Generate novel solutions by combining dimensions of a problem in unusual ways

How does Zwicky Box work?

Zwicky Box provides a step-by-step process for evaluating your options and making a structured decision. Follow the guided steps above to work through the tool with your specific situation.

When should you use Zwicky Box?

Use Zwicky Box when you face a decision that would benefit from structured analysis rather than gut instinct. It is particularly valuable for high-stakes or irreversible decisions where the cost of being wrong is significant.

Zwicky Box: How to Use It (Decision…

Zwicky Box: How to Use It (Decision… | Faster Than Normal

Use this when you need to generate a large, structured field of possible solutions — not by brainstorming freely, but by defining the dimensions of a problem and then systematically combining values across those dimensions. The Zwicky Box forces combinatorial creativity: it surfaces options that no individual brainstorm would produce because the human mind doesn't naturally traverse a multi-dimensional possibility space.

Section 1

What This Tool Does

Brainstorming has a dirty secret. It feels generative — sticky notes accumulating, energy rising, ideas bouncing — but the output is almost always clustered in a narrow band of the solution space. People anchor on the first idea voiced. They riff on each other's suggestions, which produces variations on a theme rather than genuinely different concepts. And they unconsciously stay within the boundaries of what they've already seen work, because the brain's pattern-matching machinery is optimised for retrieval, not invention. A 2003 study by Mullen, Johnson, and Salas found that individuals working alone consistently generated more ideas — and more diverse ideas — than equivalent brainstorming groups. The problem isn't effort. It's architecture. Unstructured ideation lacks a mechanism for forcing the mind into unfamiliar territory.

Fritz Zwicky, the Bulgarian-born Swiss astrophysicist working at Caltech in the 1940s, understood this problem from a radically different angle. Zwicky wasn't solving business problems. He was trying to enumerate every possible type of jet propulsion system — a design space so vast that intuition was useless. His solution was what he called "morphological analysis": decompose the problem into its independent dimensions (he called them "parameters"), list the possible values each dimension could take, and then systematically examine the combinations. The result is a multi-dimensional matrix — a box, in the geometric sense — where each cell represents a unique configuration of choices across all dimensions. For jet propulsion, the dimensions included things like thrust medium (air, water, particles), energy source (chemical, nuclear, solar), and motion type (translational, rotational, oscillating). The combinations numbered in the thousands. Most were absurd. Some were known. And a handful were genuinely novel configurations that no expert had considered, because no expert naturally thinks in six-dimensional combinatorial space.

The tool migrated from astrophysics to engineering, then to product design, then to strategy. The Swedish Morphological Society, founded in the 1990s, formalised its application to policy analysis and defence planning. But the core mechanism hasn't changed since Zwicky's original formulation. You decompose a problem into its fundamental dimensions, enumerate the options within each dimension, and then traverse the resulting matrix to find combinations that are novel, feasible, and worth pursuing. The cognitive shift is from "What's a good idea?" — which invites pattern-matching — to "What are all the possible configurations?" — which forces systematic exploration.

The power, and the danger, is in the combinatorics. A box with five dimensions and four values per dimension produces 4^5 = 1,024 unique combinations. Six dimensions with five values each: 15,625. Most of these combinations are nonsensical or impractical. That's the point. The Zwicky Box doesn't promise that every cell is viable. It promises that the viable cells you'd never have imagined are sitting there, waiting to be noticed, in a region of the solution space your intuition would never have visited on its own.

Section 2

How to Use It — Step by Step

Instructions on the left. Worked example — "How might a mid-market B2B SaaS company redesign its onboarding experience to reduce time-to-value from 45 days to under 14?" — on the right.

Step 1 — Decompose

Identify the independent dimensions of the problem

This is the step that determines everything. You're asking: what are the fundamental axes along which a solution to this problem can vary? Each dimension should be genuinely independent — changing the value on one dimension shouldn't force a change on another. Aim for 4–6 dimensions. Fewer than four and the box is too simple to surface anything surprising. More than seven and the combinatorial explosion becomes unmanageable. The art is in choosing dimensions that are specific enough to be actionable but abstract enough to allow genuinely different values. "Colour of the button" is too granular. "Business model" is too broad. "Primary learning modality" or "who does the configuration work" — that's the right altitude.

Worked example

Onboarding redesign dimensions

The team identifies five independent dimensions of the onboarding experience:

D1 — Learning format: How does the customer learn the product? D2 — Configuration agent: Who does the setup work? D3 — Success metric trigger: What event signals "onboarded"? D4 — Pacing model: What controls the speed of progression? D5 — Support channel: How does the customer get help when stuck?

