Over the past five years, extreme weather events have become more frequent and more costly. In the United States, billion-dollar weather events rose from a historical average of 9 per year to more than 23, while the annual economic impact now exceeds $149 billion. This environment doesn’t just mean more operational interventions. It means growing pressure on cost recovery processes.

In that process, which involves contracts, work orders, technical reports, and vendor invoices, AI is already proving its ability to accelerate information analysis and improve cost traceability, increasing both the speed of recovery and the percentage of costs actually recovered.

When an extreme weather event strikes, utilities activate protocols to restore service as quickly as possible. During the emergency, multiple crews and vendors work on network restoration. In many cases, mutual aid agreements with companies from other regions are also activated.

But once the emergency is over, a less visible and equally complex process begins: reconstructing what happened, documenting every task performed, and reconciling all that documentation.

During those days, large volumes of information are generated. Assistance contracts, work orders, technical reports, material records, and vendor invoices are all part of the documentation needed to substantiate the event’s costs.

Carolina Charrie, Energy and Utilities specialist at Quanam, explained that one of the main challenges lies in the variety of formats used to record that information:

“It’s a perfect storm of documentation. The biggest problem is that the terminology often doesn’t match: the service is described one way in the contract, another way in the work order, and yet another in the invoice. That makes reconciliation extremely difficult.”

Once the information is gathered, utilities must organize it to submit reports to various parties, from agencies like FEMA in the case of public utilities and cooperatives, to regulators or insurers for private companies. Every expense must be correctly classified, linked to the corresponding documentation, and submitted in the required format. In practice, this process can take several months.

The problem isn’t just operational. It’s financial. Not all event costs are eligible for recovery, and even among those that are, the likelihood of recovering them depends heavily on the quality and consistency of the documentation submitted.

There’s also a critical time factor. Industry estimates suggest that around 28% of FEMA claims take more than six years to close, against a backlog exceeding $73 billion. For utilities, this means financing operations for years, which can represent a cost of nearly 17% annually on disputed amounts, eroding the real value of what is eventually recovered.

On top of that, the administrative burden can require dedicated teams for months, with internal costs ranging from tens of thousands to over half a million dollars per event.

The case of Central Florida Electric Cooperative (CFEC) illustrates this clearly. After four hurricanes between 2023 and 2024, it accumulated more than $45 million in damages and, due to delays in reimbursements, had to finance repairs with debt. This generated over $6 million in unrecovered costs and interest, nearly 14% of the total, and led to a rate increase of around 10%.

In this context, artificial intelligence is emerging as a key tool for capturing financial value. The goal isn’t to replace human work, but to accelerate tasks that involve reviewing large volumes of information.

As Nadia Bellati, Project Manager and AI Lead at Quanam, explains:

“Artificial intelligence makes it possible to process large volumes of documentation in different formats, whether PDFs, images, or digital forms, and transform that information into structured data. From there, contracts, work orders, and invoices can be linked automatically, something that was previously done by hand.”

In utilities already moving in this direction, these solutions are implemented as an integrated workflow covering the entire process end to end.

In the first stage, documents generated during the event are fed into the system in their original formats. The AI normalizes that information and converts it into structured, comparable data.

The system then automatically extracts the relevant data from each document: amounts, dates, service descriptions, work codes, and vendors. It applies classification and matching logic to link each invoice to the corresponding work order and contract, identifying matches and flagging terminology inconsistencies that previously required manual review.

Once the information is processed, specialized teams carry out human validation of the results, reviewing flagged inconsistencies and ensuring traceability of every data point. This step is especially critical in a regulated sector where every decision must be defensible before auditing bodies.

Finally, the system generates reports with all supporting documentation organized and linked to each cost item, making review and submission to the relevant authorities far more efficient.

In recent sector experiences, reconciliation processes that previously took up to six months have been reduced to six to eight weeks through the incorporation of AI.

In regulated sectors like utilities, it’s not enough for a model to produce results. Reports must be traceable, explainable, and auditable.

That means being able to reconstruct which information was used, how the data was classified, and how each conclusion was reached. For that reason, human validation remains an essential part of the process.

Artificial intelligence functions here as an amplifier of human capacity, enabling teams to analyze large volumes of information faster and with greater precision.

When an extreme event strikes, the challenge isn’t just repairing physical infrastructure. It’s also ensuring that systems and data remain available to reconstruct what happened and resume operations.

As Ernesto Rapetti, Senior Data and AI Consultant at Quanam, points out:

“In these scenarios, it’s not just about having a backup. Organizations need architectures that allow them to keep functioning, even partially, even if part of the infrastructure is affected.”

That’s why many utilities combine on-premise infrastructure with cloud resources. This hybrid model allows processing capacity to scale when needed, for example to analyze large volumes of post-storm documentation, without maintaining that capacity year-round.

For organizations looking to improve these processes, the first step is usually not technological. It’s analytical. Before implementing new tools, it’s essential to understand how the current process works.

Identifying bottlenecks, analyzing how long each stage takes, and assessing what information is available and in what formats helps pinpoint where the greatest inefficiencies arise.

In many cases, document reconciliation emerges as one of the first areas where AI can deliver concrete improvements.

As extreme weather events become more frequent and economically damaging, strengthening the entire post-event operations cycle becomes essential to optimizing cost recovery.

In this environment, AI is beginning to extend beyond document reconciliation, from assisted generation of documentation for auditors and regulators, to digital capture of field data at the source.

The result is a more integrated process, with greater traceability and better prepared to scale in the face of future events.

Utilities that move in this direction won’t just recover more. They’ll recover faster. And that difference, in a high-frequency event environment, translates directly into financial and operational advantage.

Let’s keep the conversation going.

Every storm tests not just your infrastructure, but the financial processes behind it. If you want to explore how artificial intelligence can improve cost recovery in your organization, let’s talk.

Carolina Charrie

Energy & Utilities Specialist

Nadia Bellati

Project Manager & AI Lead

Ernesto Rapetti

Senior Data & AI Consultant

Alejandro Acle

Journalist