How The World Works | Complementarity

A key without a lock is a strangely shaped piece of metal.

A lock without a key is a secure inconvenience.

Together, under the right design, they become an access system.

The value is not located in either object alone.

It appears in the fit.

The same pattern appears everywhere. A smartphone becomes more useful with compatible software and networks. A trained worker becomes more productive with suitable tools. A railway becomes more valuable when stations connect to places people actually need to reach. A school curriculum becomes more effective when assessment, teacher knowledge and learning resources point in the same direction.

This is complementarity.

Complementarity is one of the reasons isolated improvements so often disappoint. The thing we upgraded may genuinely be better. The surrounding system may simply not contain what the upgrade needs in order to become useful.

Quick Read

Complementarity exists when the value, productivity or usefulness of one input rises because another input is present.

In consumer economics, complementary goods are often used together. OpenStax gives familiar examples such as coffee and sugar or golf clubs and golf balls: when the price of one rises enough to reduce its use, demand for the complement can also fall.

In organisations, complementarity can be deeper. Training, decentralised decision-making, incentives, information systems and work design may produce much more value as a coherent bundle than when introduced one at a time.

NBER evidence from steel production found that coherent systems of modern work practices were associated with substantially higher productivity, while isolated practices showed little effect. Other NBER work finds evidence that decentralisation and performance pay can be complementary.

The central question is:

What else has to be true for this thing to become valuable?

The One-Sentence Answer

Complementarity works when two or more inputs interact so that the marginal benefit of one becomes greater when the others are present, making coordinated bundles more productive than isolated additions.

The Difference Between Addition and Fit

If two independent machines each produce ten units, putting them side by side may produce twenty.

That is addition.

If machine A produces little without machine B, but together they produce far more than either could alone, the relationship is complementary.

Complementarity is therefore a statement about interaction.

The system does not merely contain more components.

The components change one another’s value.

The Complementarity Chain

input A → limited value alone → input B changes feasible use of A → joint capability → higher marginal return → more adoption of the bundle → supporting ecosystem

This can become self-reinforcing.

Better software makes hardware more useful. More hardware adoption attracts developers. More software attracts users. Training makes tools more productive. Productive tools justify more training.

Complementarity can therefore help build path dependence.

Complements Are Not Merely Things Used Together

We should be precise.

Two things can often appear together without being strongly complementary.

People may consume them together because of habit or convention. A statistical correlation is not enough.

The stronger claim is that changing one changes the value or demand for the other.

For organisational practices, the test becomes harder because high-performing firms may adopt many good practices at once. Researchers therefore need careful identification to separate true complementarity from simple co-adoption.

NBER work by Athey and Stern explicitly discusses the empirical difficulty of testing complementarities in organisational design.

Cross-Price Elasticity: The Consumer Version

In standard microeconomics, complements often have negative cross-price elasticity.

If the price of one good rises and people buy less of it, demand for a good used with it can also fall.

OpenStax uses the example of coffee and sugar: a higher price for one can reduce demand for the other when the goods are consumed together.

This is useful but narrow.

The richer systems idea asks whether one component raises the productivity, feasibility or value of another even when no consumer purchase is involved.

Hardware Needs Software

A computer with no useful software is less valuable than the same machine inside a rich software ecosystem.

Software also needs hardware, operating systems, standards and distribution.

This joint dependence explains why technology transitions rarely happen one component at a time.

New hardware may fail because applications are missing. New applications may fail because the installed hardware base is too small. A technically superior standard can lose because the complementary ecosystem around an incumbent is stronger.

The product is partly the network of complements around it.

Skills Need Tools

A highly trained worker using poor equipment may underperform a moderately trained worker using excellent equipment.

Training and technology can therefore be complements.

The reverse also matters. Installing sophisticated technology without building user capability can create an expensive shelf of unused potential.

This is one reason technology projects should budget for training, workflow redesign, support and feedback rather than treating the hardware or software purchase as the intervention.

Autonomy Needs Information

Decentralising decisions can improve speed and local adaptation.

But autonomy without information can simply move ignorance closer to the problem.

Local decision-makers need relevant data, competence, authority boundaries and incentives aligned with the wider objective.

This is why organisational practices often arrive in bundles.

Performance pay may interact with decentralisation. Team-based work may need training and information sharing. Quality systems may need measurement and local problem-solving authority.

One practice can be weak because the complement is missing.

