Module 4: Leverage Points and Change – Lesson 3
This lesson is just one part in our series on Systems Thinking. Each lesson reads on its own, but builds on earlier lessons. An index of all previous lessons can be found at the bottom of this page.
Every system resists change. Not out of stubbornness, and not irrationally — but because stability is what systems are built to produce. When you intervene, you are not pushing against a passive structure. You are pushing against a structure designed to push back.
Understanding that resistance is not a problem to overcome but a signal to read is what separates effective change from repeated, exhausting failure.
What You’ll Learn
- Six diagnostic steps for anticipating resistance before it stops you, each with its input, method, and output
- Why systems generate resistance as a feature, not a flaw
- The six forms resistance takes, and which diagnostic step detects each one
- How a £200 million hospital records rollout illustrates these principles in action
- Why treating resistance as feedback is the most reliable path to lasting change
How Can Leaders Anticipate Resistance Before It Derails Change?
Leaders can anticipate resistance through six diagnostic steps: mapping stakeholder incentives, identifying loops that defend the status quo, assessing system bandwidth, surfacing hidden mental models, running safe-to-fail experiments, and making progress visible to those bearing the costs of change.
Each step below states what you need before you start it, how to perform it, and what it produces. The outputs chain: the stakeholder table feeds the loop map, the loop map tells the probe what to watch for, and what the probe finds is what gets reported back. Run them in order the first time through. The two sections after this one explain why the resistance these steps find is structural rather than personal, and map each form it takes back to the step that detects it.
Stakeholder and incentive mapping identifies who benefits from the current arrangement and what they stand to lose. A change effort can promise collective gain while quietly threatening the authority of a single team — and that team’s resistance can stall the entire initiative. The question is not whether people are being unreasonable. The question is whether the change has accounted for what they will lose.
- Input: the scope of the change, and a list of every role whose work, reporting line, budget, headcount, or standing moves as a result.
- Method: for each role, write one line on what it controls today and one line on what it controls after. The difference between those two lines is the loss — write it as a noun, not a feeling. Then mark two things against it: whether the loss arrives before or after the promised gain, and whether the role can stall the rollout on its own. Sort by those two marks together.
- Output: a table — role, controls now, controls after, loss, when the loss lands, ability to stall. The rows where an early loss meets real stalling power are the change’s risk register, and they are where every later step starts looking.
Loop mapping makes visible the incentive structures, metrics, and cultural scripts that are actively defending stability. These loops are rarely written down. They live in what gets measured, what gets rewarded, and what happens to people who deviate. Naming them is a drawing exercise, and the notation is the loop-and-arrow language set out earlier in the course.
- Input: the stakeholder table from step 1, plus the evidence of what the organization actually rewards — performance metrics, bonus and promotion criteria, the last two years of who got promoted for what, and the story people tell about the last person who tried this.
- Method: build a causal loop diagram. Write the behavior you want to change as a variable, and name every variable as a quantity you could imagine going up or down — “time spent on handover”, “number of exceptions escalated” — never as an action or a judgment. Ask what feeds that variable, draw an arrow from cause to effect, and sign it: + if they move the same way, – if they move opposite. Keep following the chain until an arrow lands on a variable already on the page. That closed path is a loop. Count its minus signs — an odd number makes it balancing rather than reinforcing, and a balancing loop around the behavior you want to change is a loop defending the status quo. Mark where the effect takes weeks or months to arrive; those delays are why people conclude the change failed before it has had time to work. Name each balancing loop for the mechanism driving it, in the organization’s own words.
- Output: a diagram in which every balancing loop defending the current state is drawn, signed, and named. Each named loop is a redesign target, and the map is finished when you can say in one sentence which variable a proposed intervention changes and which loop that breaks.
Bandwidth assessment is a capacity check. No matter how well-designed a change effort is, an organization already at full operational load cannot absorb it — and load is carried by teams, not by organizations, so an average across the whole business will hide the team that is already underwater.
- Input: every initiative landing on the same teams during the change window — not only yours — and, for each affected team, its current operating load and its record from the last few rollouts.
