Analogical Reasoning
Solve new problems by drawing parallels to familiar ones — the model generates its own relevant examples from training data, finding structural similarities across domains without needing manually crafted demonstrations.
Introduced: Analogical Reasoning was introduced by Yasunaga et al. in 2023. The technique enables the model to self-generate relevant problems and solutions as analogies before solving new problems. Instead of manually providing few-shot examples, the model recalls similar problems from its training data — identifying structural parallels, extracting shared patterns, and transferring proven solution strategies to the novel challenge at hand.
Modern LLM Status: Self-generated reasoning is increasingly built into modern LLMs as part of their default reasoning capabilities. However, the explicit “recall similar problems first” instruction remains valuable for complex novel problems where you want the model to ground its approach in proven analogies from related domains. When facing genuinely unfamiliar territory, prompting the model to deliberately draw on analogous cases produces more creative, better-grounded solutions than generic reasoning alone.
Let the Model Find Its Own Examples
Analogical reasoning transforms the model into an expert who recalls similar cases before tackling a new one. Rather than you providing examples manually, the model draws on the vast library of problems and solutions encoded in its training data — finding structural matches that a human might never think to look for. A networking problem might find its best analogy in fluid dynamics; a management challenge might mirror an ecosystem’s balance.
The model becomes its own example curator. When presented with a novel problem, it searches its knowledge for structurally similar situations — problems that share the same underlying dynamics even if the surface details differ completely. It then extracts the patterns that made those analogous solutions work and maps them onto the current challenge, producing answers that are both creative and grounded in proven approaches.
Think of it like a seasoned consultant who, when faced with a client’s unique problem, draws on decades of experience across industries to say “I’ve seen something like this before” — then adapts what worked in that analogous context to fit the new situation precisely.
When a model reasons from scratch about a novel problem, it has no anchor — no proven strategy to build upon. Analogical reasoning provides that anchor by having the model first recall cases where similar dynamics played out. The model doesn’t just theorize; it draws on what actually worked in comparable situations. This grounds creative problem-solving in empirical precedent, reducing hallucination risk and producing solutions that inherit the battle-tested logic of their analogues.
The Analogical Reasoning Process
Four stages from novel problem to analogy-grounded solution
Problem Presentation
Present the novel problem clearly, including its constraints, goals, and context. The richer the problem description, the better the model can identify meaningful analogies. This stage frames the challenge and gives the model the raw material it needs to search for structural parallels.
“We need to design an onboarding flow for a complex B2B SaaS product with 50+ features. Users currently abandon during setup because they’re overwhelmed by options.”
Analogy Recall
The model generates similar problems it has encountered across different domains. It looks for structural matches — situations that share the same underlying dynamics (complexity management, progressive disclosure, user guidance) even if the surface context is entirely different. The model may draw from gaming tutorials, museum exhibit design, or apprenticeship models.
“This is similar to how video games handle tutorial design — introducing mechanics gradually through guided levels. It also parallels how IKEA stores guide customers through a predetermined path, exposing them to categories progressively rather than all at once.”
Pattern Extraction
Identify the shared structural patterns between the analogies and the current problem. Strip away the surface details to reveal the underlying principles that made the analogous solutions effective. These extracted patterns become transferable strategies that can be adapted to the new context.
Shared patterns identified: (1) Progressive complexity — start with one core action, add layers only after mastery. (2) Guided pathways — reduce decision paralysis by offering a curated default route. (3) Achievement feedback — reward completion of each stage to motivate continuation. (4) Just-in-time disclosure — show features only when contextually relevant.
Solution Transfer
Apply the extracted patterns to solve the original problem. Map each principle from the analogies onto the specific context of the current challenge. The solution inherits the proven logic of the analogues while being adapted to fit the unique constraints and goals of the new situation.
Transferred solution: Design the onboarding as a 5-stage “quest” — Stage 1 sets up the core workflow (like a game’s first level), each stage unlocks the next feature group with a progress bar, contextual tooltips introduce advanced features only when the user reaches the relevant screen, and a “setup score” provides achievement feedback.
See the Difference
Why analogy-grounded reasoning produces richer solutions
Direct Prompting
How should we reduce employee burnout in our fast-growing startup?
Offer flexible hours, provide wellness programs, hire more staff to distribute workload, and encourage regular breaks. Set clear boundaries between work and personal time.
Analogical Reasoning
Analogy 1: Athletic overtraining syndrome — athletes who train without recovery cycles experience declining performance despite increased effort. Solution: periodized training with structured rest.
Analogy 2: Crop rotation in agriculture — planting the same crop repeatedly depletes soil. Solution: rotate crops and let fields lie fallow.
Pattern: Systems that extract continuously without replenishment degrade. Recovery must be structural, not optional.
Implement “sprint rotation” — teams cycle between high-intensity delivery sprints and recovery sprints (internal tooling, learning, experimentation). Like athletic periodization, this builds recovery into the system rather than treating it as an individual responsibility. Track team velocity across cycles to calibrate the rhythm.
Practice Responsible AI
Always verify AI-generated content before use. AI systems can produce confident but incorrect responses. When using AI professionally, transparent disclosure is both best practice and increasingly a legal requirement.