The NBER Steel-Plant Lesson

A classic empirical study by Casey Ichniowski, Kathryn Shaw and Giovanna Prennushi examined steel finishing lines and found that coherent systems of innovative human-resource practices were associated with substantially higher productivity.

The interesting result was not merely “teams are good” or “training is good.”

Individual practices adopted in isolation showed little productivity effect compared with bundles involving teams, flexible job assignments, employment security, multi-skill training and incentive pay.

The system of practices mattered.

This is complementarity in operational form: the value of one practice depends partly on the presence of others.

Why Best Practices Fail When Copied

An organisation visits a high-performing institution.

It notices one visible feature.

Stand-up meetings. Open offices. Performance bonuses. Data dashboards. Small teams. One-to-one devices. A particular curriculum.

It copies the feature.

Nothing much happens.

The copied practice may have been a complement inside a larger architecture.

Remove it from the system of skills, incentives, culture, standards and information that made it productive and the visible object remains while the causal mechanism disappears.

This is why benchmarking needs dependency analysis, not sightseeing.

Complementarity Creates Bundles

If A raises the value of B and B raises the value of A, adoption tends to cluster.

Firms do not merely buy one isolated technology. They reorganise work around it. Cities do not merely build rail. They connect stations, land use, walking routes, ticketing and feeder services. Schools do not merely introduce a syllabus. They align teacher development, materials, assessment and progression.

This creates system bundles.

Bundle thinking changes investment logic: the cheapest component individually may not produce the cheapest working system.

The Missing Complement Problem

A system buys the visible input and forgets its partner.

Laptops without connectivity.

Data without analytical capability.

Autonomy without accountability.

Training without opportunities to use the skill.

Public transport without safe access to stations.

New curriculum without time for teachers to understand it.

The intervention appears weak because the production function expected a pair and received one half.

Complementarity and Bottlenecks

Complements can behave like a chain.

If every component is necessary, the weakest one can limit the whole.

An excellent train system has reduced value if the final kilometre is inaccessible. A strong curriculum has reduced impact if students cannot read the instructions. A powerful model has limited use if data quality is poor. A brilliant strategy fails if execution capacity is missing.

This gives complementarity a diagnostic edge:

The highest-return investment may be the missing complement, not another upgrade to the strongest component.

Balanced Growth

When complements are strong, uneven investment can create stranded capability.

Build generation without transmission and electricity cannot reach users. Build ports without inland logistics and cargo stalls. Train specialists without diagnostic equipment and skill cannot be fully used.

Balanced growth does not mean every component receives equal spending.

It means complementary constraints are developed enough that one investment can actually activate another.

Complementarity and Standards

Standards create complements by making components fit.

A standard socket increases the value of compatible devices. A common data format increases the usefulness of software that can exchange files. Container standards increase the joint value of ships, ports, cranes, trucks and warehouses.

Compatibility converts separate assets into a system.

This is one reason standards can generate enormous network value without being glamorous themselves.

See How Standards Work.

Complementarity and Interoperability

Interoperability lets independently built systems become complements.

A map becomes more useful when transport data can feed route planning. A payment service becomes more useful when merchants and banks can exchange recognised messages. Educational records become more useful when they can travel across systems with shared meaning.

Without interoperability, potential complements remain isolated islands.

See How Interoperability Works.

Complementarity and Substitution

Substitutes replace one another.

Complements increase one another’s value.

But the same two technologies can be both, depending on the task.

AI may substitute for a human drafting task while complementing human verification. Public transport may substitute for car travel while complementing walking and cycling. A calculator substitutes for manual arithmetic in some contexts while complementing higher-level mathematical modelling.

The relationship belongs to the task, not permanently to the object.

See How The World Works | Substitution.

Complementarity and Nonlinearity

Complementarity creates nonlinear returns.

Add A alone and little happens.

Add B alone and little happens.

Add both and the system crosses into a more productive state.

This is why linear cost-benefit analysis can underestimate bundles if it evaluates each component in isolation.

See How The World Works | Nonlinearity.

Complementarity and Path Dependence

Once a bundle is installed, each complement can reinforce the others.

Users learn the system. Suppliers specialise. Standards spread. Infrastructure grows. Replacing one component now requires compatibility with all the rest.

The complementarity that created value can later create lock-in.

This is not a contradiction.

The same interdependence that makes a system productive can make it harder to change.

