Better questions, not verdicts: Reviewing our global portfolio’s impact with AI
By Jackie VanderBrug
Since Heading for Change’s inception, it has been a privilege to witness the portfolio grow from the seed of an idea into a global portfolio of eighteen private market funds, each investing at the intersection of climate, nature and biodiversity solutions and gender equity.
We built this portfolio to be broad and wide on purpose, underpinned by the mission of creating an illustrative global portfolio of diverse climate funds across the spectrum of climate themes, geographies, stages with each leveraging a gender lens that fits its context. The range was always the point. It allows the portfolio to demonstrate in an experiential and context-specific way how the climate and gender investing thesis take shape, with no one-size fits all approach.
But we believe that insights, patterns, and archetypes can travel across funds that look very different. If that’s true, then the very heterogeneity that makes the portfolio hard to read is also what makes it valuable to read. That combination, a deliberately varied portfolio, relatively lean team, and a conviction that learning crosses boundaries, is what led us to build a light-touch, AI-powered system to surface patterns across a sample of portfolio funds.
What we built is deliberately simple. First, we create a structured baseline from a fund's entry documents, their commitments and targets, and our investment team's assessment of their strengths, capabilities and trajectory. Then we review the fund's later reporting against a fixed set of questions drawn from our climate and gender integration framework, the same lens we use at diligence. It asks nothing new of the funds, it just reads what they already wrote. We're comparing what fund managers actually report against what we anticipated or expected they would do about gender integration when we invested - not against some generic industry standard.
Because we're often one of the few investors on the cap table pushing hard on the climate and gender investing nexus, our due diligence memos captured commitments that other investors may not have asked for. We looked for a level of integration that a GP reporting to a wider set of investors will rarely offer on its own. So when a fund's generic reporting shows less than our baseline expected, that gap is often structural, not a sign the work stopped. It’s a cue for where to ask questions, not a basis for conclusions.
AI is genuinely good at two things here: it holds a consistent lens across messy, differently structured documents, which is hard for a person to do from memory across a stack of reports, read months apart. It is also good at noticing what is absent, not only what is present. It’s not easy to notice when something was never mentioned if you are reading for content. But if you ask the same fixed questions every time, a gap becomes obvious because the question doesn’t get answered.
What we found
This method was designed to not hand us verdicts about funds. It enables us to land on better questions, sharpen them over time, and provide a clear signal of where to focus conversations and future impact measurement pathways. Four things came out of using it across our portfolio.
A single extra year of reporting turns a flat number into a trend. Comparing a report to just one prior year (not reading in isolation) caught something a single-year read would have missed entirely: a metric that used to appear in the fund’s reporting had quietly disappeared or was expanded upon. On its own, the newer report looked complete. Next to last year’s, it looked like a deliberate choice. In other cases, that same two-year comparison confirmed the opposite: a commitment made at entry showed up, unprompted, in every report since, which is its own kind of finding worth noting. That’s one you get from reading two years instead of one - a small, low-cost change to any process.
The method has a hard limit, and that’s where relationships take over. AI can only read what’s on the page, not what’s happening at the fund or the conversations held in person where facial expressions reveal a deeper tell. It can’t tell you whether a missing metric means a fund stopped doing the work, or simply stopped reporting it, and no amount of careful reading resolves that. This is the edge of what the method can do. From here, it’s the relationship that takes over: the trust built with a fund, and the honest conversations that trust makes possible. We were never just checking boxes against what a fund promised on paper or what they lit up about in conversations. The read sharpens what we bring to the relationship, and creates a strong foundation to build off.
The meaningful output isn’t a verdict, it’s a better question and it gets more valuable over time. This method let us ask a specific, informed question, like how a particular trend was tracked, instead of something vague. And that gets more useful the longer we do it: a pattern spotted in year one is just a question. The same pattern for five or ten years, watching the same signals across the same funds, becomes something we can act on, something for the climate and gender finance field to pay attention to, and sometimes something we can help fix or enhance.
Across the portfolio, it flags exactly the kinds of things worth raising directly with a fund. Run across the portfolio the method has the potential to surface concrete things worth asking about face-to-face. For example, a gender-specific commitment made at entry, later reported through a different, easier-to-hit metric that does not quite answer the original question. Or a fund that started out talking about climate and gender together, and now leans toward whichever frame the market currently rewards. Elsewhere, we've seen funds carry a gender-specific commitment made at entry across multiple reporting cycles, long after the due diligence conversation that created it. We've also seen funds that started out treating climate and gender as separate workstreams move toward reporting on them together, echoing how our own thinking on the intersection has evolved. None of these prove much on their own, but each is exactly the kind of thing worth raising directly with a fund, rather than quietly noting in a memo.
If you are thinking of trying this yourself, here is what we would tell you:
Keep it simple. All you need is the same (or slightly tweaked) set of questions, asked the same way, every year - not a new system or process to maintain.
Take note when something is missing. Don’t explain away a gap in the reporting. A missing metric is information in itself, and not a hole to smooth over.
Be upfront about what this is and isn’t. You’re reading what a fund reports, not what a fund actually does. Say that clearly.
Remember the point: We’re looking for better questions, not a grade. The read hands you what to ask, but the relationship with the fund carries the rest.
We would like to compare notes with others doing this work, so please do reach out if you’ve tried something similar as we’d love to hear about your experience and learnings. Afterall, learning is most useful when it is collective.