How Education Works | Education Policy Pilots & Scaling

HEW-NODE-0052

How Education Works → Improvement Mechanics → Education Policy Pilots & Scaling

An education reform can work beautifully in twelve schools and fail badly in twelve thousand.

Nothing mysterious has to happen.

The pilot may have had unusually motivated leaders, extra coaching, a small implementation team, direct access to researchers, generous temporary funding and enough attention to solve every problem by hand. At scale, those hidden supports disappear. Procurement takes longer. Training becomes thinner. data arrive later. Teachers interpret the reform differently. Regional offices adapt it. Political timelines compress rollout. The programme keeps its name while losing the operating conditions that made it work.

This article studies that gap between promising idea and reliable system practice.

It is adjacent to, but does not replace, School Evaluation & External Review, District & Regional Education Offices, or Education Costing. Evaluation asks whether quality is strong. Middle-tier offices translate policy into local support. Costing converts ambition into resource requirements. This node focuses on the change pipeline itself: how an intervention is tested, learned from, adapted, expanded and eventually absorbed into normal system operations.

The Short Answer

Education policy pilots work when they are designed to answer a decision, not merely to demonstrate an idea.

Scaling works when the system understands which parts of the intervention are essential, which can adapt to context, what capabilities and resources expansion requires, how implementation will be supported, how progress will be monitored, and what conditions must be met before the next stage grows.

IIEP-UNESCO’s 2026 implementation framework emphasises four recurring drivers of successful policy implementation: contextual adaptation, stakeholder ownership and engagement, institutional and human capacity, and monitoring, evaluation and learning embedded throughout implementation. OECD implementation work similarly argues that effective change depends on coherent design, stakeholder engagement and an environment capable of sustaining the reform rather than assuming that a policy announcement automatically becomes classroom practice.

A pilot proves that something can work somewhere. Scaling must prove that a system can make it work repeatedly.

1. Start With the Problem, Not the Intervention

“We should introduce tablets” is not a problem statement.

“Students in remote schools cannot reliably access current learning materials, and textbook replacement takes two years” is closer.

Pilots should begin with a defined constraint so the system can later judge whether the intervention actually removed it.

2. A Solution Can Succeed While the Problem Remains

A pilot may deliver every planned workshop and distribute every device without changing the educational outcome that justified the programme.

Outputs are not the same as solved problems.

3. Define the Decision the Pilot Must Inform

Before launch, ask what will happen after the evidence arrives.

Will the programme stop if outcomes do not improve?
Will it expand only if cost stays below a threshold?
Is the test about feasibility rather than impact?
Must it work in rural and urban settings?
Is the question whether teachers can implement it without external coaches?

A pilot without a decision question can continue indefinitely because no result clearly counts as success or failure.

4. Build a Theory of Change Before Building the Programme

A theory of change explains how activities are expected to produce outcomes.

training
→ stronger teacher knowledge
→ changed classroom practice
→ better student learning
→ sustained capability

Each arrow contains assumptions that can fail.

5. Make the Assumptions Visible

Training may assume teachers have planning time. Digital learning may assume electricity and connectivity. A new curriculum may assume textbooks arrive before term begins.

Scaling fails when pilot assumptions are mistaken for universal conditions.

6. Separate the Core From the Adaptable Shell

Every intervention has elements that may be essential and others that can vary.

Core: the mechanism that must remain for the intervention to work.
Adaptable: language, schedule, examples, delivery channel, staffing pattern or local sequence that can change while preserving the mechanism.

Scaling requires knowing the difference.

7. Fidelity Does Not Mean Copying Every Surface Detail

A rural school and a dense urban school may need different schedules. That is not automatically a failure of fidelity.

The better question is whether the adaptation preserves the causal mechanism.

8. Adaptation Without Boundaries Can Dissolve the Reform

If every school changes every component, the system may eventually call many different practices by one programme name.

Then evaluation becomes meaningless because no common intervention remains.

9. Pilot Site Selection Changes What You Learn

Choosing only high-performing volunteer schools answers, “Can this work under favourable conditions?”

That may be a useful first question. It is not the same as, “Can ordinary schools across the system implement this?”

10. Representative Pilots Need Deliberate Variation

If scaling is the goal, the pilot may need different:

school sizes;
geographies;
income contexts;
leadership strength;
teacher experience;
connectivity levels;
languages;
student needs.

Variation reveals where the intervention bends or breaks.

11. Early Pilots Can Intentionally Use Easier Sites

Sometimes the first task is to prove the mechanism before testing difficult contexts.

