Innovation Atlas · Methods · generated 2026-08-20 · do not edit by hand

Frameworks carry their evidence, or say they have none.

One record per framework: what it is for, the source that defines it, and the honest strength of its support. The methods are not ranked against each other — a mapping tool, an experimentation cycle and an explanatory theory share no axis, and pretending otherwise would assert what the evidence cannot carry (ADR-0008). 2 strong · 5 weak · 2 none — and none is an acceptable answer; it only bars the method from being cited as a reason.

MTH-0004 · S5 Test

Hypothesis-driven experimentation (the scientific approach)

support: strong

Defined in: SRC-0025 D The Lean Startup; SRC-0007 D Testing Business Ideas

Supported by: SRC-0031 A The scientific-approach RCTs (Camuffo, Cordova, Gambardella & Spina and replications); SRC-0029 B Experimentation Works: The Surprising Power of Business Experiments

What it is

Treat the idea as a theory: make assumptions explicit, rank them by risk, test the riskiest with a deliberate experiment carrying a pre-declared decision rule, and act on the result — including termination. The common core that Lean Startup, Testing Business Ideas and corporate A/B programmes all operationalise.

What supports it

The best evidence in the methods layer. Five RCTs (SRC-0031: the 2020 Management Science trial plus the four-trial 2024 replication, 759 firms total) show causal effects on termination of weak ideas and decisive pivoting. SRC-0029 grounds the organisational version: firms running experiments at industrial scale with authority to act on results. This is the one method the atlas can call supported without blushing.

Known limits

The RCT outcomes are early-stage — termination, pivots, early revenue — not long-run survival or returns. And the discipline is socially expensive: a pre-declared kill threshold binds the sponsor too, which is exactly why organisations quietly drop it.

Where it is used

The S5 gate itself: no build budget without a named hypothesis, a named experiment and a kill threshold declared before data arrives. This record is the citation for that rule.

MTH-0008 · S7 Govern

Organizational ambidexterity (explore/exploit separation)

support: strong

Defined in: SRC-0009 B Lead and Disrupt

Supported by: SRC-0009 B Lead and Disrupt

What it is

Run exploitation and exploration as structurally separate units — different metrics, cadences and cultures — integrated only at the senior-team level, so the core business cannot smother experiments and experiments cannot destabilise the core. O'Reilly and Tushman's three disciplines: ideation, incubation, scaling.

What supports it

SRC-0009, rated B: systematic case evidence across firms and decades that structural separation with senior-team integration beats both full integration and full spin-off. Grounded, not replicated — B-grade support is the ceiling here, and per the ratings it carries a design choice, not a gate.

Known limits

Survivor-heavy case base, and the method's cost scales with organisation size — a separate exploratory unit below a certain firm size is a courtesy title for one distracted person. Says little about when to fold exploration back in.

Where it is used

The S7 argument for keeping exploratory budget and metrics out of the line business's quarterly logic. Cited when designing mandates, not when judging any single concept.

MTH-0001 · S2 Frame

Opportunity scoring (refined ODI)

support: weak

Defined in: SRC-0011 D Jobs to Be Done: Theory to Practice

Supported by: SRC-0011 D Jobs to Be Done: Theory to Practice

What it is

Opportunity = importance + max(importance − satisfaction, 0), normalised to 0–10 for use alongside other scoring dimensions. Importance and satisfaction are rated by the customer, not by us.

What supports it

SRC-0011 (Ulwick), rated D. The formula's defence is arithmetic rather than empirical: it double-weights importance and, unlike plain gap analysis, does not penalise needs that are already well served. That reasoning stands on its own. The published success rates attached to the method are vendor-reported and unreplicated, and are not used here as justification.

Known limits

- Assumes the customer can articulate a stable desired outcome. Many of ours cannot. - Silent on needs created by regulation rather than by the customer, which is a large share of our portfolio. - Produces a number, and numbers survive contexts their assumptions do not. Carry the rating with the score.

Where it is used

Concept intake and the S2 framing worksheet.

MTH-0003 · S5 Test

Build–Measure–Learn (Lean Startup loop)

support: weak

Defined in: SRC-0025 D The Lean Startup

Derived from: SRC-0006 D The Four Steps to the Epiphany

Supported by: SRC-0031 A The scientific-approach RCTs (Camuffo, Cordova, Gambardella & Spina and replications)

What it is

The loop: state the riskiest assumption, build the smallest artefact that tests it, measure against a pre-declared threshold, then pivot or persevere. Descends directly from Blank's customer development (SRC-0006) with the test-fast vocabulary that made it a movement.

What supports it

The RCT line (SRC-0031) supports the loop's scientific core — hypotheses, tests, decision rules — with replicated causal evidence. It does not test the brand as practised: innovation accounting, engines of growth and the rest of the book's apparatus remain unevidenced, which is why the strength here is weak rather than strong. The strong record is MTH-0004.

Known limits

Assumes building is cheap, shipping is reversible and failure is private. Hardware, clinical and grid-connected work violate all three; there the loop's unit is a simulation or a paper trial, not a shipped MVP.

Where it is used

The team's working vocabulary for S5 cadence. Any gate paper citing "lean" must cite SRC-0031 for the principle and name its kill threshold, per the standing experiment rule.

