Systems Thinking & Silos
A system is a set of interdependent components working together to achieve a purpose. Improving any component in isolation — without understanding how it connects to the others — often harms the system even as it improves the component. This is the core insight of systems thinking, and the reason most improvement efforts produce less than expected.
☰ Contents
What a system is — and why it matters
W. Edwards Deming defined a system as “a network of interdependent components that work together to try to accomplish the aim of the system.” Three words in that definition are doing most of the work: interdependent, together, and aim.
Interdependent means that the components affect each other. What happens in one part of the system has consequences in other parts. A change in discharge processes affects bed availability, which affects A&E queues, which affects ambulance handover times. A change in procurement affects production quality, which affects customer complaints, which affects the support team. The components are not independent — they are connected, and changes propagate.
Together means that the performance of the system is not the sum of the performance of its parts. A group of individually high-performing components that are not working together may produce a poorly performing system. Equally, a system where components coordinate well may perform better than the sum of its parts suggests. The interaction effects — the connections between components — often determine more of the outcome than the individual component performances.
Aim means that a system without a clear shared aim is not a system — it is a collection of components. Deming emphasised this point repeatedly: the aim must be defined, communicated, and used to align the behaviour of all components. Without a shared aim, components optimise for their own local objectives, and the result is sub-optimisation.
Deming used the orchestra as his central example of a well-functioning system. Each musician is excellent individually. But the conductor does not try to optimise each musician’s individual performance — the conductor optimises how they work together. A virtuoso violinist who plays louder than the score requires is not adding value to the system: they are disrupting the whole for individual distinction. The aim of the orchestra is the music, not the individual performances. Every institution that measures individual or departmental performance rather than end-to-end outcomes is ignoring this lesson.
Sub-optimisation — improving the parts at the expense of the whole
Sub-optimisation is what happens when a component of a system is improved in a way that harms the overall system performance. It is the direct consequence of managing components rather than the whole. It is extremely common, often invisible, and almost never intentional.
The mechanism is straightforward: each component is measured on its own metrics and held accountable for its own performance. The rational response for each component manager is to improve those metrics. But because the components are interdependent, improving one metric often shifts burden, cost, or risk to another component. The other component’s metrics deteriorate. From outside, it appears that one function is improving and another is failing. From inside the system, what is actually happening is a transfer of a problem rather than its elimination.
Procurement reduces cost per unit
Switches to cheaper supplier. Unit cost falls. Procurement target met.
Production rework rate rises
New components have higher defect rate. Rework time increases. Customer delivery delays follow.
A&E meets 4-hour target
Patients are rapidly assessed and moved to wards. A&E throughput metric improves.
Ward overcrowding increases
Wards receive patients before beds are clinically ready. Pressure transfers upstream. Readmissions rise.
Discharge team meets bed-turnover target
Patients discharged faster. Bed occupancy metric improves. Cost per bed-day falls.
Readmission rate increases
Patients discharged before they are ready return within 30 days. Community care demand spikes. Total cost rises.
Sales team maximises order volume
All customer requests accepted. Sales targets met. Commissions paid.
Operations is overwhelmed with variety
Non-standard orders flood production. Delivery reliability falls. Customer satisfaction drops.
In each case, the component manager made a rational decision given their metrics. In each case, the system outcome was made worse. The problem is not the people — it is the measurement system that disconnects component performance from system outcome.
How silos form and why they persist
A silo is a component of a system that has become effectively disconnected from the system aim. It operates according to its own internal logic, optimises for its own metrics, and treats other components as external constraints rather than interdependent parts of the same system. Silos are not created by bad people. They are created by measurement systems, incentive structures, and organisational designs that reward local performance over system performance.