See How The World Works | Path Dependence.

Complementarity and Irreversibility

Deep complementary systems can create practical irreversibility.

Changing one part may require changing many others.

This is why modularity is valuable: it preserves complementarity through interfaces while reducing the number of components that must be changed together.

Good architecture seeks strong cooperation inside modules and lower coupling across replaceable boundaries.

See How Modularity Works.

The Pairing Error

Some interventions fail because the complement is present but paired badly.

More autonomy plus incentives tied to the wrong metric can intensify gaming. More technology plus poor workflow can accelerate confusion. More data plus weak governance can produce faster bad decisions.

Complementarity does not mean every combination is beneficial.

Inputs can also be antagonistic.

The interaction sign matters.

The Sequence Problem

If A depends on B, installing A first can make A look like a failure.

Sometimes complements should arrive together.

Sometimes one must arrive first to prepare the system.

Teacher training may need to precede new curriculum implementation. Data standards may need to precede system integration. Grid upgrades may need to precede large-scale electrification.

Complementarity therefore creates a scheduling problem.

Sequence can determine whether the bundle ever reaches the state in which its joint value becomes visible.

The Coordination Problem

When complementary investments belong to different actors, each may wait for the other.

Charging infrastructure is less attractive without electric vehicles. Electric vehicles are less attractive without charging infrastructure. Developers hesitate to build applications for a small platform. Users hesitate to adopt a platform without applications.

This can create a coordination failure even when the fully developed system would be valuable.

Standards, commitments, subsidies, anchor customers or coordinated rollout can help cross the chicken-and-egg gap.

The Capability Stack

Many advanced outcomes sit on a stack of complements.

A researcher needs domain knowledge, methods, instruments, data, compute, publication channels and a community capable of criticism.

A surgeon needs training, theatre, instruments, anaesthesia, imaging, nursing, sterilisation, blood supply, records and postoperative care.

A modern port needs shipping lanes, cranes, standards, customs, data systems, roads, rail, warehouses, finance and labour.

The visible expert or machine is often only the top of a complementary stack.

Complementarity and Civilisation

Civilisation scales through complements.

Writing becomes more valuable with education and archives. Roads become more valuable with vehicles, maps and maintenance. Money becomes more useful with trust, law and institutions. Science becomes more productive with instruments, universities, journals and open criticism.

No single infrastructure explains the whole.

Capability is stacked.

Complementarity in Education

Teaching is full of complements.

Vocabulary supports reading comprehension. Background knowledge supports inference. Retrieval supports fluent access. Feedback becomes useful when the student can understand it. Practice becomes more powerful when tasks are correctly sequenced. Motivation becomes more productive when the learner has a feasible route.

This is why “just practise more” often fails.

Practice may be the missing ingredient.

Or practice may already be abundant while the missing complement is representation, vocabulary, attention, feedback quality or prerequisite knowledge.

The Small-Group Complement

A small class is not automatically better because it has fewer students.

Its value depends on complementary teaching behaviour.

More observation. Better diagnosis. Faster feedback. More opportunities to explain. Tasks adjusted to the actual state. Comparison among different solution routes.

Without those complements, small class size can become an expensive room with empty chairs.

The complement is what converts capacity into learning.

Complementarity and AI

AI capability is highly complementary.

A model becomes more valuable with good data, relevant context, usable interfaces, domain expertise, verification, tool access and clear accountability.

A powerful model inside a poor workflow can create little value.

A weaker model inside an excellent retrieval, verification and user-feedback system may outperform it on the actual task.

This is why model comparison alone is often the wrong level of analysis.

The product is the stack.

Complementarity Can Create Inequality Between Systems

Once capability is complementary, advantage can compound.

A well-resourced organisation buys better tools. Its skilled staff use them effectively. Stronger results attract more talent. Better data improves decisions. Better decisions justify further investment.

A weaker organisation buys one component but lacks the complements needed to activate it.

The gap widens.

This is a nonlinear inequality mechanism produced by bundle effects.

