Alexandra Danielsson, director at Canbury Insights, and Carolyn Saunders, pensions partner at Dentons, argue that with AI, pension funds and other asset owners can finally reopen exclusion policies that no longer stand up to scrutiny
Exclusion policies are the daylight saving time of responsible investment. Pension funds inherited their exclusion lists from an analogue era, and whilst the tools have changed, the frameworks have not. Everyone knows they’re broken. Nobody has fixed them.
Europe has been trying to abolish daylight saving time for decades. The original rationale – wartime energy conservation – is universally acknowledged to be obsolete. The EU formally voted to end it in 2019, but the clocks still change twice a year. The reason is not that anyone believes the system works; it is easier to leave things unchanged than try to coordinate agreement across competing interests. Exclusion policies in pension portfolios follow the same logic: a sensible response to the data, technology and consensus of an earlier era, now carried forward largely unexamined, because reopening them is harder than reviewing them.
Exclusion frameworks draw on four rationales – values, financial risk, international norms, and beneficiary preferences – that can overlap and conflict. Within each category, some exclusions are absolute, and others are revenue-based, but the same challenge applies: both the choice of activity and the thresholds of 5, 10, or 25% appear calibrated as much to tracking error and portfolio impact as to the strength of evidence.
Put plainly, exclusions are largely mathematical, not principled. A backwards-looking revenue threshold measures today’s business mix, not tomorrow’s – a diversified company with new extraction licences and weak transition commitments can sit comfortably below the line, while a utility actively investing in renewables with residual fossil fuel exposure can be caught above it. Improvers get caught and laggards get missed.
Exclusion lists are also taken as read. Most were inherited from previous decision-makers or taken off-the-shelf from a data provider, designed for simpler corporate structures when a tobacco company was a tobacco company. Today, holding companies sit above subsidiaries whose revenue mix looks materially different from consolidated accounts. Supply chains stretch across geographies and tiers in ways that revenue thresholds were never designed to capture. The result is a static list generating outcomes its original authors did not intend, and missing harms they would almost certainly have flagged.
Reviewing exclusion policies properly has historically required more analyst time than any pension fund or asset owner could justify
Exclusion frameworks, in short, do not withstand scrutiny. They are inconsistent in their rationale, dependent on the quirks of whichever data source a fund uses, and shaped more by what is easy to measure than by what is genuinely material. Pulling the obvious levers – lowering the threshold, broadening the activities covered – can introduce its own problems: a structural energy underweight, exclusions concentrated in emerging markets, and reduced transition incentives for companies already facing capital constraints. None of which sits comfortably with a just-transition framing.
AI changes the resourcing calculus
There is a practical reason this has been hard to fix. Reviewing exclusion policies properly has historically required more analyst time than any pension fund or asset owner could justify. The judgement was outsourced, and whatever definitional choices came back tended to be accepted as if they were neutral.
They are not neutral. There have been instances where commercial screening sources have quietly removed flags on companies whose underlying activities have not changed, while international references have continued to list them. Same company, same public facts, materially different outcomes depending on which data source a fund relies on. A revenue line tells you what a company sells; it tells you very little about how, and by whom, those products are deployed.
AI changes this, but not in the way the technology is often marketed. The answer is not a faster version of the same generic screen. It is analysis configured around a fund’s own policies, its specific holdings, and the ethical and financial-risk questions that are actually live for its beneficiaries. Tools can weigh disparate evidence across corporate disclosures, government contracts, news reporting and NGO analysis, in any language, at scale. The critical distinction is between AI that accelerates a generic output and AI that amplifies genuine in-house expertise, codifying the judgement of a specialist analyst and extending it across thousands of issuers simultaneously.
Resourcing, in other words, is no longer the binding constraint. The industry can finally go back to first principles. Are today’s ethical norms still the ones embedded in current screens? Are the activities being excluded actually the ones with material transition risk, or is the screen still pattern-matching on yesterday’s extraction industries while missing steel, cement and transport? And are funds relying on inputs that the same evidence base could plausibly contradict?
A legal perspective
Given the potential shortcomings of exclusion policies, pension fund trustees who fail to review them risk breaching their fiduciary duties. In the UK, exclusions need to be justified either on financial materiality or on the legal “test” for non-financial factors, which requires trustees to have good reason to think members would support the exclusion, and the exclusion must have no risk of significant financial detriment. Trustees who rely on inherited lists without applying more sophisticated analysis tools may find it difficult to evidence that they have exercised their discretion appropriately, leaving them open to challenge.
Reopening exclusions is uncomfortable, but failing to do so is risky. Trustees do not get thanked for reopening settled questions. But the cost of leaving the clocks unchanged – inherited, inconsistent, and increasingly indefensible – is now higher than the cost of finally moving forward.
Alexandra Danielsson is director at Canbury Insights, and Carolyn Saunders is pensions partner at Dentons
