The impact gap: where impact finance falls short, and who bears the cost

We coded 186 source-checked records of shortfall that a source records, finds or alleges. They were selected because they describe failure, so they show how failure is documented, not how often finance fails. Most record harm that is alleged but that no body has found to have occurred. Few complaints ever receive a ruling, so the absence of a finding is not evidence that no harm occurred.

1. A finding, and how it arose

In February 2022 a complaint about the International Finance Corporation's (IFC) microfinance investments in Cambodia was registered with the Compliance Advisor Ombudsman (CAO), the IFC's independent accountability office. It concerned borrowers of Cambodian banks and microfinance institutions that IFC had financed. In October 2025 the CAO reported its investigation to the IFC Board. It found that IFC failed to comply with its Sustainability Framework during pre-investment due diligence and in supervising its microfinance investments, "contributing to harm to vulnerable borrowers, including Indigenous Peoples". The CAO recorded that borrowers lost land and productive assets and suffered reduced livelihoods, food insecurity, health impacts and disrupted schooling, along with fear and instances of threats and reprisals. IFC Management disagreed that the Sustainability Framework applied. The IFC Board later rejected the finding, according to Human Rights Watch.

IFC lent to financial intermediaries, and the people harmed were borrowers one step further along the chain. According to the case record, IFC considered impacts on microfinance borrowers outside the scope of its framework. The finding concerned a funder's diligence and oversight of lending made through intermediaries.

2. What the evidence base is

The evidence base contains 186 source-checked items. Each is one documented instance, finding or pattern in which a source records, finds or alleges that impact finance, blended finance or development finance delivered less than it intended. "Source-checked" means the quote appears at the linked source and the codes were independently checked. It does not mean the claim is proven. A further 18 items were set aside as not verifiable: 12 because their source could not be reached when the independent check was run, and 6 because the check did not support their coding. We excluded 17 records: 10 as out of scope, 3 as opinion rather than evidence, 2 as superseded by a newer edition and 2 because they reported no shortfall. Items were coded and checked under a written codebook, with independent re-checking and a spot check of 20 items; see the method note.

There are three source types. Complaints number 112: cases lodged with development banks' accountability mechanisms and registered from 1 January 2021. They are a purposive sample from seven mechanisms, not a census. Evaluations number 44: reports by evaluation units, national audit offices and market surveys, using the latest edition and an older one only where no newer exists. News items number 30: reports published between 1 November 2025 and 27 October 2026. Across the evaluation and news types, 21 items come from aligned organisations such as watchdogs and think tanks. By source tier, 112 items come from accountability mechanisms, 33 from evaluators, 21 from aligned organisations and 20 from news outlets.

Complaints show what affected people, or those acting for them, managed to lodge with a mechanism, which requires that a mechanism exists and that people know of it. Evaluations show what independent evaluation units and funders chose to examine, and we coded only their findings of shortfall, not their findings of success. News shows what journalists covered. A few evaluations predate 2021, such as a 2019 review of a UK development finance institution, because no later edition exists.

Where several items report the same case (Cambodia, Tanzania, Ecuador), we count the case once in findings: the 16 items recording found harm cover 12 distinct cases. Frequencies in this article describe the evidence base, not the market.

3. Where the record shows delivery falling short

We classify each item by the impact dimension in which delivery fell short, adapting the five dimensions of impact developed by the Impact Management Project and now maintained by Impact Frontiers. We use "risk" more narrowly than the norms do, for foreseeable harm that was not managed, and "contribution" for the investor's claimed contribution. Of 186 items, 92 concern what happened: the outcome did not occur or was negative. Forty-three concern risk, where harm was foreseeable and unmanaged but the source does not show it occurred, or where a process such as consultation was skipped. Twenty-five concern how much, meaning scale, depth or duration. Twenty-one concern who, where benefits missed the intended group. Five concern contribution, where the investor's claimed contribution is unproven in four items and measured as short in one.

Of the 186 items, 126 record harm that is alleged but that no body has found to have occurred. They include complaints where a mechanism found the lender compliant, or non-compliant without finding that harm followed. Sixteen record harm that a mechanism, court, regulator or evaluator has found. Nineteen record a shortfall measured against a stated target. Twenty-five record impact that is unproven, because outcomes were not measured or cannot be attributed to the investment.

