How AI disruption tracking would handle Tropical Storm Bertha today

Tropical Storm Bertha is useful precisely because it did not give operators the courtesy of a long runway. On May 27, 2020, it formed suddenly off the South Carolina coast, reached 50 mph winds, was named roughly an hour before landfall, and came ashore near Charleston before most organizations could have moved from weather monitoring to transportation action.[1]

That compressed clock is the real test for AI logistics disruption tracking. A slow-moving hurricane gives a supply chain team time to convene, debate thresholds, and reroute freight through ordinary escalation paths. Bertha gave almost none of that. The one documented operational disruption in the provided materials was outside ordinary freight movement: SpaceX scrubbed a Crew Dragon launch because high winds affected the booster recovery zone.[1] That does not make Bertha a known supply chain failure. It makes it a clean benchmark for a harder question: when a weather risk forms fast, does today’s alerting stack reach the right logistics decision-makers before they are already improvising?

The 2020 Bertha Clock Left Almost No Operational Slack

The uncomfortable part of Bertha is not its scale. Reported damage was about $130,000, minor by hurricane-season standards.[1] The uncomfortable part is the sequencing. A storm that becomes operationally relevant only after naming is already too late for many logistics choices: a truck is already moving toward a terminal, a drayage appointment is already set, a warehouse shift is already staffed, and a supplier inside the risk area may already have a loading commitment.

In 2020, a logistics team watching public weather alerts could have known a coastal system was developing. What it likely did not have, unless it had built the capability itself, was an automated chain from meteorological signal to network exposure: which inbound purchase orders touched the coastal Carolinas, which carriers had freight in motion, which customer orders depended on those lanes, and who had authority to approve a reroute or hold.

That distinction matters. Storm tracking as a map is not the same as disruption tracking as a workflow. A logistics manager does not need another colored cone if the cone arrives after the execution window has closed. They need the system to say, early enough to matter, which shipments, suppliers, facilities, lanes, and commitments are exposed.

What Would Surface Today

A modern AI disruption platform would not have made Bertha a large event. It also would not have guaranteed a perfect seven-day warning for a storm that formed quickly. The practical change is narrower and more valuable: the first usable signal would be connected automatically to logistics exposure, instead of waiting for a named-storm headline to trigger manual investigation.

The difference is not that AI knows the future. The difference is that the investigation starts before a person has to ask for it. If a distribution center near Charleston is irrelevant to a company’s current flow, the alert can stay low. If the same area contains a sole-source supplier, an urgent inbound load, or a recovery-critical spare part, the alert needs to move fast and land with someone who can act.

The Signal Has To Become A Shipment Question

The stronger platforms now compete on the handoff between external risk and internal execution. project44 says its Disruption Management Agent monitors more than 8 billion data sources and 100,000 news posts hourly across more than 120 risk categories; it also reports 75% faster disruption identification and 40% savings on disruption-related costs.[2] Those are vendor-reported performance claims, not neutral benchmarks, but the architecture points to the right operational problem: the weather signal must be joined to shipment, facility, and order context.

Everstream Analytics describes a broader risk-data engine, processing 20 billion daily data points and combining NOAA GFS/GEFS, ECMWF, and proprietary AI models; the company also identifies customers including Unilever, Campbell’s, Schneider Electric, and Google.[3] For a Bertha-type case, the important part is not the large number by itself. It is whether model output, weather alerts, and local signals are converted into an exposure list quickly enough for transportation and procurement teams to decide what to do.

A useful alert would look less like “storm near South Carolina” and more like a work queue. It would identify open purchase orders from exposed suppliers, loads already tendered into affected lanes, customer orders that depend on those loads, and alternatives that are still feasible. If the alert cannot answer those questions, it may be accurate weather intelligence and still be weak logistics intelligence.

Where The Advance Window Comes From

The advance window comes from combining several imperfect signals rather than waiting for one official event label. Weather models can show developing risk before naming. News and local feeds can show closures, emergency preparations, and infrastructure stress. Supplier mapping shows which nodes matter to the company. Shipment visibility shows what is already committed. Automated impact analysis ranks which exposures deserve interruption.

That is why the seven-day claim needs careful handling. TraxTech says predictive AI systems can prevent 60–75% of unplanned production stoppages by providing 3–7 days of advance warning.[4] A Johnson & Johnson AI system, cited through WorldCertification rather than a primary Johnson & Johnson source in the provided materials, reportedly detected 85% of major supply disruptions an average of seven days before impact.[5] Those figures support the broader maturity argument. They should not be read as proof that every fast-forming tropical storm will produce a clean seven-day logistics alert.