- Method: count three things per team, per week. First, concurrent changes: how many separate initiatives require this team to learn or do something differently in the same window. Second, hours of impact per person — training, running the old and new process side by side, rework, and the meetings the change adds — as a share of the working week. Third, absorption evidence: of the last few changes this team was given, how many are still being done the new way, and how many quietly reverted. Set the threshold from that third count rather than from a published benchmark: the load at which this team’s last reversion happened is the number that predicts the next one. Then plot the change against the calendar and move it until no team crosses its own line.
- Output: an impact profile by team and week with the crossing points marked, and one decision per team — go, stage, or defer — with the sequencing that keeps each team under its line.
Mental model surfacing gets at assumptions so embedded they are nearly invisible: which roles are considered “real” work, who has standing to make decisions, what counts as success. Surfacing them takes what Chris Argyris called double-loop learning: questioning not just how things are done but why they are done that way. Argyris drew the distinction in Harvard Business Review in 1977 — single-loop learning corrects the error and leaves the assumption that produced it untouched; double-loop learning goes after the assumption. Without it, even well-funded change efforts grind against invisible boundaries.
- Input: the notes or recollection of the meetings where the change stalled, and any arrow on the loop map that you could draw but not explain.
- Method: three passes. Collect the sentences people said, verbatim, that carry an unexamined rule — “that’s not how it works here”, “the clinicians will never accept it”, “that isn’t real work”. Then, for each one, ask the double-loop question out loud in the room: not how to do the thing better, but what would have to be true for that rule to be the right one. Write the answer down as a plain belief statement, in the words of the person who holds it, and put their name against it. Finally, test each belief against the record — name the most recent time it held, and the most recent time it didn’t. A belief nobody can date is doing more work than it has earned.
- Output: a short list of belief statements, each with an owner, the evidence on both sides, and one of three marks: holds, retire, or unresolved. The ones marked “holds” are constraints to design around; the ones marked “retire” are the change’s real work.
Safe-to-fail experiments replace grand rollouts with disciplined probes. The term is Dave Snowden’s, from the Cynefin framework: in a complex domain you probe first and read what comes back, because the system cannot be understood in advance of being disturbed. Small experiments do not just limit risk; they make resistance specific and visible. A pilot reveals which loops activate, which stakeholders mobilize, and where bandwidth actually runs out. That intelligence is more useful than any theoretical change plan.
- Input: the named balancing loops from step 2 and the top rows of the stakeholder table — the specific mechanisms you want to watch activate — and a list of the units small enough to run a pilot in.
- Method: pick the unit that contains the loop, however inconvenient that is. A pilot that carries the task but not the mechanism will come back clean and tell you nothing. Before it starts, write the prediction down: which loop activates, who objects, what breaks first, and roughly when. Agree the two thresholds in the same sitting — the result that would make you scale, and the result that would make you stop — because both get renegotiated once the numbers are in the room. Run it long enough for the delays you marked on the map to play out; a probe that ends before the feedback arrives has measured enthusiasm rather than adoption. Then compare what happened against the prediction.
- Output: a record of prediction, observation, and the gap between them — the gap is the finding, not the pass or fail — plus an amended loop map carrying whatever the probe revealed that the map had missed.
Visible progress addresses the information failure that breeds cynicism. If the people bearing the costs of change cannot see its benefits, they will reasonably conclude the effort is not worth it. What keeps the case alive through the difficult middle is not the volume of reporting but who it is addressed to.
- Input: the column from step 1 recording who bears the cost and when they bear it, and from step 5 the measures that actually moved.
- Method: report to the people paying, in the currency they are paying in. A team whose week got two hours longer needs to see its own two hours, not an adoption percentage for the whole business. Publish the open problems next to the wins — named, dated, each with an owner and a date it will be looked at again — because a report with no failures on it gets read as marketing and then stops being read. Fix the interval and keep it short enough that a missed one is noticed. And open every report with what changed because of the last round of feedback; that line is the whole point, and it is the first one to get dropped.
- Output: a standing report with a named owner, a fixed cadence, and an open-issues list that carries forward — and, in the record over time, items visibly moving from open to closed.
Why Do Systems Resist Change?