All 16 findings of harm sit in the what dimension, where 75 further items are allegations and one is a measured shortfall. In risk, 35 items are allegations and 8 are unproven. In how much, 13 items are unproven, 11 are measured shortfalls and one is an allegation. Many evaluations report that impact could not be established. An evaluator who finds no outcome data has not shown that outcomes were poor.

A complaint states what the people who lodged it say happened. Mechanisms close cases on eligibility grounds, when parties enter dispute resolution, or when requesters withdraw. None of these tests whether the harm happened. Complainants usually have to document harm with no access to project records, while the lender holds them. A case that ends without a published finding is often the intended result of dispute resolution. This article therefore says "complaints allege" wherever no finding exists, and it names no institution or project on the strength of an allegation.

4. What the sources say about causes

Each item can carry up to three root causes, recorded only where the source evidences them, so counts exceed the number of items. Safeguards and governance failures, meaning environmental and social safeguards, due diligence or oversight that failed, appear in 89 items. A participation gap, where affected people were not consulted or had no voice in design and remedy, appears in 64. Measurement gaps appear in 20 items, policy or regulatory gaps in 9, and the distance created by funding that flows through intermediaries in 9. Unproven additionality appears in 6, and capital mismatch and incentives or fees in 4 each.

Twenty items state no evidenced cause. We did not infer one.

Actors differ by cause. Where safeguards or governance failed, the decision traced to a development bank or multilateral lender in 33 items, to a government in 27, to an enterprise in 23 and to an intermediary in 6. Participation gaps follow governments in 27 items, enterprises in 16 and development banks in 13. "Actor" here means the party whose decision the failure most directly follows from. It is not a finding of legal responsibility. Accountability mechanisms rule only on the lender's compliance; where a government is coded as actor, any finding still concerns the lender.

5. How causes link to shortfalls

Causes link to particular kinds of shortfall. Of the 89 safeguards and governance causes, 61 lead to the what dimension, where a negative outcome is reported, most often as an allegation, 22 to risk, 5 to who and 1 to how much. Participation gaps follow a similar path: of 64, 38 are linked to what, 16 to risk, 9 to who and 1 to how much.

Measurement gaps run elsewhere. Of their 20 appearances, 15 lead to how much, 3 to risk, 1 to who and 1 to contribution. Unproven additionality leads to contribution in 5 of its 6 appearances and to how much in the other. A scheme that cannot show it made a difference has not met the first of the four tests set out in section 9.

Many of these items were lodged by the people affected or by those acting for them. Items in the what dimension name local communities 29 times, Indigenous peoples 13, workers 13, displaced communities 12 and low-income borrowers 9. Fishers and other resource users appear 6 times, small firms 5 and smallholders 4. In the risk dimension, local communities appear 13 times and ecosystems and biodiversity 8. One item can name up to three groups.

6. Who bears the cost, as the record shows it

The people named are local communities in general (51 items), Indigenous peoples (19), displaced communities (15), workers (14), low-income borrowers (11), fishers and other resource users (6), small firms (6), smallholders (5) and women (3). Two further categories, ecosystems and biodiversity (11) and taxpayers or public budgets (2), are not people and are listed separately. Harms to women are rarely recorded separately in complaints or evaluations, so a count of 3 shows how they are recorded, not how often they occur.

Severity is assigned by a rubric and not measured, and for allegations it reflects what is alleged. All six items naming fishers and resource users are rated high. Of 15 items naming displaced communities, 13 are high. Of 19 naming Indigenous peoples, 11 are high, 7 medium and 1 low. Of 11 naming low-income borrowers, 7 are high.

In Togo, artisanal fishers took their concerns to the World Bank Inspection Panel, which registered their request in September 2021. The Panel investigated the World Bank's compliance on a coastal resilience project. It found that the project's insufficient understanding of artisanal fishing, in particular beach seine fishing, and of the different groups who depend on it, caused many of the concerns raised. That lack of understanding meant the project could not identify the full nature of its impact. The Board approved a Management Action Plan in September 2023.

