Define the organization’s convictions, framework, evidence standard, governed definitions, and non-negotiables before adapting measurement to a funder.
In short: Build an Organization Evidence Model by recording five connected layers: the change your organization is committed to, the framework explaining how that change happens, the evidence standard used to judge claims, the governed definitions that keep evidence comparable, and the rules for aligning with funders or external standards. The result is not another strategy document. It is an organization-owned evidence position that teams can use in proposals, delivery, learning, and reporting.
In short: An Organization Evidence Model is the governed expression of what an organization believes should change, how its work may contribute, what evidence it will trust, and which definitions or ethical boundaries it will not rewrite for each funding opportunity.
A mission statement explains why the organization exists. A Theory of Change, Logic Model, Logframe, or Results Framework organizes its causal logic. A data dictionary defines the evidence fields. The Organization Evidence Model connects these artifacts so that the same organizational position survives across programs, people, systems, proposals, and reports.
This is especially important for organizations working across sites, partners, programs, or donors. Without an explicit model, each application becomes an opportunity to rename outcomes, change denominators, accept a new reporting category, or overstate what the available evidence can prove. The drift often becomes visible only when numbers cannot be reconciled at reporting time.
In short: Each layer answers a different question and produces a different governed artifact. Skipping a layer creates predictable confusion later.
| Layer | Question | Output |
|---|---|---|
| Conviction | What change matters, what have we learned, and what will we not distort? | Mission commitments, field beliefs, hypotheses, cautions, and non-negotiables |
| Framework | How could activities contribute to change? | Outcomes, pathways, assumptions, external factors, risks, and contribution boundaries |
| Evidence standard | What evidence is credible enough for this decision? | Decision-specific rules for quantitative, qualitative, longitudinal, comparative, and participant evidence |
| Dictionary | What does every concept, category, and calculation mean? | Stable definitions, formulas, sources, timing, disaggregation, ownership, permissions, and versions |
| Alignment | Where do our model, the funder’s context, and a standard genuinely match? | A sourced crosswalk of matches, translations, extensions, gaps, conflicts, and non-negotiables |
In short: Separate what the organization values from what it has demonstrated. Preserve both, but label them differently.
Ask five questions in plain language:
The last distinction protects integrity. “We believe accessible employment should include job quality” is a conviction. “Our program improves job quality” is a claim requiring evidence. “Participants retained jobs for 90 days” is a finding only when its population, source, missingness, and calculation are defined.
In short: Keep the framework the organization actually uses, then crosswalk equivalent elements. Do not rename everything into one format before checking whether the meanings match.
A Theory of Change may emphasize pathways and assumptions. A Logic Model may organize inputs, activities, outputs, and outcomes. A Logframe may connect objectives, indicators, verification sources, and assumptions. A Results Framework may organize results and indicators at several levels. The model should govern at least:
If several teams or partners use different frameworks, record the correspondence and the unresolved differences. A crosswalk is more honest than declaring that two boxes are equivalent because their labels sound similar.
In short: An evidence standard defines what is credible enough for a particular use. Operational learning, participant support, funder reporting, board oversight, public communication, and causal claims do not require identical evidence thresholds.
For a weekly service decision, a recent case note and current attendance record may be sufficient. A public causal claim requires a much stronger design. A participant story can explain mechanism or context, but should not be presented as prevalence without a defensible sampling method.
In short: The dictionary converts organizational intent into fields that can be collected and interpreted consistently. It must govern qualitative concepts and longitudinal links as well as numeric indicators.
For every approved concept, record a stable ID, canonical name, plain-language definition, concept type, framework position, calculation or interpretation rule, evidence source, method, disaggregation, collection timing, owner, permissions, quality rule, limitations, version, and external mappings. The implementation details belong in the next Define lessons; here, the essential point is that the dictionary carries the organization’s model into actual workflows.
