuses. Tracking burn against a fixed fee once delivery starts is the workplan and resource tracking pack, and building the argument afterwards is the storyline tool.

Price the tree before you prune it

Take the Word documents and CSV sheets blank, or send River the question and the day count and get back the allocation as a list of what is not being bought.

Edit with AI