A Theory of Change is an explicit, testable account of how and why actions are expected to contribute to outcomes in a particular context. It links a desired impact backwards to preconditions, mechanisms, assumptions, evidence and indicators.
In one minute
A usable Theory of Change distinguishes:
- impact: the longer-term condition to which the programme intends to contribute;
- outcomes: changes in behaviour, capability, access, performance or conditions;
- outputs: direct products and services delivered;
- activities and inputs: what the programme does and uses;
- causal links and mechanisms: why one condition is expected to influence another;
- assumptions and external factors: what must hold and what the programme does not control;
- indicators and evidence: how progress, failure, side effects and rival explanations will be examined.
The arrows are claims, not decoration. Each important link should be discussable, evidenced and revisable.
Best for: designing or evaluating a programme whose outcomes depend on behaviour, context and multiple actors.
Avoid when: the work is a simple delivery checklist, the desired outcome is imposed without affected-group participation, or the diagram is used to promise attribution.
The problem it addresses
Programmes often count activity—sessions delivered, systems launched, users enrolled—while assuming outcomes will follow. Theory of Change surfaces the missing causal steps. It asks who must change what, through which mechanism, under which conditions and what else could explain the result.
The method does not prove that the programme caused an outcome. It guides design and evaluation; causal inference still depends on the evidence strategy and context.
When to use it
Use Theory of Change when:
- a programme has several outcomes and causal stages;
- success depends on adoption, institutions, partners or context;
- funders, delivery teams and affected groups hold different assumptions;
- indicators focus on outputs rather than meaningful change;
- evaluation questions and data collection need a causal structure;
- an existing intervention must adapt without losing its intent.
When not to use it
Do not use Theory of Change:
- to reverse-engineer a justification for a fixed solution;
- to imply a linear world when feedback and adaptation dominate;
- to treat stakeholder consensus as causal evidence;
- to omit harms, distributional effects or people excluded from the pathway;
- to claim attribution from a completed diagram;
- to create a dense “everything map” with no decision use.
Inputs required
- a bounded population, geography, time horizon and decision use;
- affected-group perspectives and stakeholder power analysis;
- a precise impact ambition and non-negotiable safeguards;
- evidence on mechanisms, barriers and contextual conditions;
- known alternatives and external contributors;
- evaluation capacity, indicators and data governance;
- an owner and cadence for revising the theory.
Step-by-step process
1. Define the purpose and boundary
State whether the theory supports design, funding, implementation, evaluation or adaptation. Define population, place, horizon and exclusions.
2. Describe the impact and safeguards
Write the longer-term condition as a contribution, not a promise. Add outcomes the programme must not worsen, including access, burden, rights and environmental effects.
3. Work backwards to outcomes
Ask which prior conditions must exist for the impact, then which earlier outcomes enable them. Keep behaviour and system conditions observable.
4. Add outputs, activities and inputs
Connect what the programme controls directly to the earliest outcomes. Do not label attendance or deployment as an outcome when it is an output.
5. Explain causal mechanisms
For each important arrow, state why the change should occur for whom and under what circumstances. Cite evidence and uncertainty.
6. Surface assumptions and context
Record participation, trust, incentives, capacity, policy, infrastructure and partner behaviour that must hold. Name external factors and power asymmetries.
7. Test rival pathways
Ask what else could produce the outcome, where the chain could fail and whether feedback loops alter the sequence. Remove arrows that cannot be explained.
8. Design indicators and evidence
Choose a small set of output, outcome, mechanism, equity and unintended-effect indicators. Define source, denominator, timing and responsible owner.
9. Set learning decisions
For each critical assumption, define the evidence that would continue, adapt, stop or redesign the intervention. Schedule theory reviews, not only performance reports.
AI automation lens
AI can trace claims to cited evidence, flag missing assumptions, compare pathway versions and summarise authorised qualitative evidence.
It must not:
- infer community needs from proxy data without participation;
- fabricate causal links or evidence strength;
- optimise indicators while hiding distributional harm;
- collapse contested values into one generated impact statement;
- claim programme attribution from correlation;
- decide acceptable harm or consent.