The mistake is forgetting that the first pilot was deliberately favourable and treating its result as national proof.

12. Document the Pilot Support Package

How much coaching did schools receive? How often did experts visit? How quickly were technical problems solved? Were materials hand-delivered?

These supports are part of the intervention whether or not the project brochure mentions them.

13. Heroic Effort Is a Hidden Subsidy

A pilot manager may personally answer teachers at midnight, rewrite materials on weekends and solve procurement problems through informal relationships.

That can save a pilot. It cannot be the operating model for a national system.

14. Track Labour, Not Only Money

Projects often record purchased materials but undercount staff hours, travel, coordination, data cleaning and management attention.

Scaling needs the full resource footprint.

15. Costing Should Begin During the Pilot

Education Costing should not wait until leaders have already promised expansion.

Unit cost, fixed cost, variable cost and transition cost should be measured while the intervention is still small enough to understand.

16. Pilot Unit Cost Can Be Misleadingly High

Design, research and setup costs are spread across a small number of schools.

Some costs may fall with scale.

17. Pilot Unit Cost Can Also Be Misleadingly Low

Donated software, volunteer expertise, temporary grants and unpaid overtime can make the pilot look cheaper than normal operations will be.

Scaling should remove artificial subsidies from the steady-state estimate unless they are genuinely durable.

18. Economies of Scale Are Not Guaranteed

Printing may become cheaper per book. Coaching may become more expensive if qualified coaches are scarce. Remote regions may add logistics costs.

Each resource behaves differently as volume grows.

19. Diseconomies of Complexity Matter

Ten schools can be coordinated in one messaging group. Ten thousand schools need governance, escalation, reporting, support tiers and reliable logistics.

Scale creates organisational cost even when the intervention itself is unchanged.

20. Implementation Capacity Is a Real Resource

Policies consume attention from ministry staff, regional officers, principals, teachers, finance teams, procurement teams and data staff.

A system running ten major reforms at once may have enough money and not enough implementation capacity.

21. Reform Load Can Overwhelm Schools

A school may simultaneously face a new curriculum, assessment reform, digital platform, safeguarding procedure and reporting requirement.

Each reform can be sensible alone and collectively impossible.

22. Sequence Reforms Around Dependencies

Teacher training before materials arrive may decay before teachers can use it. A data system launched before identifiers are cleaned creates duplicate records faster.

Implementation plans should follow dependency order.

23. Procurement Can Become the Critical Path

A reform may be pedagogically ready and operationally blocked because devices, books, furniture or services cannot be purchased on time.

Education Procurement belongs inside scale planning from the beginning.

24. Supply Chains Change the Intervention

A literacy programme with books is not the same intervention when half the schools receive books six months late.

Implementation fidelity includes availability of required inputs.

25. Teacher Training Is Often the Largest Human Scaling Challenge

The pilot may train 80 teachers directly with expert facilitators. National scale may require 80,000.

The training model itself must therefore be scalable.

26. Cascade Training Can Dilute Meaning

Expert trains trainer, trainer trains regional trainer, regional trainer trains school trainer, school trainer trains colleagues.

At each layer, content can simplify, distort or lose the reasoning behind the practice.

27. Training Needs Practice and Feedback

A presentation about a new pedagogy is not equivalent to being able to perform it in a classroom.

Teacher Professional Learning should include modelling, rehearsal, feedback and follow-up where the reform requires changed practice.

28. Coaching Is Powerful and Expensive

Intensive coaching can support implementation quality. It also requires skilled people, travel time and manageable caseloads.

Before scaling a coaching-heavy model, the system should know where enough capable coaches will come from.

29. Build the Support Workforce Before the Rollout Wave

If every school adopts the programme on the same date but the support system is still being recruited, early problems become permanent habits.

Support capacity should lead demand, not chase it.

30. The Middle Tier Is Often the Scaling Engine

District & Regional Education Offices can translate central policy into local support, aggregate problems, coordinate training and identify where implementation is drifting.

A reform that bypasses the middle tier during the pilot may later discover that national scale depends on it.

31. Local Ownership Is Not a Public-Relations Exercise

Teachers and school leaders often know which parts of a reform collide with timetable, workload, language, facilities or student needs.

Engagement improves implementation when it changes design, not merely when stakeholders are informed.

32. Resistance Can Contain Operational Information

Not every objection is correct. But repeated resistance may signal workload, incentive or feasibility problems that pilot teams missed.

Implementation teams should distinguish unwillingness from unworkability.