Step 2 — Enumerate

List 3–6 possible values for each dimension

For each dimension, brainstorm the distinct options. These should be genuinely different approaches, not minor variations. Include at least one value that feels unconventional or counterintuitive — that's where the novel combinations will come from. Don't filter for feasibility yet. If "the customer does zero configuration" seems impractical, include it anyway. The box's job is to generate; filtering comes later. Write the dimensions as rows and the values as columns (or vice versa) to create the visible matrix.

Worked example

Populating the values

D1 — Learning format: (a) Self-paced video library, (b) Live cohort-based workshop, (c) Interactive in-app guided tasks, (d) 1:1 concierge walkthrough

D2 — Configuration agent: (a) Customer self-serve, (b) Vendor implementation team, (c) AI-assisted auto-config from existing data, (d) Certified third-party partner

D3 — Success metric trigger: (a) First workflow completed, (b) Integration with existing tool live, (c) Team adoption threshold (e.g., 5+ active users), (d) Customer-defined milestone

D4 — Pacing model: (a) Fixed schedule (Day 1, Day 3, Day 7…), (b) Behaviour-triggered (next step unlocks on completion), (c) Customer-chosen pace, (d) Deadline-driven (go-live date works backward)

D5 — Support channel: (a) Email/ticket, (b) Dedicated Slack channel, (c) AI chatbot with escalation, (d) Embedded community forum with peer answers

Step 3 — Combine

Traverse the matrix to generate specific configurations

Now you walk through the box. Each combination — one value selected from each dimension — defines a distinct solution concept. You don't need to examine all 1,024 combinations (in a 4×4×4×4×4 box). Use three strategies: (1) Random traversal — pick values at random from each dimension and see what emerges. Do this 10–15 times. (2) Anchor-and-vary — fix one dimension at an unusual value and systematically vary the others. (3) Constraint-driven — impose a real-world constraint ("must cost under $X per customer") and find all combinations that satisfy it. Write each combination as a one-sentence concept description.

Worked example

Generating novel configurations

Combination A: Live cohort workshop + AI auto-config + team adoption threshold + deadline-driven pacing + Slack channel. "Customers join a weekly cohort. AI pre-configures the instance from their CRM export. The cohort works toward a shared go-live date. Support happens in a dedicated Slack channel where cohort members also help each other."

Combination B: In-app guided tasks + customer self-serve + first workflow completed + behaviour-triggered pacing + AI chatbot. "Pure product-led onboarding. No human touch. The app guides users through setup step by step, unlocking the next task only when the previous one is done. AI chatbot handles questions."

Combination C: 1:1 concierge + AI auto-config + customer-defined milestone + customer-chosen pace + embedded community. "White-glove start with AI doing the heavy lifting on data migration. The customer defines what 'onboarded' means for them. They control the pace. Long-term support shifts to a community forum."

Combination A is the one nobody in the room had considered before the box forced it. The cohort model combined with AI auto-configuration and deadline-driven pacing is a genuinely different architecture than anything the team had discussed in six months of onboarding improvement work.

Step 4 — Filter

Evaluate combinations against feasibility and strategic fit

Most combinations will be impractical, redundant, or internally contradictory. That's expected. Apply three filters sequentially: (1) Internal consistency — does this combination make logical sense? A "self-serve configuration" paired with "1:1 concierge walkthrough" is contradictory. Discard. (2) Feasibility — can you build this within your resource and time constraints? (3) Strategic differentiation — does this combination create something meaningfully different from what exists in the market or what you're doing today? You should end up with 3–8 viable concepts from an initial field of dozens.

Worked example

Filtering the onboarding concepts

From 15 generated combinations, the team discards 7 for internal contradictions (e.g., "fixed schedule" paired with "customer-chosen pace" is incoherent) and 4 for feasibility constraints (the company lacks the engineering capacity to build AI auto-config in Q1). Four concepts survive:

Combination A (cohort + AI config + deadline-driven) — novel, feasible by Q2, strong differentiation
Combination B (pure product-led) — feasible now, but similar to competitors
Combination C (concierge + AI + community) — novel for enterprise tier, feasible with existing CS team
A fourth hybrid that emerged during discussion

The team selects Combination A for prototyping, with Combination C as the enterprise variant.

Step 5 — Develop

Flesh out the top concepts into testable prototypes

Take your 2–3 surviving combinations and develop each into a concrete concept with enough detail to evaluate or test. Define the user journey, the resource requirements, the key assumptions, and the fastest way to validate those assumptions. The Zwicky Box got you to a novel configuration. Now you need to determine whether that configuration actually works — and that requires prototyping, customer interviews, or a small-scale pilot, not more matrix analysis.