Systems resist change because they are designed to maintain equilibrium. Every feedback loop, every incentive structure, every informal norm functions to preserve the current state. When a change effort pushes the system toward a new equilibrium, these mechanisms activate to push it back.
This is not dysfunction. A system that could not resist disruption would be brittle — unable to distinguish signal from noise, unable to maintain any coherent function. The same stability that makes resistance frustrating is what makes the system reliable under normal conditions. The challenge for change leaders is not to defeat that stability. It is to find the specific loops holding it in place and change what those loops measure or reward — which is what the six steps above are for.
As a general rule, the stronger a system’s current performance, the stronger its resistance to change — because more loops, resources, and identities are invested in the current equilibrium.
In our work at Subverse, that rule turns up most often in brand and communication rollouts, and it rarely arrives as disagreement. A new set of language gets built, approved, and installed. It then holds in the places we touched and reverts everywhere else. The pages we wrote keep saying what they were built to say, while the next campaign brief, the next product page, the next recruiter post drift back within a couple of quarters.
The loop doing that is a measurement loop. Whoever writes the next thing is being measured on turnaround, reach, or volume, and never on whether the piece agrees with everything else the organization has said. New language costs time in that loop and returns nothing to it, so the loop rejects it — without a single objection being raised in a meeting. Resistance then scales the way this lesson predicts: the strongest pushback comes from the channel performing best on its own numbers, because that team has the most to lose by putting a working figure at risk for a change it did not ask for.
The intervention is not a better rollout or more training. It is changing what the loop measures. When the review question becomes whether a piece carries the same meaning as everything around it, the loop that was reverting the work starts defending it instead. That is the test we hold our own engagements to: set something we built beside something the client’s team made a year later, and the two should carry the same meaning, whatever the difference in format or polish. Where they do, the loop changed. Where they don’t, we changed the language and left the system that produces it alone.
What Forms Does Resistance Take?
Resistance to change takes six primary forms in most organizational systems: incentive misalignment, fear of loss, capacity constraints, defensive feedback loops, hidden mental models, and classic system traps like policy resistance and fixes that fail.
Incentives are the most visible. Status, authority, and resource allocation are tied to the current structure, and people whose position depends on how things work now have rational reasons to oppose change. Fear of loss amplifies this. Daniel Kahneman and Amos Tversky’s prospect theory, published in Econometrica in 1979, established that people weight potential losses more heavily than equivalent gains — losses loom larger. So even the beneficiaries of a change resist it when the losses arrive before the gains do.
Capacity is less visible but equally powerful. Organizations can only absorb so much disruption at once. When change efforts compete with existing operational demands, bandwidth runs out before adoption happens — and the system reverts to what it knows.
Feedback loops defend the status quo through incentive schemes, performance metrics, and cultural norms that reward current behavior and penalize deviation. Until those loops are recognized, new initiatives will be absorbed rather than adopted.
The system traps have the longest paper trail of the six. Donella Meadows named the first of them policy resistance in Thinking in Systems: when several parties pull the same stock toward incompatible goals, every effective intervention drags it further from somebody else’s target, and the effort each party spends holding its position is what keeps the system parked where none of them want it. Peter Senge’s fixes-that-fail archetype describes the sibling pattern — the remedy that relieves the symptom, removes the urgency that would have forced a real repair, and lets the original problem return larger. Both sit low on Meadows’ leverage hierarchy, which is why they survive so much well-funded attention.
The six forms map onto the six steps. Each form is what a particular step is looking for, and what that step finds is what the intervention has to change.
| Form of resistance | Diagnostic step that detects it | What the intervention has to change |
|---|---|---|
| Incentive misalignment | Step 1 — stakeholder and incentive mapping | What the role is measured and rewarded on, before anyone asks it to work differently |
| Fear of loss | Step 1 — the column recording when the loss lands relative to the gain | The sequence: bring the gain forward to meet the cost, and report to the people paying in the currency they are paying in (step 6) |
| Capacity constraints | Step 3 — bandwidth assessment | The calendar: stage or defer until every affected team sits under its own reversion line |
| Defensive feedback loops | Step 2 — loop mapping | What the balancing loop measures or rewards, not the people inside it |
| Hidden mental models | Step 4 — mental model surfacing | The beliefs that fail the record: retire those, design around the ones that hold |
| System traps: policy resistance, fixes that fail | Step 5 — safe-to-fail experiments, in the gap between prediction and observation | The goal or rule the trap is serving, higher up the leverage hierarchy than the symptom |
The most common mistake in change management is treating resistance as individual psychology — a motivation problem — when it is almost always a structural one. The solution is redesigning the loops, not re-energizing the people.