33. Incentives Can Distort Pilot Evidence

Pilot schools may receive grants, equipment, recognition or additional staff that disappear at scale.

The intervention may be accepted partly because of the package around it.

34. Measure Implementation, Not Only Outcomes

If learning outcomes do not improve, the system needs to know whether the theory was wrong or the programme was never actually implemented.

Both questions matter.

35. Implementation Indicators Should Track the Mechanism

training completed;
materials available;
practice used at expected frequency;
coaching delivered;
student participation;
required timetable time protected;
technology functioning;
adaptations documented.

Do not measure every possible activity. Measure the parts needed for the theory to operate.

36. Outcome Measures Need Baselines

“Scores rose to 68” means little if we do not know where they started or what happened in comparable schools.

Evaluation design should be agreed before results are known.

37. Not Every Pilot Needs a Randomised Trial

A feasibility pilot may ask whether schools can schedule the programme or whether data can be collected reliably.

Evaluation method should match the decision question.

38. Strong Causal Claims Need Stronger Designs

If leaders want to claim that the intervention caused an outcome, they need a design capable of separating the intervention from other changes.

That may require comparison groups, randomisation, quasi-experimental methods or other credible approaches depending on context.

39. Quantitative Results Need Implementation Context

An average effect can hide that urban schools improved while remote schools could not operate the model.

Scaling decisions need to know where, for whom and under what conditions the intervention worked.

40. Qualitative Evidence Finds Mechanisms and Friction

Interviews, observations and implementation logs can reveal why teachers changed the programme, why students stopped attending, or why a technically successful tool was abandoned.

Numbers tell us what pattern occurred. Process evidence helps explain the pattern.

41. Negative Results Are Valuable Assets

A well-run pilot that shows an idea does not work can save a system from an expensive national mistake.

Failure should not be hidden merely because the project was politically visible.

42. Stopping Rules Protect Resources

Before rollout, define conditions that trigger pause, redesign or termination.

safety failure;
unacceptable cost;
low adoption;
no evidence of the intended mechanism;
severe equity effects;
critical infrastructure unavailable.

Without stopping rules, programmes can survive because institutions become attached to them.

43. Expansion Should Use Gates, Not Hope

Instead of pilot → national rollout, use stages.

prototype
→ feasibility pilot
→ diverse-site pilot
→ controlled expansion
→ regional scale
→ national scale
→ institutionalised operations

Each gate asks whether evidence and capacity justify the next increase in exposure.

44. Stage Gates Need Explicit Criteria

A gate might require:

minimum implementation quality;
acceptable unit cost;
positive or sufficiently promising outcomes;
no major safety issue;
training capacity ready;
supply chain proven;
data collection functioning;
budget authority confirmed.

The next stage should not begin because the calendar says so.

45. Controlled Expansion Creates Learning at Larger Scale

Moving from 20 schools to 200 may reveal coordination, procurement and support problems that never existed at 20.

The expansion stage is itself another experiment in system capacity.

46. Scale Readiness Is Different From Programme Effectiveness

An intervention can be educationally effective and operationally unready for expansion.

Scale readiness asks whether the delivery system exists.

47. A Scale-Readiness Review Should Examine the Whole Chain

policy authority;
budget;
staffing;
training;
materials;
procurement;
data;
support;
leadership;
communications;
monitoring;
equity;
maintenance;
long-term ownership.

A missing dependency can become the national bottleneck.

48. Budget Approval Is Not the Same as Cash Availability

A ministry may have an approved budget while schools cannot access funds at the time they need to act.

Implementation needs the timing of money, not only the annual total.

49. Funding Rules May Need to Change Before Scale

If a reform requires recurring software licences, additional counsellors or school-level materials, ordinary funding formulas may not recognise those costs.

Temporary project grants cannot sustain a permanent reform.

50. Procurement Frameworks May Need to Precede Rollout

Thousands of schools individually buying different versions of the same service can create price variation, incompatibility and weak support.

Central or framework procurement may be appropriate where standardisation creates value.

51. Standardisation Has a Limit

One national product can simplify support but fit some contexts badly.

Scale architecture should standardise what benefits from sameness and preserve choice where context genuinely matters.

52. Data Systems Should Be Ready Before the Reform Needs Them

If leaders promise real-time monitoring but Education Management Information Systems receive data six months late, the governance model is fictional.

Monitoring design must match actual data capability.

53. Do Not Build a Parallel Reporting Empire

Pilots often create custom spreadsheets, dashboards and reporting teams.