Worked example

Developing Combination A

Concept name: Cohort Onboarding with AI Pre-Config

User journey: Customer signs up → uploads CRM export or connects API → AI auto-configures instance (workflows, fields, integrations) within 24 hours → customer is enrolled in next weekly cohort (max 8 companies) → 3 live sessions over 10 days, each 90 minutes → deadline: go-live on Day 12 → dedicated Slack channel stays active for 30 days post-launch.

Key assumptions to test: (1) AI can achieve 80%+ configuration accuracy from a CRM export. (2) Customers will commit to a fixed cohort schedule. (3) Peer dynamics in the cohort accelerate learning. (4) 12-day timeline is achievable.

Fastest validation: Run a manual "Wizard of Oz" cohort — a human does the configuration work that AI would eventually do — with 3 pilot customers. Measure time-to-value and satisfaction against the current 45-day baseline.

Section 3

When It Works Best

✓

Ideal Conditions for the Zwicky Box

Dimension	Best fit
Problem type	Design problems with multiple independent variables — product configuration, service design, go-to-market strategy, business model innovation. The tool shines when the solution space is large but structured: you know the dimensions along which solutions can vary, but you haven't explored the full combinatorial space.
Creative stage	Divergent ideation — when you need volume and variety, not convergence. Deploy the Zwicky Box after you've framed the problem but before you've committed to any particular solution direction. It's a generation tool, not an evaluation tool.
Team knowledge	Teams with deep domain expertise who are stuck in familiar patterns. The box is most valuable precisely when the team knows the problem well enough to define its dimensions but has been circling the same three solution concepts for months. Expertise enables good dimension selection; the matrix breaks the expertise trap.
Competitive context	Markets where existing solutions cluster around similar configurations. If every competitor's onboarding looks the same, the Zwicky Box helps you find the configurations nobody has tried — not because they're bad, but because nobody thought to combine those particular values.

Section 4

When It Breaks Down

⚠

Failure Modes

Failure pattern	What goes wrong	What to use instead
Wrong dimensions chosen	The entire output of the box depends on which dimensions you select. Choose dimensions that are too surface-level ("colour," "name," "price point") and you get trivial variations. Choose dimensions that aren't truly independent and you get contradictory combinations. The box amplifies the quality of your decomposition — garbage dimensions in, garbage combinations out.	Abstraction Laddering to find the right level of dimension specificity; First Principles Thinking to identify the truly fundamental variables
Combinatorial overwhelm	Six dimensions with six values each produces 46,656 combinations. Teams stare at the matrix, feel paralysed, and either abandon the exercise or cherry-pick combinations that confirm their pre-existing preferences — defeating the entire purpose. The box becomes a prop for confirmation bias rather than a tool for exploration.	Limit to 4–5 dimensions with 3–4 values each. Use structured traversal strategies (random sampling, anchor-and-vary) rather than attempting exhaustive review.
Dimension dependencies ignored	The box assumes dimensions are independent. In practice, they rarely are. "AI auto-configuration" as a configuration agent may require "behaviour-triggered pacing" because the AI needs user actions to calibrate. When dependencies exist between dimensions, many cells in the matrix are impossible — but the box doesn't flag them. Teams waste time evaluating combinations that can't exist.

The most dangerous failure mode is wrong dimensions chosen — and it's dangerous precisely because it's invisible. When the box is built on shallow or poorly chosen dimensions, the output still looks impressive. You still get a matrix. You still get novel combinations. The process still feels rigorous. But the combinations are variations on the wrong variables, like rearranging deck chairs with great systematic thoroughness. The protection is simple but requires humility: spend as much time debating the dimensions as you spend generating combinations. In practice, dimension selection should consume 40–50% of the total session time. If your team agreed on the dimensions in five minutes, you almost certainly chose the obvious ones — and the obvious dimensions produce obvious combinations.

A second layer of protection: after selecting your dimensions, ask one question of each — "If we changed the value on this dimension while holding everything else constant, would the solution feel meaningfully different?" If the answer is no, the dimension isn't doing enough work. Replace it.

Section 5

Visual Explanation

Section 6

Pairs With

The Zwicky Box is a generation engine. It produces candidate solutions. What happens upstream — problem framing — and downstream — evaluation and selection — determines whether those candidates are worth anything.

