What is portfolio analytics for impact?
Portfolio analytics for impact examines results across a group of funded organizations, programs or investments to understand patterns, differences and progress toward intended outcomes. It combines appropriate quantitative measures with qualitative evidence and the context needed to interpret them.
For a foundation or impact fund, the questions might be which partners are progressing against their plans, where outcomes differ between groups, and what support or further inquiry is needed. Those questions require more than a total of funds committed or people reached.
This guide focuses on impact and partner evidence. It is not a guide to securities performance attribution, market-risk models or a particular financial analytics product. The aim is to make portfolio conclusions useful and inspectable without claiming that every measure can be combined.
Choose the question before choosing the aggregate
Start with a decision or learning question. “How many people completed training?” calls for different data from “Which partners need support?” or “What changed for the people who completed it?” A single score cannot answer all three.
Distinguish portfolio data management, monitoring and analytics. Data management maintains records and definitions. Monitoring compares progress with the plan. Analytics investigates patterns and relationships across compatible records. They share inputs, but they are different tasks.
- Describe: what results were reported, by whom and for which period?
- Compare: how do results differ across relevant groups or stages?
- Explain: what evidence may account for the pattern, including alternative explanations?
- Act: what should the team review, change or collect next?
Use the portfolio data management guide to prepare the records and the portfolio monitoring software guide to evaluate ongoing review workflows.
Keep four kinds of information distinct and connected
A practical portfolio analysis often needs the following four groups of information. This is an organizing approach for the analysis, not a reporting standard or a requirement to force all fields into one table.
On a narrow screen, scroll the table horizontally to read all columns.
| Information | Examples | Interpretation boundary |
|---|---|---|
| Resources and financial context | Approved funding, disbursements and reported expenditure. | Money spent is not an outcome. |
| Delivery and outputs | Services delivered, completions and milestones. | Delivery does not establish sustained change. |
| Outcomes and experience | Follow-up measures, participant feedback and relevant changes. | Observed change is not automatically caused by the program. |
| Evidence and interpretation | Sources, definitions, time periods, coverage, corrections and limitations. | A clean total is only meaningful within these boundaries. |
Connect each observation to the right organization, award or vehicle and reporting period. One partner may have two awards. One person may participate in more than one program. A stable ID helps with matching, but it does not remove the need for overlap rules or appropriate permissions.
Impact Frontiers’ five dimensions offer a complementary way to examine what outcome occurs, who experiences it, its scale and duration, contribution and risk. They help frame the questions; they do not justify collapsing every dimension into one unqualified score.
Worked example: the average rate is not always the combined rate
Illustrative example. Partner A reports eight participants meeting an agreed outcome among ten followed up: 80%. Partner B reports 18 among 90 followed up: 20%. Assume the outcome, follow-up window and counting rules are compatible and the groups do not overlap.
- Partner A8 of 10 followed up = 80%
- Partner B18 of 90 followed up = 20%
- Average partner rate(80% + 20%) ÷ 2 = 50%
- Combined participant rate(8 + 18) ÷ (10 + 90) = 26%
Illustrative data. The first average gives each partner equal weight; the combined rate weights by the number followed up.
If the question is what share of all followed-up participants met the outcome, report 26%, with the denominator of 100. If the question is the average of the two partner rates, 50% is correct, but label it accordingly. Do not call one the other.
Now suppose 40 additional participants were eligible for follow-up but did not respond. The observed rate remains 26 of 100 respondents. It is not a verified outcome rate for all 140 eligible people. Show the follow-up coverage and examine whether missing responses differ systematically from those observed.
The low rate for Partner B also deserves context. Its participants may have different starting needs, or its delivery may require improvement. The aggregate alone cannot distinguish those explanations. Keep the comparison visible and investigate before assigning performance labels.
Compare cohorts and vintages at meaningful points in time
A cohort is a group defined by a shared characteristic or starting event, such as participants entering a program in the same quarter. A funding vintage groups investments or awards by their starting period. Define both explicitly before drawing a trend.
Comparing a mature three-year cohort with a newly funded cohort can confuse time available for change with performance. Where relevant, align the comparison by elapsed time: for example, six months after entry rather than a common calendar date alone.
- Matched records: compare the same entities across periods when the question is individual change.
- Repeated snapshots: show when the group’s membership changes between observations.
- Attrition: report how many eligible records lack follow-up and what is known about the gap.
- Definition changes: identify breaks in measurement rather than drawing a continuous trend without explanation.
- Corrections: retain the accepted historical version and the effect on the comparison.
A stronger portfolio average can result from adding a high-performing partner even if no existing partner improves. Check both the changing portfolio and the stable subset when that distinction matters. Neither view is universally superior; they answer different questions.
Use targets and benchmarks without pretending they are the same
A target is the result a partner or funder planned to achieve. A benchmark is a reference for comparison, such as a relevant prior cohort or external dataset. A result can exceed a modest target and still compare poorly with an appropriate benchmark.
Before using an external benchmark, check the population, outcome definition, observation period, geography and collection method. Document material differences. A familiar indicator name or framework label does not establish comparability.
For variance, state the convention. If actual completions are 40 against a target of 50, actual minus target is −10 and the shortfall is 20% of target. If the target is zero, that percentage calculation is undefined; use another meaningful comparison.
Separate financial variance from outcome variance. Underspending can reflect delay, lower cost or a changed plan. It does not, by itself, prove efficiency. Use the delivery and financial explanations together before recommending action.
Read qualitative evidence alongside the pattern
Narrative reports, interviews, site notes and open-ended feedback can help explain why a result changed or why a measure misses something important. Attach the source, date and relevant partner or cohort so a reader can assess the context.