Visual model
Text alternative: the pathway is built backwards from impact through outcomes to outputs, activities and inputs; context, assumptions, evidence and alternatives test each important link.
Interactive example
Scenario
A public-interest organisation launches an AI benefits adviser. Its draft theory says: “deploy chatbot → improve household security.” It counts sessions but has no mechanism, accessibility condition or evidence of completed claims.
Worked answer
Work backwards from secure access to eligible benefits. Required outcomes include understanding, trust, correct eligibility guidance, completed applications and timely decisions. Outputs include accessible explanations and human escalation; deployment is an activity. Critical assumptions include digital access, language coverage, current policy rules and agency processing capacity. Measure completion, accuracy, time, appeal, drop-off and distribution across groups; preserve a non-digital route.
Facilitation notes
- Invite affected groups before finalising outcomes and assumptions.
- Build backwards to reduce activity-first thinking.
- Use a separate assumptions register when the diagram becomes dense.
- Distinguish contribution from attribution.
- Mark evidence confidence and contested values visibly.
- Pair the pathway with feedback-loop analysis when dynamics matter.
Expected output
- a bounded impact and outcome pathway;
- outputs, activities and inputs clearly distinguished;
- causal mechanisms for critical links;
- an assumptions and external-factors register;
- indicators for outcomes, mechanisms, equity and harms;
- rival explanations and evidence gaps;
- continue/adapt/stop decisions and a review cadence.
Common mistakes
- Activity equals outcome. Delivery is not the same as change.
- Arrow without mechanism. Sequence alone does not explain causality.
- Linear certainty. Context and feedback can change the pathway.
- Consensus as evidence. Agreement documents a belief, not its truth.
- Only positive outcomes. Harms and distribution belong in the theory.
- Frozen diagram. A Theory of Change should change when evidence changes.
Quality checklist
- Purpose, population, place, horizon and exclusions are explicit.
- Impact is framed as contribution, not guaranteed attribution.
- Outputs and outcomes are distinguishable.
- Critical arrows name mechanisms and evidence confidence.
- Assumptions, external factors and power are visible.
- Affected groups helped test outcomes and harms.
- Indicators include mechanisms, equity and unintended effects.
- Rival explanations and adaptation decisions are recorded.
Template
| Pathway element | Causal mechanism | Assumption / context | Indicator / source | Rival explanation | Decision trigger / owner |
|---|---|---|---|---|---|
| Input → activity → output → outcome → impact | Why change should occur and for whom | What must hold | Definition, denominator and timing | What else could explain it | Continue / adapt / stop |
Knowledge check
Question: A programme delivered 2,000 training sessions. Where does this usually belong?
Answer: As an output or delivery measure. An outcome would describe a resulting change in capability, behaviour or condition, supported by evidence.
Related tools
References
- HM Treasury. The Magenta Book: Central Government Guidance on Evaluation. Official guidance (opens in a new tab). Accessed 22 September 2026.
- UK Government Analysis Function. Theory of Change Toolkit. Official toolkit (opens in a new tab). Accessed 22 September 2026.
- UK Foreign, Commonwealth & Development Office. UK Aid Connect: Theory of Change Guidance. Guidance PDF (opens in a new tab).
- Breuer, Erica, et al. “Using Theory of Change to Design and Evaluate Public Health Interventions.” Implementation Science 11, 2016. DOI (opens in a new tab).
- Archibald, Thomas, et al. “Critiquing Theories of Change.” Canadian Journal of Program Evaluation, 2016. DOI (opens in a new tab). Highlights quality and use limitations.
Method profile
- Primary output: a testable causal pathway, assumptions register and learning-oriented evidence plan.
- Decision level: programme, policy, portfolio or ecosystem intervention.
- Evidence strength: widely used evaluation practice with methodological literature; quality and causal usefulness vary substantially with process and evidence.
- Review trigger: failed assumption, changed context, differentiated harm, new causal evidence or a scheduled evaluation decision.