Use before

Abstraction Laddering

The hardest part of the Zwicky Box is choosing the right dimensions. Abstraction Laddering helps you move up ("Why does this dimension matter?") and down ("What specifically varies within this dimension?") until you find the level of abstraction where the dimensions are genuinely independent and meaningfully different. Dimensions chosen at the wrong altitude produce either trivial or unmanageable boxes.

Use before

First Principles Thinking

First Principles decomposition strips a problem to its fundamental truths — the physics of the situation. These fundamentals often map directly to the dimensions of a Zwicky Box. If you've identified that onboarding is fundamentally about learning, configuration, and validation, those become your dimensions. First Principles gives you the skeleton; the box gives you the combinatorial flesh.

Use before

Reframing

A Zwicky Box built on a poorly framed problem produces a thousand solutions to the wrong question. Reframing challenges the problem statement itself — "Are we really trying to reduce time-to-value, or are we trying to reduce perceived effort?" — before you commit to dimensions. Different frames yield entirely different boxes.

Use after

Decision Matrix

Section 7

Real-World Application

IDEO and Shimano — redesigning the American bicycle market

The scenario

In the mid-2000s, Shimano — the Japanese components giant that supplies gears, brakes, and drivetrains to most of the world's bicycle manufacturers — faced a puzzling stagnation. The US cycling market was flat. High-end road and mountain bike sales were healthy, but the vast majority of American adults — an estimated 161 million who owned bikes but rarely rode them — had effectively stopped cycling. Shimano engaged IDEO, the design consultancy, to figure out why and to design a product that could re-engage lapsed riders. The brief wasn't "design a better derailleur." It was "design a new cycling experience."

How the tool applied

IDEO's team used morphological analysis as part of their structured ideation process. After extensive ethnographic research — riding with lapsed cyclists, interviewing bike shop owners, studying the purchase and abandonment journey — they decomposed the "casual cycling experience" into its fundamental dimensions. These included: gearing complexity (number of gears and how they're controlled), maintenance burden (what the rider must do to keep the bike functional), riding posture (aggressive vs. upright), purchase channel (specialty bike shop vs. mass retail), and aesthetic identity (sport/performance vs. lifestyle/transportation). For each dimension, they enumerated values ranging from the existing market standard to deliberately unconventional options.

The existing market clustered around a single configuration: many gears, exposed cables requiring regular adjustment, forward-leaning posture, specialty shop distribution, and sport-oriented aesthetics. The morphological analysis surfaced a combination that was the near-opposite on every dimension: internally geared hub (three speeds, zero external maintenance), fully enclosed cables, upright riding position, available through non-traditional channels, and a clean aesthetic that looked more like furniture than sporting equipment.

What it surfaced

The result was the Shimano Coasting line, launched in 2007 — a category of bicycles designed around automatic shifting (the hub shifted gears based on pedalling speed, requiring no rider input), coaster brakes (pedal backward to stop, eliminating hand-brake complexity), and a design language that deliberately rejected the performance-cycling aesthetic. Three bicycle manufacturers — Trek, Raleigh, and Giant — produced Coasting bikes using Shimano's new components.

Section 8

Analyst's Take

Faster Than Normal — Editorial View

The Zwicky Box endures because it solves a problem that gets worse with expertise. The more you know about a domain, the more efficiently your brain prunes the solution space — which is useful for execution and catastrophic for innovation. Experienced teams converge fast. They know what works. And that knowledge becomes a cage. The box is a mechanical override for expert convergence. It doesn't require creativity; it requires only the discipline to define dimensions honestly and the patience to traverse combinations that feel wrong. The "feel wrong" part is the signal, not the noise. If every combination in your box feels plausible, your dimensions are too conservative.

The failure I see most often: teams treat dimension selection as a five-minute warm-up exercise and then spend an hour on combinations. This is backwards. The dimensions are the tool. A box with brilliant dimensions and lazy combination-traversal will still produce novel concepts. A box with lazy dimensions and exhaustive combination-traversal will produce a thousand variations of the same mediocre idea. I've watched teams build Zwicky Boxes where every dimension was a feature attribute — screen size, battery life, price tier, colour — and wonder why the output felt incremental. The dimensions need to operate at the level of the user's experience or the system's architecture, not at the level of spec sheets. "How does the user discover the product?" is a dimension. "What colour is the packaging?" is not.