Look for supportive, critical and contradictory evidence. If several partners describe a transport barrier, inspect who raised it, when and in response to which question. A theme mentioned in five reports is not automatically a prevalence estimate across everyone served.
Keep quotations accurate and respect their permitted use. A vivid quotation should not stand in for the entire group, and the absence of a comment does not establish the absence of a problem.
The practical question is whether the narrative changes the interpretation or the next step. If it suggests a new explanation, record it as a hypothesis to examine rather than presenting it as a settled cause.
Can SROI or other impact scores be rolled up across a portfolio?
Not safely by averaging ratios or scores without understanding how they were constructed. Compatible units, populations, valuation assumptions, time horizons and overlap rules matter. Source links are necessary for inspection, but are not sufficient to make two models compatible.
Simplified illustrative arithmetic: one project estimates adjusted social value of 200 against inputs of 100, a ratio of 2:1. Another estimates value of 800 against inputs of 200, a ratio of 4:1. If—and only if—the underlying models are compatible and values are not double-counted, the combined calculation is 1,000 ÷ 300, about 3.33:1. The simple average of 2 and 4 is 3:1 and answers a different question.
In practice, first inspect the valuation, duration, attribution assumptions and possible overlap. A portfolio figure may be inappropriate when those cannot be reconciled. Present the project-level analyses and their limits rather than creating a precise-looking combined number.
For the underlying methodology, see Social Value International’s Guide to SROI. Other composite scores need the same discipline. Retain component measures, explain weights and test how conclusions change under reasonable alternatives. A score can help organize judgment; it should not conceal the underlying evidence or trade-offs.
What AI can help with in portfolio analysis
Configured AI can help extract candidate figures, organize text, identify possible contradictions and draft a synthesis around a defined portfolio question. It works best when the source material, definitions and selected records remain available for review.
Check whether an answer identifies its population, reporting window, calculation and exclusions. Verify material figures and inspect cited passages. A fluent explanation can still use the wrong denominator or miss a changed definition.
Sopact’s portfolio workflow focuses on recurring partner collection and source-connected analysis. Test it with metrics, uploaded documents and narrative evidence from several periods. Confirm the actual supported formats, integrations and permission behavior.
A BI platform or spreadsheet can also support traceable calculations when its model and supporting evidence are designed well. The decision is which setup gives your team a maintainable path from incoming partner evidence to a reviewed answer.
The video below is a broader portfolio-reporting companion with financial-holdings examples. The impact calculations, comparability rules and limitations are in this article.
Evaluate analytics software with a question you can independently check
Use one representative analysis rather than a polished sample dashboard. Include two reporting periods, a revised value, an incompatible measure, an unreadable document and a missing partner. Agree the expected handling before testing the system.
- Team control: can an authorized owner update a definition and retain its version?
- Relationships: can one organization appear in two funds without unintended duplication?
- Coverage: are included, missing, failed and excluded records visible?
- Time: can the model compare matched records and changing cohorts distinctly?
- Qualitative context: can a finding open to supporting and conflicting passages?
- Documents: are sources, versions and access restrictions retained?
- Calculations: can a reviewer inspect numerators, denominators, filters and assumptions?
- Reproduction: can another reviewer rebuild the result from the accepted records?
Inspect the export as well as the screen. Ask what happens when a connector fails, a source document is corrected or a user loses access. Include integration, model maintenance, training and review time in the implementation estimate.
Identical AI wording is not the test of reproducibility. The underlying accepted records, definitions and calculation should be inspectable even if the assistant phrases its explanation differently.
Customer practice: make the framework usable before scaling
Kuramo Foundation’s story describes translating an internal gender-lens framework into an operational data and dashboard approach before launch. It illustrates the preparation needed to make a framework usable; it is not proof of a causal outcome or a particular portfolio return.
For your next analysis, choose one shared outcome and a few varied partners. Document what can be compared, what must remain separate and what evidence is missing. Present the result with its calculation and limitations, then use the review to improve the next collection cycle.
Continue through the Portfolio Intelligence course. For the output, use the impact report writing guide and report examples and dashboards. A useful report makes the reasoning easier to inspect, not merely the number easier to see.
Frequently asked questions
What does portfolio analytics for impact measure?
It examines outcomes, delivery, resources and relevant evidence across funded organizations or programs. The analysis depends on the question and should preserve definitions, periods, coverage and limitations.
How is analytics different from portfolio monitoring?
Monitoring follows progress against the plan. Analytics examines patterns, comparisons and relationships. Both depend on maintained records and clear definitions.
Should we average partner outcome percentages?
Only when the average partner rate is the question. For a combined participant rate, sum compatible numerators and denominators, account for overlap and state who was observed.
How should missing partner data be handled?
Show coverage and the affected measures. Do not automatically substitute zero, assume missing partners resemble reporting partners or describe a partial result as the complete portfolio.
What is a cohort or vintage comparison?
A cohort groups entities by a defined characteristic or entry event; a vintage groups awards or investments by starting period. Compare at appropriate elapsed times and disclose changes in membership or definitions.
Can portfolio analytics prove causation?
No, not by aggregation alone. Observed change and associations need appropriate evidence and analysis before attributing them to a program, enterprise or investor.
Can SROI ratios be combined?
Sometimes, after checking model compatibility, assumptions and double counting. Do not simply average ratios; inspect the underlying adjusted values and inputs, and retain project-level limits.
Does an analytics platform replace BI or finance systems?
Not necessarily. It may improve partner collection and interpretation alongside existing systems. Verify data ownership, integrations, calculations and exports for the intended workflow.