Illustrative composite example based on recurring workforce-program conditions. A funder counts a placement when a participant starts any paid job. The organization’s program model treats success as accessible, adequately paid work retained for 90 days. A delivery partner submits only a monthly aggregate placement count. Participant interviews describe transportation, schedule, and disability-access barriers that the aggregate cannot show.
| Model layer | Employment example | Decision protected |
|---|---|---|
| Conviction | A placement is not sufficient if the work is inaccessible or unstable. | Do not equate start date with durable outcome. |
| Framework | Placement → job quality → 90-day retention, with accessibility and transport assumptions. | Track the pathway, not one terminal count. |
| Evidence standard | Verified start, wage/quality fields, 90-day follow-up, and participant explanation of barriers. | Combine outcome level with mechanism and context. |
| Dictionary | placement_start, job_quality, employment_90d, access_barrier_theme | Keep each construct and missing state distinct. |
| Alignment | Report the funder’s placement count as an output view; retain quality and retention as organizational outcomes. | Meet a valid requirement without weakening the core model. |
In short: External standards belong after the organization defines its own outcome, population, unit, period, boundary, and calculation. A related label is not automatically an exact mapping.
The UN maintains the official SDG indicator framework. IRIS+ provides qualitative and quantitative metrics for impact investors through its catalog of metrics. GRI uses Universal, Sector, and Topic Standards to report an organization’s significant impacts; see the official GRI Standards. Each serves a different purpose. Record the exact version and whether the relationship is exact, partial, or contextual.
Using only the supplied strategy, program, framework, interview, learning, and governance sources, draft an Organization Evidence Model. Return: (1) mission commitments and field convictions; (2) intended outcomes and causal assumptions; (3) evidence rules by decision; (4) candidate governed concepts and definitions; (5) non-negotiables; (6) contradictions and missing information; and (7) human decisions required. Cite the source, date, and location for every populated statement. Label each as documented policy, observed practice, hypothesis, disputed interpretation, or missing information. Do not convert an aspiration into a demonstrated outcome, infer participant experience, or resolve a framework conflict without an owner’s decision.
A team can build the five layers manually with a source register, framework crosswalk, evidence-policy document, and data-dictionary spreadsheet. The difficulty begins when programs, partners, funders, and versions multiply.
Sopact Sense can centralize the approved sources, extract candidate definitions with citations, keep framework and dictionary records linked, surface conflicting language, and generate different proposal or report views from the same governed model. People still approve convictions, causal assumptions, definitions, standards mappings, access rules, and consequential decisions. AI assists with retrieval and comparison; it does not establish truth or organizational authority.
State organizational convictions and non-negotiables; choose or reconcile the framework of change; define evidence thresholds for real decisions; encode approved concepts in a governed data dictionary; and crosswalk the model to funder context and external standards. Preserve sources, owners, versions, disagreements, and missing information throughout.
No. A Theory of Change is one possible framework for explaining how change may happen. The Organization Evidence Model also governs what counts as evidence, how concepts are defined in data, how versions are controlled, and how funder or standards mappings are handled.
Yes. Use the framework that fits the organization’s decisions and obligations. The model can contain a Theory of Change, Logic Model, Logframe, Results Framework, or a crosswalk among several frameworks. The important requirement is that outcomes, assumptions, evidence, and definitions remain explicit.
Only when the underlying construct, population, timing, calculation, and use genuinely align. Otherwise retain the organization definition and record the funder measure as a translated view, valid extension, or conflict requiring discussion.
No. The framework explains intended change; the dictionary governs the fields used to observe and discuss it. A dictionary without outcome logic can make inconsistent data cleaner without making the evidence more meaningful.
AI can inventory documents, extract candidate statements, compare frameworks, draft dictionary records, and flag contradictions. People must decide what the organization believes, which causal claims are defensible, what evidence is sufficient, and which ethical boundaries or definitions are non-negotiable.
Review it when strategy, program design, population, evidence, regulation, standards, or funding commitments materially change. Version the model rather than overwriting history, and record which definitions and rules applied to each reporting period.
ChatGPT, Claude, and Gemini are fine for a quick test — but not for an answer you'll put in front of a funder or board. When it has to hold up, run it in Sopact Sense.
Build the evidence model in Sopact →