The highest-leverage modification: include one "wildcard" dimension that has nothing to do with your industry. Zwicky himself did this — he included dimensions in his propulsion matrices that came from unrelated physics domains. For a SaaS onboarding box, add a dimension like "emotional arc" (values: surprise → mastery, anxiety → relief, isolation → belonging, confusion → clarity). For a retail concept, add "temporal relationship" (one-time, recurring ritual, seasonal, on-demand). These alien dimensions force combinations that cross-pollinate from outside your category. The best Zwicky Box I've ever seen in practice had five conventional dimensions and one borrowed from theatrical stage design — "audience relationship" (passive observer, active participant, co-creator, critic). It produced three product concepts that the team's VP of Product called "the first genuinely new ideas in two years." The alien dimension was the catalyst. Without it, the box would have produced competent but predictable configurations.

Section 9

Top Resources

Discovery, Invention, Research Through the Morphological Approach — Fritz Zwicky (1969)

Primary source

Zwicky's own definitive account of morphological analysis, published near the end of his career. Dense, idiosyncratic, and occasionally combative — Zwicky had a famously difficult personality and used the book to settle scores with colleagues who dismissed his method. But the core chapters on constructing morphological boxes, selecting parameters, and conducting cross-consistency analysis remain the authoritative reference. Out of print but available through academic libraries and used book dealers.

Business Model Generation — Alexander Osterwalder & Yves Pigneur (2010)

Book

The Business Model Canvas is, structurally, a specialised Zwicky Box — nine dimensions of a business model with multiple possible values per dimension. Osterwalder and Pigneur don't cite Zwicky, but the method is morphological analysis applied to strategy. Read this for the best modern example of how decomposing a problem into dimensions and recombining values produces genuinely novel business configurations. The "patterns" chapter is particularly useful for seeing how different value combinations across the canvas produce fundamentally different business architectures.

Blue Ocean Strategy — W. Chan Kim & Renée Mauborgne (2005)

Book

The "Strategy Canvas" — plotting your offering against competitors across multiple value dimensions and then deliberately choosing a different configuration — is morphological thinking applied to competitive positioning. Kim and Mauborgne's "Eliminate-Reduce-Raise-Create" framework is essentially a structured way to traverse a Zwicky Box and select the combination that maximises differentiation. The Cirque du Soleil case study is the canonical example of what happens when you recombine dimensions that an entire industry assumed were fixed.

General Morphological Analysis: A General Method for Non-Quantified Modelling — Tom Ritchey (2011)

Book

Ritchey, who founded the Swedish Morphological Society, is the leading contemporary practitioner of Zwicky's method. This monograph covers the full modern toolkit: cross-consistency analysis (the systematic method for eliminating impossible combinations), software-assisted morphological analysis, and applications to defence planning, policy analysis, and scenario construction. The most rigorous treatment of the method's extensions and limitations available. Essential reading if you plan to use the box on problems with more than five dimensions.

Zero to One — Peter Thiel (2014)

Book

Thiel's central argument — that the most valuable companies create something genuinely new rather than iterating on existing solutions — is the strategic case for why the Zwicky Box matters. "What important truth do very few people agree with you on?" is a question that the box can help answer mechanically: find the combination of dimension values that no competitor has assembled, and ask whether it's unexplored because it's bad or because nobody thought to look there. Read alongside the Zwicky Box as the philosophical justification for systematic combinatorial exploration.

Zwicky Box

Continue exploring

What This Tool Does

How to Use It — Step by Step

Identify the independent dimensions of the problem

Onboarding redesign dimensions

List 3–6 possible values for each dimension

Populating the values

Traverse the matrix to generate specific configurations

Generating novel configurations

Evaluate combinations against feasibility and strategic fit

Filtering the onboarding concepts

Flesh out the top concepts into testable prototypes

Developing Combination A

When It Works Best

Ideal Conditions for the Zwicky Box

When It Breaks Down

Failure Modes

Visual Explanation

Pairs With

Real-World Application

IDEO and Shimano — redesigning the American bicycle market

Analyst's Take

Top Resources

This connects to...

Continue exploring

More like this, in your inbox

What This Tool Does

How to Use It — Step by Step

Identify the independent dimensions of the problem

Onboarding redesign dimensions

List 3–6 possible values for each dimension

Populating the values

Traverse the matrix to generate specific configurations

Generating novel configurations

Evaluate combinations against feasibility and strategic fit

Filtering the onboarding concepts

Flesh out the top concepts into testable prototypes

Developing Combination A

When It Works Best

Ideal Conditions for the Zwicky Box

When It Breaks Down

Failure Modes

Visual Explanation

Pairs With

Real-World Application

IDEO and Shimano — redesigning the American bicycle market

Analyst's Take

Top Resources

This connects to...