ESG data has spent decades reflecting what companies report. The sustainability factors that actually drive returns sit mostly outside that picture — and AI is now making them quantifiable for any portfolio team.
ESG data has had a structural problem that we as an industry have been too polite to name directly: it has been built around what companies choose to say about themselves.
That has meant a heavy focus on companies’ internal operations: energy use, headcount policies, governance structures, reported emissions. Useful for tracking progress for certain internal improvements – but a narrow and increasingly inadequate lens for investment decisions.
For sustainable capital allocation, the data questions that matter are about the future viability of a business model, not the sustainability of its head office.
The 4% problem in sustainable investing
Here is a useful way to think about the scale of the gap between internal and external impacts.
According to CDP, internal operations account for just 4% of a company’s total greenhouse gas emissions for the average company. The remaining 96% sits in the value chain – in the raw materials a company depends on, the products it sells, the markets those products operate in, and the end-of-life consequences of what it makes.
The same logic applies beyond climate. The largest sustainability impacts – on human health, on ecosystems, on communities – almost always occur outside the company’s own walls. They are properties of a business model, not of a corporate sustainability team.
And yet ESG analysis has been largely built around the 4% that companies can directly measure and choose whether and when to disclose. The 96% has remained largely invisible – not because it is unknowable, but because the tools to surface it have not existed at scale.
They have to be modelled from the outside.
How AI-enabled data modeling changes this
AI does not just make existing data processes faster. In the context of sustainability data, it makes entirely new kinds of analysis possible.
Until recently, building sustainability intelligence that does not depend on company disclosure required prohibitive amounts of manual research – mapping supply chains, cross-referencing scientific literature, modelling product-level impacts across value chains. Only the largest institutions, with dedicated research teams and significant budgets, could attempt it in any systematic way.
AI changes the economics of that entirely. The ability to collect, combine, clean, and interpret large volumes of heterogeneous data – scientific research, satellite imagery, trade flow data, regulatory filings, news, product databases – at relatively modest cost means that outside-in modelling is no longer the exclusive territory of sovereign wealth funds with internal quant teams.
The result is a shift from sustainability data as a collection exercise – pulling numbers from ESG reports and structuring them for downstream use – to sustainability data as an analytical product: forward-looking, model-based, and built to answer investment-relevant questions besides mere reporting requirements.
Case: NBIM’s data models for physical climate risk and controversies
Physical climate risk is the clearest example. Understanding whether a manufacturing site is exposed to flood risk requires satellite imagery and geospatial modelling — not company disclosures. Managers who rely on reported data for physical risk are missing one of the most financially material dimensions of climate exposure.
Norges Bank Investment Management (NBIM) has been building proprietary internal models for exactly this kind of analysis. They are also developing their own controversy screening, independent of what companies choose to report.
NBIM is not alone. Across the Nordics, leading managers are rethinking what sustainability data is for — and recognising that the era of pulling data from company ESG reports and feeding it into reporting templates is not the same thing as using sustainability intelligence to make better investment decisions.
What this means for your data stack
Three things worth acting on now:
- Distinguish disclosure-based from modelled data. The key question is not which ESG data provider you use — it is whether the data answers investment-relevant questions or reporting requirements. Both have a role, but the weight is shifting toward the former.
- Extend sustainability data into investment decisions, not just reporting. Most managers use sustainability data to satisfy regulatory and LP reporting obligations. That is a significant underuse. Feeding material sustainability signals into investment management besides basic exclusions is where the edge lies.
- Be deliberate about build versus buy. Building the scientific and value chain data foundations for outside-in modelling takes years. For most managers, the better path is identifying third-party providers who have done that work — while keeping internally built models for areas tied to proprietary portfolio data or specific investment theses.
The direction of travel
The shift from disclosure-first to model-based sustainability data is underway. What AI has done is dramatically lower the barrier to entry — making outside-in intelligence accessible to a much wider set of institutions. The managers building that capability now are positioning themselves ahead of a market that will catch up.
Eight years of outside-in modeling
At Upright, we have viewed the market through this exact lens since 2017. By combining financial macro-models of product and service revenue flows across global value chains with extensive scientific research, we map the actual impact, risk, and opportunity mechanisms of the global economy. Long before AI was a buzzword, we built the infrastructure to ensure capital allocation is driven by the best available financially material information, rather than pure corporate reporting.
In February 2026, Upright made this infrastructure accessible to asset managers directly through a real-time double materiality assessment tool, complementing the existing database of 50,000+ companies. By continuing to quantify what’s relevant – instead of what’s easily available – we look forward to helping asset managers to uncover the major external impacts and related financial risks and opportunities of any company or fund.
Daniel Lobo, VP Sales
Daniel Lobo is VP of Investor Sales at The Upright Project, where he works with asset managers and institutional investors to integrate science-based data into investment decision-making. Before joining Upright, Daniel spent 15 years at Bloomberg in senior enterprise sales roles working with global asset managers and asset owners