48 US states now require AI transparency in key areas. Critical thinking remains your strongest tool against misinformation.
Analogical Reasoning in Action
See how cross-domain analogies unlock creative, grounded solutions
“We’re launching an AI-powered analytics feature. Should we bundle it free to drive adoption, or charge a premium to signal value?”
Analogy recalled: Razor-and-blade model (Gillette) — give away the razor, profit from blades. Also: Spotify’s freemium model — free tier drives adoption, premium tier captures willingness to pay once value is demonstrated.
Counter-analogy: Tesla Autopilot — premium pricing from day one signals advanced capability and funds continued R&D. Early adopters pay a premium that subsidizes future improvements.
Pattern extracted: The decision depends on whether the feature is the “razor” (drives adoption of something else) or the “blade” (the revenue engine itself). If analytics makes your core product stickier, bundle it. If analytics is the product, price it to signal value.
Transferred solution: Launch with a free basic tier (the razor) that demonstrates value through limited insights, with a premium tier (the blade) unlocking real-time dashboards and predictive features. This mirrors the Spotify model and lets users self-select based on demonstrated value rather than promised value.
“Our microservices communicate through a tangled web of direct API calls. Adding a new service requires updating dozens of others. How do we decouple this?”
Analogy recalled: Postal system vs. telephone — telephones require direct connections (like API calls), but the postal system uses a central routing hub. A letter doesn’t need to know the recipient’s phone number; it just needs an address and the system routes it.
Second analogy: Airport hub-and-spoke model — instead of direct flights between every city pair (exponential connections), airlines route through hubs. Adding a new city requires only one new route to the hub, not routes to every other city.
Pattern extracted: Point-to-point connections scale quadratically. Hub-mediated routing scales linearly. Introduce an intermediary that handles routing so participants only need to know the hub, not each other.
Transferred solution: Introduce an event bus (the postal system / airport hub) where services publish events without knowing who consumes them. New services subscribe to relevant events through the bus. This reduces coupling from O(n²) direct connections to O(n) bus connections, and adding a new service requires zero changes to existing ones.
“My students struggle with the concept of mathematical limits. They can compute them mechanically but don’t understand what a limit actually means intuitively.”
Analogy recalled: Walking toward a wall — with each step you cover half the remaining distance. You never touch the wall, but you get arbitrarily close. The wall is the limit.
Second analogy: Tuning a radio dial — as you turn the dial, you get closer and closer to the target frequency. Static decreases continuously. The “limit” is the perfect frequency where static disappears, even if mechanical precision means you never hit it exactly.
Pattern extracted: A limit describes the value a process approaches as you refine the input, even if you never arrive exactly. The key insight is the convergence behavior, not the destination itself.
Transferred solution: Teach limits through the radio-tuning analogy — have students physically adjust a slider toward a target value on screen, watching outputs converge. Then connect this experience to the formal notation: “as x approaches a” means “as you tune closer to the target frequency.” The mechanical computation becomes meaningful because students have felt the convergence before formalizing it.
When to Use Analogical Reasoning
Best for novel problems that benefit from cross-domain insight
Perfect For
Challenges with no obvious direct precedent — analogical reasoning finds structural parallels in unexpected places that generic reasoning would miss.
When a fresh perspective from an unrelated field could unlock a breakthrough — biomimicry, design thinking, and interdisciplinary solutions.
Explaining abstract or technical ideas by mapping them to familiar, tangible experiences that the audience already understands.
Learning from how other industries, organizations, or systems solved structurally similar challenges — avoiding reinventing the wheel.
Skip It When
When direct solutions already exist in the same domain — no need for cross-domain analogies when best practices are already established.
When exact, verifiable answers are needed — analogies provide directional insight, not precise calculations or deterministic outputs.
Straightforward lookups or computations that have one correct answer — analogies add unnecessary complexity to questions with direct solutions.
Use Cases
Where analogical reasoning delivers the most value
Product Design
Solve UX challenges by drawing on analogies from physical spaces, games, or other digital products that solved similar user experience problems.
Business Strategy
Identify market strategies by finding analogies in other industries — how did streaming disrupt video rental, and what does that mean for your industry?
Scientific Research
Generate hypotheses by finding parallels between your research domain and established findings in related fields — biomimicry, cross-pollination of methods.
Technical Communication
Explain complex technical concepts to non-technical audiences by mapping them to everyday experiences they already understand.
Risk Assessment
Anticipate failure modes by studying how similar systems failed in other contexts — structural engineering lessons applied to software architecture.
Problem Reframing
When a team is stuck on a problem, analogical reasoning reframes it through a different lens — turning “how do we reduce churn?” into “how do ecosystems maintain biodiversity?”
Where Analogical Reasoning Fits
From manual examples to self-generated analogies
Analogical reasoning works even better when you provide one or two domain-specific examples alongside the instruction to generate more. Your examples prime the model’s search direction, while the model’s self-generated analogies expand the search to domains you might not have considered. This hybrid approach gives you the best of both manual curation and automated discovery.
Related Techniques
Explore complementary reasoning techniques
Reason by Analogy
Try analogical reasoning on your own challenges or build analogy-enhanced prompts with our tools.