At scale, parallel systems multiply teacher workload and fragment data. Mature reforms should migrate essential measures into normal system reporting where possible.

54. Institutionalisation Means the Programme Stops Being a Project

A reform is institutionalised when ordinary structures own it:

normal budgets;
job descriptions;
training systems;
procurement cycles;
data systems;
school routines;
accountability mechanisms.

If the programme disappears when the project office closes, it never truly entered the system.

55. Ownership Must Move Deliberately

Pilot teams often know the intervention better than the permanent ministry unit that will inherit it.

Handover should include procedures, rationale, data, unresolved risks, supplier knowledge and support responsibilities.

56. Knowledge Handover Is Part of Scaling

Do not transfer only documents.

Transfer the reasoning behind decisions, common failure modes, exception handling and what the pilot team learned informally.

57. Leadership Turnover Tests Institutionalisation

If a reform collapses when one minister, director or principal leaves, the system depended on a person rather than an institution.

Durable reforms survive ordinary leadership change.

58. Political Visibility Can Compress Learning Cycles

Leaders may want national rollout before a pilot has produced reliable evidence.

The larger the exposure, the more expensive it becomes to discover basic design flaws late.

59. Slow Is Not Always Better Either

A strong intervention can remain trapped in perpetual pilot mode while generations of students never receive it.

The goal is not maximum caution. It is evidence-proportional expansion.

60. Crisis Can Force Rapid Scaling

School closures, displacement or public-health emergencies may require new delivery models immediately.

In rapid scale, systems can still use feedback loops: deploy, observe, repair, standardise and expand iteratively rather than pretending the first version is final.

61. Adaptive Implementation Is Controlled Learning During Delivery

IIEP-UNESCO’s recent implementation work emphasises adaptation because education systems operate in changing political, institutional and local contexts.

Adaptation should be documented: what changed, why, who approved it, and whether the core mechanism remained intact.

62. Feedback Loops Need a Destination

Teachers can report problems every week, but if no team has authority to change materials, guidance or support, feedback is merely collected.

Learning systems require a decision owner.

63. Version the Reform

If guidance changes, schools should know which version is current and what changed.

Version control prevents different regions from implementing different historical editions unknowingly.

64. Change Logs Preserve Institutional Memory

A change log can record:

problem observed;
evidence;
decision;
version changed;
expected effect;
date;
owner.

Future leaders can then see why the reform looks the way it does.

65. Equity Can Change at Scale

A digital intervention may work well in connected pilot schools and widen inequality when expanded to places with weaker devices, electricity or home support.

Scale reviews should test who gains, who waits and who bears new burdens.

66. Average Success Can Hide Exclusion

If 90% of students benefit while a small group with disabilities loses access, the average can look excellent.

Distribution matters.

67. Regional Variation Can Be Information, Not Noise

If one district consistently implements better than another, study leadership, support, staffing, transport, language and workload differences.

Variation can reveal the conditions for success.

68. Benchmarking Should Improve the System, Not Shame Schools

Publishing simplistic league tables can encourage gaming or avoidance.

Implementation comparisons are most useful when they help locate support needs and transferable practices.

69. Common Failure Mode: Pilot the Best Schools, Scale to Every School

The programme succeeds under unusually favourable conditions and is assumed to be universally ready.

Repair: test progressively more representative and constrained settings before full scale.

70. Common Failure Mode: Scale the Name, Lose the Mechanism

Training shortens, materials change, support disappears, but the programme keeps the original label.

Repair: define core components and monitor whether they survive expansion.

71. Common Failure Mode: Promise Scale Before Costing It

Political commitment precedes realistic estimates of training, support, procurement and recurring costs.

Repair: cost the steady-state operating model before irreversible rollout commitments.

72. Common Failure Mode: Treat Training as Distribution

Slides are delivered to thousands of teachers and counted as implementation.

Repair: measure capability and classroom practice, not attendance certificates alone.

73. Common Failure Mode: Build a Project Data System That Dies With the Project

The pilot dashboard works beautifully because a special team manually cleans data. National systems never absorb it.

Repair: migrate essential monitoring into sustainable system data processes before project closure.

74. Common Failure Mode: Collect Feedback Without Authority to Adapt

Schools report the same obstacle for months while central guidance remains frozen.

Repair: establish a governed change process with decision rights and version control.

75. Common Failure Mode: Never Stop Piloting

The programme receives repeated extensions because leaders want more evidence but never define what evidence would be enough.

Repair: define scale, redesign and stop thresholds before the pilot begins.