For Bertha, the more defensible claim is that a modern system would likely have shortened the time between first meteorological concern and logistics triage. Whether that means days, hours, or minutes depends on the available model signal, the company’s network data, and how tightly the platform is connected to transportation execution.

Supplier Mapping Is Where The Alert Becomes Personal

A storm alert has different value depending on whether the company knows what sits inside the risk area. If the platform only knows owned facilities, it may miss a contract manufacturer, packaging supplier, cold-chain handoff, or regional carrier yard. If it has multi-tier supplier mapping, the same weather signal can become a ranked list of business consequences.

This is where a small storm can expose a large process weakness. The damage total from Bertha does not tell a logistics team much. The location and timing do. A supplier one county inland may be unaffected; a load approaching a coastal cross-dock may need a different appointment; a critical inbound shipment may justify a reroute even when most freight can continue. The decision is not “storm or no storm.” It is which commitments lose flexibility first.

Resilinc’s 2026 hurricane-risk framing is relevant here because it treats storm season as a network exposure problem rather than a weather-viewing exercise. Resilinc cites a 2026 forecast of 6–7 hurricanes, including 2–4 major hurricanes, and reports disruptions up 38% year over year and extreme weather alerts up 33%.[6] Those numbers do not say Bertha caused a logistics breakdown. They do say that storm-risk operations are becoming routine enough that manual watch-and-react processes deserve scrutiny.

Demand Planning Belongs In The Same Conversation

Transportation disruption is only one side of the storm problem. Demand can move too: stores may pull forward orders, customers may delay receipts, service parts may become urgent, or a product mix may change by region. ClimateAi’s FICE model is described as combining weather data with demand data for proactive planning.[7] That kind of linkage matters because the right logistics action depends on whether the exposed freight is still needed on the same schedule.

For a Bertha-like scenario, a demand-aware workflow could separate freight that should be protected from freight that can wait. A high-priority shipment feeding a storm-sensitive customer region may need escalation. A replenishment load with flexible delivery may be held rather than pushed into congestion. A platform that only sees the storm path may treat both as equal; a platform tied to demand and order priority should not.

What A Good Bertha Test Would Ask A Vendor To Prove

A supply chain leader evaluating AI disruption tracking should not ask only whether the platform tracks storms. That bar is too low. The Bertha test is sharper because the time window collapses. The platform has to show when it would have detected risk, which data source triggered the alert, how quickly it mapped the event to the company’s network, and what decision it would have put in front of the user.

• Alert provenance: the platform should show whether the trigger came from NOAA or ECMWF model output, official warnings, news ingestion, carrier data, facility alerts, or another source.
• Lead time: the vendor should distinguish between early weather concern, operational exposure alert, recommended mitigation, and confirmed disruption.
• Network visibility: the system should identify affected suppliers, facilities, lanes, shipments, purchase orders, and customer commitments, not just draw a storm boundary.
• Workflow integration: the alert should reach the transportation, procurement, customer service, or risk owner who can approve the next move.
• Performance evidence: vendor-reported savings and speed improvements should be separated from independently verified benchmarks.

The most revealing demo would replay Bertha without pretending it was a catastrophe. Load a representative supplier and shipment network. Start before the storm was named. Ask the platform to show what it would have surfaced as the coastal signal developed, what it would have ignored, and when it would have escalated from watch item to logistics action.

The Honest Conclusion From A Small Storm

Bertha should not be inflated into a major supply chain event. The documented materials support a narrow case: it was a fast-forming, short-lived storm with minor reported damage and one concrete operational consequence in the SpaceX launch scrub.[1] Its value is as a timing benchmark.

Against that benchmark, today’s AI disruption tracking tools have clearly moved the work upstream. They can ingest weather models, news, supplier maps, shipment data, and risk taxonomies at a scale that was not a normal 2020 operating baseline. They can also convert some portion of that signal into exposure analysis and recommended action before a manager has to start from a blank screen.

The buying question is not whether AI would have “prevented Bertha.” It would not. The buying question is whether a platform can turn a fast weather signal into a timely logistics decision: what is exposed, who owns it, what can still be changed, and how much of the claimed performance is proven outside the vendor’s own materials.

References

1. Tropical Storm Bertha (2020), Wikipedia
2. Disruption Management Agent, project44
3. Everstream Analytics weather and risk intelligence articles, Everstream Analytics
4. Predictive AI systems and production stoppage prevention, TraxTech
5. Johnson & Johnson AI system disruption detection citation, WorldCertification
6. 2026 hurricane forecast and supply chain disruption framing, Resilinc
7. FICE model, ClimateAi