Fifty-seven items name no affected group. Most are evaluations or portfolio findings that report a shortfall without saying who missed out. Of these, 41 are rated medium severity. Without a named group, no one can test whether the intended beneficiaries were reached.

7. The evidence gap

In the evidence base, 125 items have a public actor as the primary party, meaning a development finance institution, a multilateral bank or a government. Sixty-one have a private actor, but almost all are companies financed by public lenders: enterprises (35), intermediaries (25) and private investors (1). Most appear because a public lender's accountability mechanism or evaluation exists. Only one is a private investor.

The source tiers show why. Of the 112 complaint items, 44 trace to governments, 35 to enterprises, 22 to development banks and 11 to intermediaries. Of the 33 evaluator items, 25 concern development banks. Of the 20 news items, 13 do.

Failures surface where accountability mechanisms and independent evaluators exist. Development banks have complaint offices and evaluation units that publish. Few private impact funds publish a comparable independent complaint or evaluation function. Where a development finance institution is a limited partner, its mechanism may cover the fund, but complaints that reach it are recorded against the institution. We searched news archives, news feeds, a browser-based search and the websites of aligned organisations for documented failures by private impact funds and found very few. The candidates we examined were mostly ordinary lending or commercial private equity with no impact claim, and we excluded them. The method note has the detail.

This is a finding about visibility. It does not show that private funds perform better or worse. The evidence needed to judge them is not public, so outsiders cannot test their claims.

8. Allegation and finding

Of the 112 complaints, 12 ended with a published finding of non-compliance, 7 with a published finding of compliance, and 93 with no published finding that we could locate. Some of the 93 are open. Others were resolved through dispute resolution or ended when the requester left the process.

A compliance finding means the mechanism found the lender followed its own policies; it does not establish that no harm occurred. These cases remain counted as allegations. A non-compliance finding is also different from a finding of harm. Nine of the 112 complaints carry a finding of harm, and for the other 103 harm is alleged. The gap between 12 and 9 arises where non-compliance was found without a finding that harm followed.

Few complaints reach a finding, and the route is long. In the Cambodia case, the CAO reported to the IFC Board more than three years after registration. We did not measure time to finding and do not claim this is typical. Timing still shapes the count: of 79 complaints registered from 2023 onwards, 72 have no published finding, and cases registered in that period could rarely have reached one by October 2026. Others ended when requesters withdrew. In one such case the workers had asked for confidentiality for fear of reprisals and reported indirect threats, before reaching an agreement with the company and withdrawing.

A finding does not guarantee remedy. Management may dispute it, and action plans are often not agreed with the people affected. We did not track whether any finding led to compensation or restored livelihoods.

A complaint count says little about a lender's performance. It reflects how many people could reach a mechanism, how many could do so safely and with support, and how far each case had progressed on the day we looked.

9. What would close the gap

The gap is one of evidence, so the remedies concern evidence. IntelliSearch applies four tests to any intervention. It should be additional. It should show contribution, financial or non-financial, against a stated counterfactual. It should be measured, against a stated source. And it should be able to continue when grant or concessional funding ends.

Investors can:

• require a fund-level grievance mechanism open to affected people, and report its case log annually to limited partners;
• obtain independent verification of impact management, as Principle 9 of the Operating Principles for Impact Management sets out;
• state the investor contribution strategy (signal, engage, grow new markets, flexible capital) and the evidence for it;
• involve affected people in designing projects, monitoring them and agreeing remedy, with a grievance channel they can reach without going through the investee.

Development finance institutions can:

• publish the outcome and elapsed time of every complaint, including those closed without a finding;
• publish whether each finding led to remedy agreed with complainants;
• adopt and fund a reprisal-response protocol;
• record the affected group in portfolio evaluations, given that 57 items name none.

Policymakers can:

• apply existing anti-greenwashing rules, such as the Financial Conduct Authority's under the Sustainability Disclosure Requirements, to impact claims, and require evidence that can be inspected;
• make access to an independent grievance mechanism a condition of public or concessional money flowing to private impact funds.

IntelliSearch is a UK research and advisory firm for impact capital, and it publishes the method behind this analysis in the accompanying method note so that others can inspect it. To correct an item or exercise a right of reply, write to hello@intellisearch.org.