What the Research Says About AI for Climate-Driven Disruptions

The evidence is strong on disruption, thinner on AI mitigation

Any serious study of AI for supply chain disruption from climate extremes starts with an imbalance. The climate-disruption baseline is increasingly well documented across public agencies, academic work, and business-continuity reporting. The evidence that AI reduces enterprise losses from those disruptions is much narrower: useful in some tactical settings, promising as an early-warning layer, but not yet validated across multiple seasons of actual extreme events.

That gap matters because the evaluation question has shifted. A planning director no longer needs to be persuaded that storms, floods, heat, and drought can disturb supply chains. The harder question is whether a specific AI product changes the decision chain quickly enough to reduce the cost of those events.

The disruption baseline is no longer speculative

The strongest part of the record is the climate-risk side. The UN event count reported by Yale Environment 360 is not a supply-chain metric by itself; it does not say how many purchase orders failed or how many plants missed production schedules. But it does establish the exposure environment in which those operational failures now occur: more frequent major weather and climate events over a multi-decade comparison window.[1]

NOAA’s 2024 U.S. disaster tally adds the cost dimension. Twenty-seven billion-dollar-plus events in one year, with losses above $182 billion, is a macroeconomic figure, not a software benchmark. Still, it sets the scale of the hazard environment that procurement, logistics, and inventory teams are being asked to absorb.[2]

Supply-chain-specific monitoring points in the same direction. Interos reported 94.5 million businesses at risk in 2025, up 48% from 2024, while Resilinc’s figures cited in Everstream’s 2025 analysis point to both more disruptions and a sharp rise in extreme-weather alerts.[2][3] These numbers do not prove that every company will face a climate-driven failure this year. They do show why risk teams are under pressure to move beyond static supplier lists and annual risk reviews.

The preparedness evidence is less complete in public form, but the available figures are uncomfortable. A Harvard Business Review study of 12,000 supplier sites found that only 11% were fully prepared for weather disruption, while 93% of surveyed firms in China and Taiwan and 49% in the United States reported increased climate volatility, according to summaries of the study.[4] The gap is not just about awareness. It is about whether supplier-site risk is mapped, reviewed, and connected to sourcing, inventory, and continuity decisions before the event.

McKinsey’s 2020 analysis framed the same issue in revenue terms: an unprepared company could lose about 35% of annual revenue from a single climate-driven shock, compared with about 5% for a prepared company.[5] That estimate should not be read as a universal prediction. Sector, geography, substitutability, and inventory posture all matter. Its value is directional: preparedness changes the size of the downside.

Where AI evidence currently reaches

The peer-reviewed AI literature is more cautious than most sales language. The 2025 Nature Communications paper on integrated artificial intelligence for complex climate risk proposes FATES, a framework for connecting forecasting, anticipation, threshold identification, exposure identification, and vulnerability assessment. It is useful because it describes the kind of integrated early-warning chain enterprises would need. It is also useful because it states the limitation plainly: “most efforts to use ML have focused on hazard forecasts and have yet to trickle down the early warning chain.”[6]

That sentence is the dividing line evaluators should keep in view. A model can improve hazard forecasting and still leave the enterprise exposed if nobody changes replenishment, allocation, supplier qualification, transportation mode, or customer-commitment logic. The operational claim begins only when the forecast changes a decision.

C2ES’s 2025 report on climate risk and supply chain resilience is helpful here because it treats climate risk as cascading rather than isolated. A flood does not stop at a supplier gate; it can move through transportation links, energy availability, labor access, inventory buffers, and customer commitments.[7] AI tools that only flag the original weather hazard may be valuable, but they are handling the first part of a longer chain.

The Business Continuity Institute’s 2024 discussion of weather-event resilience similarly supports the urgency of better preparation, continuity planning, and cross-functional response.[8] It does not, however, turn AI into a proven mitigation layer. Preparedness research and business-continuity guidance can justify investment in better sensing and planning. They cannot substitute for measured AI outcomes.