The reinforcing loop that creates and sustains silos follows a predictable pattern:
- The organisation measures component performance because system performance is harder to attribute
- Component managers optimise for component metrics because those are the metrics they are accountable for
- Sub-optimisation produces friction at the boundaries between components
- Each component interprets boundary friction as evidence that the adjacent component is failing
- Components invest in protecting themselves from adjacent failures rather than in system improvement
- The boundaries become harder to cross, the silos more defended, and the system performance worse
- Management responds by measuring component performance more rigorously — reinforcing the original problem
Once silos are established, the people inside them cannot easily see the system-level problem. Each component has a coherent internal narrative in which the problem is caused by the adjacent components. From inside the silo, the proposed solution is always: give us more resource, better cooperation from the others, or clearer boundaries. None of these are wrong, exactly — they would help at the margin. But they do not address the system-level measurement problem that created the silos in the first place. Breaking a silo trap requires authority to change the measurement system, not just authority over one of the components.
The NHS as a silo system
The NHS provides the most extensively documented example of silo-induced sub-optimisation in any public institution. Primary care, secondary care, community care, mental health, social care, and public health each have separate funding, separate accountability structures, separate targets, and separate professional cultures. Each has been improved individually and measured independently. The system they collectively constitute has not improved at the same rate — and in many measures has deteriorated.
The A&E crisis is a silo problem. Patients arrive at A&E partly because primary care capacity is insufficient to absorb their needs earlier. They cannot be discharged from A&E because ward capacity is insufficient. They cannot be discharged from wards because community and social care capacity is insufficient to support them at home. Each component has improvement programmes. The system does not improve because no component has accountability for the end-to-end outcome. See the NHS A&E analysis for the detailed data.
Deming’s System of Profound Knowledge
Deming’s System of Profound Knowledge (SoPK) is his answer to the question: what does a manager need to understand to improve a system rather than harm it? It has four components, each of which addresses a distinct category of management error. The components are interdependent — deficiency in any one of them causes characteristic failures, and the failures interact.
Appreciation for a System
Understanding interdependence, the aim of the system, and the consequences of sub-optimisation. Managing the whole, not the parts. The orchestra principle: the aim is the music, not the individual performances.
Knowledge of Variation
Understanding the difference between common cause and special cause variation. Knowing when to act and when to leave the system alone. Managing by facts rather than by reaction to noise.
Theory of Knowledge
Understanding how learning happens: prediction, test, observation. Improvement requires a theory that makes a testable prediction, not just action. The PDSA cycle is the operational implementation of this component.
Psychology
Understanding how people are motivated, how fear destroys information flow, how measurement systems shape behaviour, and why people do what the system makes rational rather than what produces good outcomes.
The four components are not a checklist. They interact. A manager who understands variation but not systems will make statistically sound decisions that are locally correct and systemically harmful. A manager who understands psychology but not variation will create the right culture for tampering to flourish. The full framework is required. Most management training provides fragments of one or two components and leaves the rest.
Deming’s 14 Points — his prescriptions for management transformation — are the operational expression of the System of Profound Knowledge. Each Point addresses a specific management error that arises from ignorance of one or more of the four components. Point 11 (“eliminate numerical quotas”) is a Psychology and Variation point. Point 3 (“cease dependence on inspection”) is a Systems and Variation point. The Deming’s 14 Points concept page develops this in full.
The feedback diagram
Deming used a feedback diagram to show how a system connects its components to its aim through information flow. The diagram is not complex, but it encodes a principle that most organisations violate: information about outcomes must flow back to the people and processes that produce them. Without feedback, a system cannot learn. It continues to produce the same outputs regardless of whether those outputs are achieving the aim.
Adapted from Deming, W.E. Out of the Crisis (1982). The feedback loop (blue dashed) is the element most organisations omit: outcome information must flow back continuously to every stage, not just be collected at the end. The design loop (purple dashed) shows that feedback should also drive system redesign — not just process correction.
The diagram encodes several important points that are easy to miss:
- The feedback loop goes all the way back to suppliers. If incoming material or information quality is a cause of poor outcomes, that signal must reach the supplier, not just the production stage that receives it.
- Each stage is both a customer and a supplier. Stage B receives from Stage A and supplies to Stage C. The quality of the relationship at every boundary determines whether the whole system works. Silo thinking treats each stage as independent. Systems thinking treats each boundary as a design decision.
- The design and redesign loop connects outcomes to system structure. Outcome feedback that only adjusts parameters (turns up a setting, retrains a person) operates within the existing system design. Outcome feedback that changes the system design is what produces structural improvement. This is the difference between the PDSA cycle as a corrective loop and the PDSA cycle as a learning loop.
