Last week, I outlined my updated approach to building an adaptation and resilience investment universe using Claude. I talked about how taking a sector-by-sector approach to adaptation and resilience should help maximise the opportunity set for returns generation throughout the cycle.

However, the sector component is only half the story. The sector tells you what a company does—critical for assessing whether it provides an adaptation solution or how demand for its product is exposed to physical risk.

The other side of the coin—and the trickier one—is geography: where a company operates, where it sources inputs, and where it sells its products.

This week, I want to explore the very real challenges that geographic exposure presents for investors seeking adaptation and resilience opportunities.

Investors with exposure to climate investing in recent years have had a lot to contend with: an uneven, non-linear energy transition, shifting political dynamics, and uncertainty around demand and technological evolution.

Yet, compared to adaptation and resilience, transition investing is arguably a simpler endeavour when it comes to measuring exposure and success. That’s because mitigation has a single, core metric: carbon emissions.

A tonne of carbon emissions is the universal climate mitigation language. Companies and investors can use it to assess everything from business model intensity to the impact of new technologies or policies on warming trajectories. A tonne of carbon in the US is the same as a tonne in Germany, China, Japan, or East Timor.

Adaptation and resilience investing, by contrast, lacks a single unifying metric. Physical climate risk—and the opportunities it creates—is inherently location-specific.

Physical climate risk is local. Different hazards occur in different places, with varying frequencies and intensities. Consider a few well-known examples:

Tornado Alley cuts through the central plains of the United States.

Source: Scientific American (as of 2023)

Tropical storms form around the equator, affecting the southern US, the Caribbean, and parts of Asia.

Source: BBC (as of 2026)

Droughts devastate sub-Saharan Africa and create acute water stress in parts of North and South America.

Source: NASA (as of 2026)

Different geographies face different risks—and different risks require different solutions.

As a result, exposure to physical climate risk—and the resilient investment opportunity—depends on a combination of what a company does and where it does it. There is no single, simple metric to capture this.

What this means for portfolio construction

Over the past week, I’ve been working through what this complexity means in practice when it comes to building an adaptation and resilience portfolio.

There is a growing ecosystem of data providers offering physical risk exposure data, typically for risk management purposes. Increasingly, this data can also be used to identify investment opportunities.

However, vendor selection is critical. The usefulness of the data ultimately depends on the quality of the underlying methodology.

For this experiment, I am deliberately using open-source providers that I trust—and that subscribers can access themselves.

I’ve long been a fan of Probable Futures (www.probablefutures.org), a rigorous, open-source platform for forward-looking physical climate risk data. I’ve used it extensively in research and have returned to it here as I apply that same rigour to portfolio construction using AI.

It is not a complete source of geographic data on extreme weather risk but it is a very solid foundation to build upon.

Source: Probable Futures (as of 2026)

Some aspects of working with this data have been greatly simplified by Claude; others have required close scrutiny and active questioning of its outputs.

For example, when prompted to use Probable Futures data, Claude generated extensive, well-structured Python code to access the API and analyse geographic exposure across multiple hazards—saving me hours of work.

However, when mapping operational exposure to physical risks, Claude struggled with important nuances—highlighting broader challenges for investors trying to capture this opportunity set.

• First, it limited results to a pre-existing portfolio of 89 names identified earlier. This approach ranks risks within a fixed universe but fails to identify new resilient opportunities—meaning potentially attractive companies are missed.
• Second, it defaulted to a geographic focus on Southeast Asia without instruction. This kind of concentration overlooks the complexity of global supply chains.
• Third, when it came to mapping company exposures to physical risk Claude relied on single-point location data (e.g., headquarters or primary operation), rather than reflecting the full geographic footprint of operations.

To be fair, capturing the full complexity of global operations (let alone full supply chains) and mapping them against multiple physical risks is an enormous challenge. It’s precisely what makes adaptation and resilience investing so difficult—and so interesting.

What is next for the AI-assisted fund construction project

Figuring out how to build a systematic approach to match multiple weather hazards to multi-geography operations and customer bases to find resilient companies feels like playing 4D chess.

There is no neat solution to share this week. Rather, it remains a work in progress. I am working on options for gaining as much access to the resilience opportunity set as possible while accepting that a full mapping for all companies is highly unlikely with the data and disclosures available to me at this stage.

This stage of fund construction will take time and a lot of iteration to work out so I suspect next week’s Resilient Investor article will be more of a regular research piece as Claude and I work on this in the background.

Please do get in touch if you’ve got any specific questions on this process or any suggestions on approaches that support geographic mapping by emailing me at [email protected].

Subscribe below to follow our progress as well as regular thematic research.