MTH-0006 · S2 Frame

Jobs to be done

support: weak

Defined in: SRC-0011 D Jobs to Be Done: Theory to Practice; SRC-0028 D Competing Against Luck: The Story of Innovation and Customer Choice

Supported by: SRC-0011 D Jobs to Be Done: Theory to Practice

What it is

Frame demand as the progress a customer is trying to make in a circumstance — the job — and specify desired outcomes for it, rather than segmenting by who the customer is. Two dialects: Ulwick's outcome-driven machinery (SRC-0011) and Christensen's narrative theory (SRC-0028).

What supports it

D-rated practitioner sources on both dialects — consulting portfolios and the milkshake anecdote, with Ulwick's success rates vendor-reported and unreplicated. The lens reliably produces better interview questions than demographic segmentation does; that observation is craft, and craft is what weak-via-D-sources means here.

Known limits

Assumes an articulable, stable job. Regulation-created needs — a large share of infrastructure work — have no customer who can voice the job, and multi-actor B2B purchases have several conflicting ones.

Where it is used

S2 framing interviews and the outcome statements that MTH-0001 then scores. The jobs lens shapes questions; it never appears in a gate paper as a finding.

MTH-0007 · S1 Scan · S7 Govern

Open innovation (boundary-crossing search and licensing)

support: weak

Defined in: SRC-0027 C Open Innovation: The New Imperative for Creating and Profiting from Technology

Supported by: SRC-0027 C Open Innovation: The New Imperative for Creating and Profiting from Technology

What it is

Deliberately sourcing ideas, technology and routes to market across the firm boundary — scouting, licensing in and out, partnerships, spin-outs — instead of defaulting to internal R&D for everything.

What supports it

Chesbrough's argued case studies (SRC-0027, rated C): coherent framing, no tested proposition. Von Hippel's user-innovation evidence (SRC-0003) supports one specific channel — users as originators — better than the general open-beats-closed claim is supported anywhere.

Known limits

Appropriability: openness leaks exactly as easily as it absorbs, and the framework offers no test for when the licence out is the business walking out. The academic literature after 2003 multiplied definitions faster than evidence.

Where it is used

S1 scouting posture and the make-buy-license question at the S7 portfolio gate, as framing and checklist. Any claim that an open arrangement will raise returns needs its own evidence, case by case.

MTH-0009 · S5 Test

Experiment selection from a catalogue

support: weak

Defined in: SRC-0007 D Testing Business Ideas

Supported by: SRC-0007 D Testing Business Ideas

Adoption: 1M+ practitioners, vendor-reported (Strategyzer), unaudited

What it is

Pick the test from a catalogued library of 40+ experiment types, matched to the assumption at risk and ranked by evidence strength, cost and setup time — instead of inventing an experiment ad hoc or defaulting to a survey.

What supports it

SRC-0007, rated D: the catalogue is craft with no outcome data of its own. Its premise — that deliberate experiments with decision rules beat intuition — is the part SRC-0031 evidences; the specific catalogue and its evidence-strength rankings are the authors' judgment.

Known limits

Built for digital and service propositions; the catalogue thins out exactly where our work gets hard (hardware, certification, regulated infrastructure). Evidence-strength rankings inside the book are asserted, not measured.

Where it is used

The S5 worksheet: every riskiest assumption is paired with one named experiment from the catalogue and a pre-declared kill threshold. The pairing discipline is ours; the menu is the book's.

MTH-0002 · S3 Concept

Business Model Canvas

support: none

Defined in: SRC-0026 D Business Model Generation

Adoption: 6M+ practitioners, vendor-reported (Strategyzer), unaudited

What it is

One page, nine blocks: customer segments, value propositions, channels, customer relationships, revenue streams, key resources, key activities, key partnerships, cost structure. A shared drawing convention for what the business is supposed to be.

What supports it

Nothing that clears the bar. Uptake is enormous but vendor-counted; the one study associating heavy Canvas use with accelerator pitch results (271 teams) was exploratory and measured competition performance. The honest statement: this is the most widely spoken business-model notation, and notation is what it is — support_strength none is not a criticism of a map for being a map.

Known limits

A canvas records beliefs; it cannot test one, rank one, or notice it is wrong. Nine tidy boxes also flatter tidy businesses — platform, regulated and multi-sided models fight the template.

Where it is used

Every concept entering S3 gets drawn as a canvas before it may request experiment budget. The drawing feeds the assumption list that SRC-0007-style experiments then attack.

MTH-0005 · S2 Frame · S3 Concept · S5 Test

Design thinking (d.school five modes / IDEO HCD)

support: none

Defined in: SRC-0032 D Design Thinking Bootleg (Stanford d.school); SRC-0033 D The Field Guide to Human-Centered Design (IDEO.org)

What it is

A structured discovery process — empathize with users, define the problem, ideate widely, prototype cheaply, test early — taught as five modes (Stanford) or three phases (IDEO). In practice: a facilitation repertoire for getting a team out of the building and out of its first idea.

What supports it

Teaching practice and two decades of consulting portfolios; no outcome data that clears the bar. The craft value of specific methods (structured interviews, rapid prototyping) is real and partly borrowed from evidenced traditions, but "design thinking" as a package has never been shown to cause better innovation outcomes. Strength none, used anyway, per the methods rule.

Known limits

Front-loaded: everything before viability. The process manufactures desirability insight and says nothing about who pays or what a test should kill. Workshop theatre is a documented failure mode — five modes performed in an afternoon, nothing decided.

Where it is used

S2 discovery work, using the kits' method cards under our own evidence rules. It feeds hypotheses to MTH-0004; it never replaces it.