What Does Navigating Resistance Look Like in Practice?
In October 2014, Cambridge University Hospitals moved Addenbrooke’s and the Rosie off paper and onto a single electronic record, becoming the first hospital in England to run a fully paperless patient record system. The program cost around £200 million. Inspectors from the Care Quality Commission arrived six months later, and by September 2015 the trust had been rated inadequate and placed in special measures — a sequence documented in Katie Purohit’s case study The National Health Service’s ‘special measures’: Cambridge (Health Services Management Research, 2020). The six diagnostic lenses account for what happened, and for how the trust climbed back out.
The case for switching was strong: faster access, fewer errors, smoother handoffs between departments. But a system does not process the case for change. It processes the disruption.
Stakeholder mapping would have found clinical staff facing a workload spike with no offsetting relief. That group’s resistance is the one the evidence weights heaviest. In a study of 223 nurses across four university hospitals, Cho, Kim and Choi found resistance to change carried the largest total effect on resistance behavior, at 0.65, while perceived usefulness pulled the other way at −0.33 (Factors associated with nurses’ user resistance to change of electronic health record systems, BMC Medical Informatics and Decision Making, 2021). What moves that resistance is the benefit showing up in the day’s work. By the time a rollout is underway, the explaining has usually been done twice over.
Loop mapping would have exposed the paper the system could not hold. Inspectors found staff confused about what to do with records that had no home in the new system, and discharge information going out inaccurate — a documentation loop that had kept the old process viable, now broken with nothing behind it. Bandwidth assessment would have registered a trust absorbing a full estate’s worth of process change in a single cutover while running at capacity. And mental model surfacing would have reached the assumption underneath the schedule: that a records system is an IT deployment rather than a redesign of how clinical work is recorded, coordinated, and trusted.
This is where safe-to-fail experiments were missing. Cambridge went live across the trust at once. A ward-level pilot would have surfaced the same failures at a scale the organization could absorb — the orphaned paper records, the inaccurate discharge summaries, the clinicians unable to reach information they needed, all of which the Care Quality Commission’s April 2015 inspection went on to find at full scale. The resistance was not unpredictable. It was never given anywhere small to appear first.
The trust kept the system and started reading the resistance it had generated. Clinicians were brought into the work of fixing what they had been handed, and visible progress did the rest — the failures inspectors had named became the list the organization worked through in public, against a regulator’s published account of them. Cambridge left special measures in January 2017, with the same system still running. Resistance, treated as feedback, eventually produced the rollout the original plan should have been. It is a great deal cheaper to collect that feedback from one ward than from a regulator.
Conclusion
Resistance is not the enemy of change. It is change’s most reliable diagnostic tool.
Systems push back because they are built to endure. Every loop, every norm, every incentive structure is doing its job — preserving function against disruption. When a change effort encounters that resistance, it has encountered the system’s operating logic. The six steps above are how you read it: write down who loses and when, draw the loops holding the current state in place, count what each team can absorb, put names against the beliefs, probe at a scale you can afford to be wrong at, and report back to the people paying for it.
The most persistent failure in organizational change is treating resistance as a willingness problem when it is a structural one. The leaders who produce lasting change are not the ones who overwhelm resistance. They are the ones who read it accurately, and let it shape a better path.
Course Index
- Module 0: Introduction to Systems Thinking
- Module 1: Components of Systems
- Module 2: Feedback Loops and Causality
- Module 3: Mental Models and Paradigms
- Module 4: Leverage Points and Change
- Module 5: Systems Archetypes
- Module 6: Applying Systems Thinking to Your World