The best public deployment evidence is tactical, not enterprise-wide

The clearest public AI deployment example in the research set is ClimateAi’s Hurricane Ian case involving a roofing materials producer. ClimateAi says its forecasting helped the producer anticipate regional demand before Hurricane Ian and pre-position product, generating $15 million in additional sales.[9]

That is a meaningful operational result, and it is stronger than a generic statement about predictive accuracy. It links a forecast to an inventory and demand-positioning decision, then to a commercial outcome. It also has clear limits. It is a vendor-originated case study, focused on one product category, one company context, and one hurricane. It demonstrates a plausible and commercially valuable use case for weather intelligence. It does not demonstrate sustained enterprise resilience across suppliers, lanes, commodities, or multiple event seasons.

The distinction is not pedantic. Capturing demand before a storm and reducing losses during a broader supply shock are related but different outcomes. A roofing materials producer may benefit from higher near-term demand after a hurricane. A semiconductor assembler, food distributor, hospital network, or apparel brand may face a different problem: lost supply, constrained transportation, spoilage, service-level penalties, or customer allocation. The same weather signal can produce different decisions depending on margin structure, substitution options, safety stock, and customer commitments.

Market growth is not effectiveness evidence

Spending is moving faster than validation. Precedence Research estimated the AI in supply chain market at $7.15 billion in 2024 and projected it to reach $192.51 billion by 2034.[10] That forecast is useful as a signal of investment appetite and vendor expansion. It should not be read as evidence that AI reduces climate-driven disruption losses.

This is where procurement teams often get squeezed. The disruption numbers are large enough to justify urgency. The software market is large enough to create executive attention. But between those two facts sits the missing evidence layer: before-and-after outcome data from real disruptions, across enough events to separate model value from luck, seasonality, inventory overhang, demand spikes, or heroic manual intervention.

What evaluators can reasonably ask vendors to prove

The right standard is not perfection. Climate risk is messy, and multi-year validation will take time. But the evidence request should match the size of the claim. A vendor claiming better climate-risk visibility owes different proof than a vendor claiming lower enterprise losses.

• For hazard forecasting: ask for forecast skill, lead time, geographic resolution, false positives, and false negatives against known historical events.
• For early warning: ask who received the alert, how much earlier it arrived than the prior process, and whether the alert reached the team authorized to act.
• For planning recommendations: ask which inventory, sourcing, allocation, logistics, or customer-commitment decision changed because of the model.
• For financial impact: ask for event-level comparisons against a baseline, including stockouts, expedite cost, lost sales, service level, working capital, and write-offs.
• For resilience claims: ask for multi-season evidence across actual extreme events, not a single storm, pilot, or retrospective model run.

A serious vendor should be able to separate model performance from operational performance. If the proof stops at “we predicted the event,” the claim is weather intelligence. If the proof shows that planners moved inventory, changed sourcing, or protected service levels during the event, the claim starts to become supply-chain performance evidence. If the proof shows lower losses across repeated climate disruptions, then the vendor is approaching the resilience claim many buyers are actually being sold.

The current research-backed position

The available evidence supports a narrow, useful conclusion: AI can improve parts of the climate-risk decision chain, especially where weather intelligence feeds tactical inventory or sourcing decisions. The evidence does not yet support a broad claim that AI has been independently shown to reduce enterprise supply-chain losses from climate extremes at scale.

For now, the evaluator’s rule should be simple: treat AI for climate-driven supply chain disruption as potentially valuable, but require multi-season outcome data tied to actual extreme events before accepting claims of sustained resilience. Detection accuracy, market growth, and a single successful case are not enough.

References

1. How Climate Change Is Disrupting the Global Supply Chain, Yale Environment 360
2. Protecting Your Supply Chain from Extreme Weather: Steps to Minimize Risk, Interos
3. Climate Change Is Accelerating Supply Chain Disruption, Everstream Analytics
4. How Exposed Is Your Supply Chain to Climate Risks?, Harvard Business Review, May 2022
5. Could climate become the weak link in your supply chain?, McKinsey & Company, August 2020
6. Early warning of complex climate risk with integrated artificial intelligence, Nature Communications, March 2025
7. Assessing the Landscape of Climate Risk and Supply Chain Resilience, Center for Climate and Energy Solutions, September 2025
8. Climate Challenges: Enhancing supply chain resilience to weather events, The Business Continuity Institute, November 2024
9. Accurate Hurricane Forecasting Helps Roofing Materials Producer Come Out on Top, ClimateAi
10. AI in Supply Chain Market, Precedence Research