What is collective impact?
Collective impact is an approach in which organizations coordinate their work around a shared goal for a complex social problem. Partners agree how their different activities contribute, how they will learn together and how progress will be assessed. A partnership might involve public agencies, businesses, community groups, educators or funders.
The approach is associated with five conditions: a common agenda, shared measurement, mutually reinforcing activities, continuous communication and backbone support. John Kania and Mark Kramer described these in their 2011 article. They are conditions for organizing collaboration, not a guarantee of success.
This guide focuses on making shared measurement practical without forcing every partner into an identical survey or exposing individual records unnecessarily. It also explains how the measurement supports the other parts of the partnership.
The five conditions, in practice
Scroll horizontally to see all columns →
| Condition | What partners need to agree | Question to resolve early |
|---|---|---|
| Common agenda | The problem, intended change and broad approach. | Do affected communities and partner organizations mean the same thing by success? |
| Shared measurement | A manageable set of measures and reporting conventions. | Which results can be compared, combined or examined separately? |
| Mutually reinforcing activities | Distinct contributions that support the shared goal. | Where do services overlap, leave gaps or depend on another partner? |
| Continuous communication | Regular opportunities to exchange evidence and resolve differences. | What information is needed before the next decision? |
| Backbone support | Coordination capacity, responsibilities and resources. | Who maintains agreements, follows up and supports partners? |
Shared measurement is important, but it is not the only condition that matters. Relationships, resources, decision authority and community participation affect whether evidence can be used. A technically consistent dashboard cannot resolve an unresolved disagreement about whose priorities count.
Include community influence throughout the work
Subsequent practice has emphasized equity and the role of people affected by the work. The Collective Impact Forum treats equity as integral to the approach, rather than an optional addition after the five conditions are implemented.
Make participation concrete. Who helps choose the questions? Who can challenge an interpretation? How are people supported to participate? Who decides when partners disagree? An invitation to comment is different from sharing decision authority. State the arrangement honestly.
Look beyond averages. A partnership may improve access overall while leaving a particular location or group behind. Plan how to examine those differences without disclosing sensitive information about small groups.
An example: a regional employment partnership
Imagine a regional partnership that wants to improve access to suitable employment. A training provider offers assessed practice, employers explain recruitment needs, a transport organization addresses access barriers and a community group gathers participants’ experiences. This is an illustrative example, not a documented customer result.
The common goal does not require every organization to deliver the same service. Their contributions are different. Their evidence also differs: assessments, interviews, transport records and participant feedback answer different questions.
A useful shared question might be: “Where are people losing access between training and suitable work?” That question is more actionable than asking every partner to upload every available metric. A theory of change can help explain the expected connections and the assumptions behind them.
Choose a small shared core and preserve local detail
Begin with the decisions the partnership will make together. For each, identify the evidence needed and whether it already exists. Select a small shared core; let partners retain additional questions needed for their own work.
In the employment example, shared fields might include reporting period, service type, a defined completion measure and follow-up coverage. A specialist partner may need additional questions about accessibility or learning support. Those local questions need not become compulsory for everyone.
Shared measurement does not necessarily mean one form, one database or one survey for the whole network. It means partners understand the measures they are using and the claims those measures can support. Existing systems may supply the data if their definitions, coverage and timing are suitable.
Scroll horizontally to see all columns →
| Field | Example agreement |
|---|---|
| Measure | Participants completing the agreed training requirements. |
| Unit | Distinct participants within one named program and period. |
| Definition | Completion requires the specified attendance and assessment conditions. |
| Timing | Completion date falls within the reporting quarter. |
| Source | Program completion records with review responsibility identified. |
| Exceptions | Withdrawals, missing assessments and corrections shown separately. |
| Version | The agreed definition and date it took effect. |
If partners use different completion requirements, the figures may describe different things. Preserve the difference or agree a valid common measure. Relabeling columns does not make the underlying results comparable.
Decide when individual linkage is needed
Use individual linkage when the question genuinely concerns the same person across services or time, and when the necessary permissions and arrangements support it. For example, examining a participant’s transition from training to employment may require matched observations.
Other questions can be answered using partner-level or population-level evidence. A partnership reviewing service availability across districts may not need names or individual histories. Anonymous resident feedback should not be quietly joined to identifiable records.