76. A Strong Pilot-to-Scale Operating Model

Define the educational problem.
Define the decision the pilot must inform.
Build the theory of change.
Identify core and adaptable components.
Choose sites appropriate to the learning stage.
Record the full support package.
Measure cost and staff time.
Measure implementation as well as outcomes.
Collect qualitative evidence on friction and adaptation.
Test equity effects.
Define stop, redesign and scale thresholds.
Expand in stages.
Build training, procurement, data and support capacity before each wave.
Use regional and district structures deliberately.
Version changes and document adaptations.
Move the reform into normal budgets and routines.
Continue monitoring after institutionalisation.

77. A Scale-Readiness Gate

Is the problem still important at scale?
Is the causal mechanism plausible and evidenced?
Are core components explicit?
Has the intervention worked in sufficiently varied contexts?
Is implementation quality measurable?
Is steady-state cost affordable?
Is budget timing workable?
Can required materials be procured and delivered?
Can enough people be trained and supported?
Are data systems ready?
Are equity risks understood?
Does the middle tier have capacity?
Are legal and policy authorities in place?
Is there a feedback and adaptation process?
Who will own the programme after project teams leave?

78. A Scaling Dashboard

sites activated;
sites implementation-ready;
staff trained;
staff demonstrating required capability;
materials delivered on time;
core-component fidelity;
documented adaptations;
support cases per school;
coach caseload;
unit cost;
budget release timing;
procurement delays;
student reach;
outcome change;
equity gaps;
regional variation;
critical incidents;
school workload indicators.

79. Worked Example: A Reading Intervention

A reading programme works in twenty primary schools. Teachers receive five days of direct expert training, monthly coaching and complete classroom book sets. Student reading improves.

A rapid national rollout would require 40,000 teachers. There are not enough expert trainers or coaches, and book procurement would take eighteen months.

Instead, the system expands to 200 schools across varied regions. It tests a shorter training model with structured rehearsal, builds a coach-certification pathway, contracts book production early, and measures which support components most strongly predict implementation quality. Only when training fidelity, book delivery and coach capacity meet defined thresholds does the next expansion begin.

The intervention scales because the delivery system scales with it.

80. Worked Example: A Digital Attendance Platform

A digital attendance system works in fifty connected urban schools. Teachers mark attendance quickly and district staff can identify absence patterns the same day.

When expansion reaches remote schools, intermittent connectivity causes failed submissions and duplicate records. The system pauses the next rollout wave, adds offline capture with synchronisation, strengthens learner-identity matching, and tests the revised version in low-connectivity sites.

Adaptation preserves the mechanism—timely trustworthy attendance—rather than preserving the original technical design at all costs.

81. Worked Example: A New Teacher-Mentoring Model

A mentoring pilot improves retention among beginning teachers. In the pilot, experienced mentors support four new teachers each.

National workforce data show there are not enough qualified mentors to maintain that ratio. Rather than dilute the model silently, the system tests a tiered design: intensive mentoring for highest-need entrants, group mentoring for others, protected mentor time and regional specialist support. Outcomes and workload are compared before broader adoption.

Scale changes the resource equation, so the system redesigns deliberately rather than pretending nothing changed.

82. What Good Looks Like

The problem is explicit. The theory of change is visible. Core components are known. Pilot sites are chosen for the question being asked. Hidden support is counted. Costs include labour and transition. Evaluation distinguishes implementation from outcomes. Adaptation is documented. Expansion uses readiness gates. Training capacity exists before rollout. Procurement and data are on the critical path. Regional offices know their role. Equity is monitored. Negative evidence can stop the programme. Feedback changes versions. Project structures gradually disappear as normal systems take ownership.

The reform becomes ordinary without becoming hollow.

83. The World Return

Education systems are full of good ideas.

The difficult part is not discovering that a skilled teacher, committed principal or well-supported pilot team can make something work.

The difficult part is building conditions in which ordinary institutions can reproduce the result without extraordinary rescue.

That is the moment a reform changes category.

It stops being a demonstration and becomes infrastructure.

The classroom practice is supported by training. Training is supported by budgets. Budgets are supported by policy. Materials arrive through procurement. Problems travel through regional offices. Data show where implementation is weak. Feedback changes the next version. Leadership can change without erasing the system.

Scale is not making the pilot bigger.

Scale is rebuilding the conditions for success so they can survive the size of the real world.

The true unit of scaling is not the programme. It is the system’s capacity to reproduce the programme’s essential mechanism.

How Education Works | Education Policy Pilots & Scaling — How Small Trials Become System Change Without Losing What Made Them Work