- Without the feedback loop, the system is open. An open system — one with no feedback from outcomes — cannot improve. It produces its current outputs indefinitely, regardless of whether those outputs achieve the aim. Most quality systems collect data but do not route it back to where it would change behaviour. The data is used for reporting, not for learning.
Systems thinking in the NHS
The NHS is one of the most extensively studied examples of a system that is managed as if it were a collection of independent components rather than an interdependent whole. The consequences are visible in the data.
The NHS A&E analysis on this site shows that A&E performance has not shown a sustained structural improvement despite decades of investment, multiple reorganisations, and repeated government programmes. The Bootstrap CUSUM analysis finds no lasting change points. The reason, from a systems thinking perspective, is that most interventions targeted individual components — A&E capacity, GP access, discharge processes — rather than the system that connects them.
The dementia diagnosis rate analysis tells a similar story. A target was set for a single component (diagnosis rate). Services were redesigned to meet that target. The target was met. But the system that the target was embedded in — community support, carer capacity, post-diagnosis care — was not redesigned alongside it. Meeting the diagnostic target without changing the post-diagnosis system shifted burden to services that were not designed to absorb it.
Each NHS trust is accountable for its own performance. Each clinical directorate within a trust is accountable for its own metrics. Each component of the pathway is measured and managed independently. The patient experiences the entire pathway as a single system. The patient’s experience is measured, if at all, in surveys that arrive months after the pathway is complete, too late and too aggregated to drive improvement in any specific component. The feedback loop is so attenuated that the system cannot learn from its own outputs. Bootstrap CUSUM on outcome measures — applied in real time to the end-to-end pathway rather than to individual component metrics — is the structural fix.
Plain language summary
💬 Systems thinking in plain language
A system produces outcomes. Components produce outputs. Managing components produces outputs. Managing the system produces outcomes.
If your problem keeps recurring despite repeated fixes, the fixes are addressing component outputs rather than system outcomes. The system is unchanged, so it continues to produce the same outcomes.
The practical questions: What is the aim of the whole system? Who is accountable for end-to-end outcomes rather than component metrics? Where does outcome information flow back to, and how quickly? What happens at the boundaries between components — and who owns those boundaries? Where is the constraint that limits the whole system, regardless of how well individual components perform?
If you cannot answer the first question, the system does not have a shared aim. If you cannot answer the second, nobody is accountable for what matters. If you cannot answer the third, the feedback loop is broken and the system cannot learn.
Where this fits in the 7-step method
Step 1 (List symptoms): If the symptom list includes problems that are owned by different functions and where each function believes the others are responsible, the Silo Trap is operating. List symptoms by the end-to-end pathway experience, not by departmental ownership.
Step 3 (Root cause): Causal Loop Diagrams are the systems thinking tool for root cause analysis. They show reinforcing and balancing loops, not just linear causal chains. The silo trap is a reinforcing loop: measurement of components drives sub-optimisation, which creates boundary friction, which reinforces defensive silo behaviour, which perpetuates component measurement. The loop must be identified before it can be broken.
Step 4 (Dominant constraint): The Theory of Constraints is systems thinking applied to throughput: find the single bottleneck that limits the whole system and address that, not the components that are most visible or most complained about. The Theory of Constraints page develops this.
Step 5 (Complete solution): A solution that improves one component without changing the measurement and accountability structure that created the silo will be absorbed by the system. The complete solution must include the feedback loop redesign: who receives outcome information, how quickly, and with what authority to act on it.
Step 7 (PDSA — Study): The outcome measure for a systems-level fix should be an end-to-end measure, not a component measure. If the Bootstrap CUSUM test is applied to a component metric, it will detect component improvement while missing whether the system aim has been achieved. Choose the measure that captures what the system is for.
Systems Thinking & Silos sits in the Systems & Constraints group — alongside Theory of Constraints, Ashby’s Law, and Failure Demand. All four describe different mechanisms by which component-level thinking fails to produce system-level improvement.