The Complementarity Audit

Define the outcome. What capability are we trying to create?
List the major inputs. Tools, skills, information, infrastructure, incentives, standards?
Test independent value. What does each input accomplish alone?
Test joint value. Does A become more useful when B is present?
Find missing complements. What prevents the visible investment from becoming productive?
Find antagonists. Which inputs interfere rather than reinforce?
Check sequence. Which complement must arrive first?
Check coordination. Are different actors waiting for one another?
Check standards. Can the components actually fit?
Check bottlenecks. Which weak complement caps the whole stack?
Check scale. Does the value of the bundle rise as adoption grows?
Check path dependence. Will the bundle become hard to change later?
Check modularity. Can complements cooperate without becoming inseparable?
Measure the bundle. Are we evaluating the system rather than isolated components?

When the Complementarity Lens Fails

Complementarity can become an excuse for endless scope.

“This project needs ten more complementary projects before it can succeed.”

Sometimes that is true.

Sometimes it is a refusal to test whether the original intervention has independent value.

Good complementarity claims should specify the interaction mechanism and be testable.

Not everything belongs in the bundle.

The Minimum Viable Complement Set

A useful design question is not “What is the perfect ecosystem?”

It is:

What is the smallest set of complements required for this capability to become genuinely usable?

This keeps systems thinking from becoming paralysis.

Build the minimum working bundle.

Observe.

Then add the next complement where evidence shows a constraint.

How Complementarity Connects to the Rest of the World

Substitution: substitutes replace; complements increase joint value.
Nonlinearity: joint inputs can create more than additive returns.
Standards: standards let components fit and become useful together.
Interoperability: exchange enables independently built complements to cooperate.
Capacity: a missing complement can become the binding bottleneck.
Path dependence: complementary ecosystems reinforce installed systems.
Irreversibility: tightly interdependent bundles are harder to unwind.
Modularity: good interfaces preserve complementarity while protecting replaceability.
Coordination: complementary investments may need simultaneous or sequenced adoption.
Incentives: autonomy or technology can require aligned rewards to produce value.
Information: decentralised action needs decision-relevant information.
Second-order effects: adding one complement can change demand for many others.

Questions a Reader Can Now Ask

What else must be present for this investment to work?
Does this input have value alone?
Does another input raise its marginal value?
Which missing complement is limiting the system?
Are we copying a visible practice without its hidden bundle?
Do the components share standards and interfaces?
Must they be adopted together or in sequence?
Who owns each complementary investment?
Could coordination failure prevent joint adoption?
Does the bundle create lock-in?
Can modularity preserve future replacement?
What is the minimum viable complement set?

Frequently Asked Questions

What are complementary goods?

They are goods often used together so that consumption of one tends to enhance consumption of the other. In demand analysis, they often show negative cross-price elasticity.

How is complementarity different from teamwork?

Teamwork is coordinated human activity. Complementarity is a more general relationship in which one input becomes more valuable because another is present. The inputs can be people, technologies, standards, skills or infrastructure.

Why do isolated best practices fail?

Because the observed practice may rely on complementary skills, incentives, information or infrastructure in the original organisation. Copying only the visible component can remove the mechanism that made it work.

Can complements also create lock-in?

Yes. A rich complementary ecosystem can increase value while making switching harder because replacing one component requires compatibility with many others.

What is the most useful design question?

Ask what missing complement prevents the current investment from producing its expected value.

Research Basis and Further Reading

OpenStax, Principles of Economics: complements.
OpenStax, Cross-Price Elasticity of Demand, distinguishing substitute and complementary goods.
Casey Ichniowski, Kathryn Shaw & Giovanna Prennushi, “The Effects of Human Resource Management Practices on Productivity”, NBER, on coherent bundles of complementary work practices.
Bryan Hong, Lorenz Kueng & Mu-Jeung Yang, “Estimating Management Practice Complementarity between Decentralization and Performance Pay”, NBER.
Susan Athey & Scott Stern, “An Empirical Framework for Testing Theories About Complementarity in Organizational Design”, NBER.

The Larger Idea

We like to point at heroes.

The bridge. The train. The teacher. The computer. The hospital. The scientist. The policy.

But most important capabilities are not single objects.

They are arrangements.

A train is track, signalling, stations, power, maintenance, timetables, ticketing and the places at both ends. A teacher is curriculum, subject knowledge, classroom conditions, student readiness, assessment and time. A computer is hardware, software, standards, electricity, networks and people who know what to ask it to do.

Take away the complement and the hero shrinks.

Add the right complement and an ordinary component can become extraordinary.

Some things are valuable because of what they are. Others become valuable because the world around them finally fits.

How The World Works | Complementarity — Why Some Things Become More Valuable Together