Impact investing has a comparability problem. AI offers a path forward

Impact investing now represents $1.57 trillion in assets under management, spanning nearly 4,000 organizations worldwide and growing at over 20% annually, according to Global Impact Investing Network’s (GIIN) 2024 impact investing market report.

More capital than ever is being deployed with the explicit intention of generating measurable social and environmental outcomes.

Yet as the market matures, a structural weakness is becoming harder to ignore: the industry still lacks a uniform way to compare impact results across organizations and that undermines its credibility and ability to allocate capital effectively as an industry.

Without comparability, capital cannot flow to the most impactful solutions, high-performing models struggle to stand out and claims of impact remain difficult to verify. This creates a comparability bottleneck that risks slowing future growth.

Impact is inherently contextual. A kilogramme of waste diverted in Nairobi – where diversion is closely tied to livelihoods within informal systems – does not represent the same outcome as a kilogramme diverted in Barcelona, where it signals efficiency gains within mature infrastructure.

Efforts to standardize metrics risk erasing the very context that gives impact its meaning.

However, the lack of standardization creates its own problem. When data cannot be compared at all, capital allocation becomes opaque. Organizations that measure impact rigorously but in ways that do not align with dominant frameworks risk becoming analytically invisible.

Today’s landscape reflects this trade-off. The coexistence of IRIS+, the Global Reporting Initiative (GRI), Sustainability Accounting Standards Board (SASB)/International Financial Reporting Standards, the Sustainable Finance Disclosure Regulation, International Finance Corporation Performance Standards and dozens of other frameworks – key components of the environmental, social and governance and impact measurement landscape – means that even similar indicators are defined with different units, scopes and thresholds.

GRI may report energy use in megawatt-hours across the full value chain, while SASB focuses on core operations and uses gigajoules. Each approach is internally consistent but across a portfolio, these differences compound into noise rather than insight.

This is not a new problem in capital markets. Financial markets only scaled after accounting standards were introduced, creating comparability. Credit ratings established a shared language for risk.

The World Economic Forum’s Stakeholder Capitalism Metrics, adopted by more than 50 leading companies, represent a similar attempt to bring consistency to non-financial reporting.

Impact investing is now facing its own version of this infrastructure challenge: without a common basis for comparison, scale alone cannot deliver efficiency or credibility.

If comparability cannot be solved through standardization alone, the question becomes: what will it take to make impact data usable at scale? Three shifts are emerging:

First, treat comparability as translation rather than uniformity. Investors do not need identical indicators. They need reliable ways to interpret differences: to determine when metrics reflect the same underlying outcome, when they should be grouped but kept distinct and when they should not be aggregated at all.

Emerging approaches such as the Common Framework developed by the Common Approach, alongside systems like GIIN’s IRIS+ and the UN Development Programme’s SDG Impact Standards, point toward this model: structured enough to enable alignment, yet flexible enough to preserve context.

Second, use artificial intelligence (AI) to enable classification and harmonization at scale. AI will not (and should not) define what counts as impact. But it can dramatically improve how impact data is processed.

By classifying indicators based on meaning rather than wording, identifying equivalencies across reporting systems and flagging risks such as double-counting, AI makes large-scale comparability operational.

In a pilot developed by Agile Impacts with the Inter-American Development Bank across more than 500 investment operations, this approach achieved around 90% precision in indicator matching. Human judgment remained essential but what was previously infeasible at scale became practical.

Third, invest in governance, not just technology. A portfolio-quality system still requires human decisions: what qualifies for inclusion, where harmonization is reasonable, how double counting is prevented and which assumptions are acceptable. Technology enables these functions. It does not determine what should count.

The most important part of this solution is not AI but rather a change in how we frame the problem.

Today, most standards try to solve two distinct challenges at once: how companies measure impact more effectively and how we compare that data across the market. These are not the same problem.

Helping a company measure more, with higher quality and greater consistency, requires flexibility: room for each organization to capture what is operationally meaningful and what drives its competitive advantage.

Comparing results across portfolios, on the other hand, requires shared logic for grouping, translating and aggregating indicators regardless of how each organization chose to measure them.

By bundling both challenges into a single framework, the field has been forced into an impossible tradeoff: the more you standardize for comparability, the less room companies have to measure what actually matters to them and the less they measure, the less useful the data becomes for everyone.

The paradigm shift is to separate these two problems. One track focuses on helping companies measure better: with more context, more granularity and greater alignment with business value. The other focuses on comparability as a distinct analytical layer that translates across measurement systems rather than constraining them.

AI can serve both, but in different ways: supporting richer data collection on one side and enabling scalable classification and harmonization on the other. What makes this shift possible now is the technology.

Only with today’s tools can we realistically take heterogeneous data from thousands of organizations and translate it into decision-ready evidence without forcing uniformity at the source. This changes the question from “how do we get everyone to measure the same way?” to “how do we make sense of what everyone is already measuring?”

Simply reframing the question can unlock the solution we’re really after.