When linking is justified, define the matching method, who can access identifying information, how ambiguous matches are reviewed and how errors are corrected. Do not merge people merely because their names are similar. Record unmatched cases rather than forcing a complete-looking dataset.
Worked example: when two totals cannot simply be added
Suppose Partner A reports 80 distinct participants and Partner B reports 60 during the same quarter. If 20 people used both services, the combined number of distinct people is 80 + 60 − 20 = 120. The sum of partner counts, 140, describes participation across the two services, not 140 different people.
That calculation depends on knowing the overlap. If appropriate linking is unavailable, report the 140 service-participant records and explain that the count is not deduplicated across partners. Do not invent an overlap estimate to produce a more attractive total.
For percentages, combine compatible numerators and denominators rather than averaging partner percentages without considering their meaning. If one partner observes employment at 30 days and another at six months, keep those results separate unless a justified comparison is available.
The same caution applies to qualitative evidence. Twenty mentions of a barrier are not automatically twenty people, and a theme found in one partner’s interviews may not represent the whole region.
Plan a reporting cycle partners can maintain
- Confirm the question and period. Send the definitions and explain what decision the return supports.
- Collect the agreed core. Reuse reliable registration information and existing records where appropriate; request updates when relevant.
- Retain context. Keep the partner, source, date, question version and relevant supporting documents.
- Review exceptions. Identify missing periods, changed definitions, duplicate records and conflicting explanations.
- Interpret together. Ask partners and affected groups what the pattern might mean.
- Record the response. Name the action, owner and next review, including unresolved questions.
Choose the cadence around the work. An operational issue may need a quick response, while an employment outcome may need months to become observable. “Continuous communication” does not mean every measure must update in real time.
What the backbone team should own
The coordination team needs capacity to maintain agreements, answer questions and support partners with uneven resources. Collecting a spreadsheet is only part of that responsibility. Someone must resolve definition changes, explain missing evidence and prepare findings for joint review.
Keep a change record for the dictionary and reporting rules. If an indicator changes, preserve the earlier definition and explain whether historical results can still be compared. Give partners a clear route for correcting a submitted figure.
Access should follow purpose. A central reviewer may need an approved aggregate and source documentation without needing every participant’s personal information. Make ownership, export access and review responsibilities explicit before expanding collection.
What software can support—and what partners must decide
Software can help organize recurring returns, preserve source context, identify exceptions and make reviewed results easier to revisit. AI assistance may help classify narrative evidence or locate passages. Reviewers still need to check meaning, contradictory accounts and whether a calculation uses compatible records.
Sopact is relevant when partners contribute varied responses and documents that need shared definitions, analysis and governed reporting. Test a small workflow using realistic submissions. Include a changed question, a missing return and a correction, rather than evaluating only a polished dashboard.
A system does not create agreement, prove causality or make a partnership equitable. Those responsibilities remain with the people governing the work. For a practical collection plan, use the Membership & networks course; for evaluating results, use Impact Measurement & Reporting.
Report what the partnership knows and will do next
Separate partner activity, observed outcomes and claims about the partnership’s contribution. Explain definitions, overlap, coverage, timing and important limitations. Include adverse findings and what partners disagree about. Make the report useful to contributors and affected communities as well as funders.
Use How to Write an Impact Report and the report examples for the reporting step. For place-based assessment, continue to community impact.
Watch: distinguish delivery from outcomes
This companion explains the difference between what partners deliver and the changes they seek. Apply it alongside the shared definitions and aggregation checks above.
Frequently asked questions
Does collective impact require one shared database?
No. Partners need workable agreements on measurement and learning. The technical arrangement should fit the questions, permissions and capacity of the partnership.
Must every partner use the same survey?
No. Agree the common measures needed for shared decisions and preserve useful local questions. Check meaning and timing before comparing results.
Is shared measurement the only reason collective impact succeeds?
No. Coordination, relationships, resources, community influence and the other conditions matter. Measurement supports decisions but cannot guarantee results.
Can we add all partner participant counts?
Only if the intended total and overlap are clear. Summed counts may describe service participation rather than distinct people across the partnership.
How is collective impact different from community impact?
Collective impact describes an approach to coordinated work. Community impact describes effects on a community. A partnership may pursue community impact through collective work.
Choose your next step
ASME
Related network practice: gathering stakeholder evidence. This story does not establish a collective-impact outcome.
Read the customer story →