A practical look at how emergency management organizations can evaluate AI based on operational need, organizational readiness, risk, and its ability to support better decision-making.
Artificial Intelligence and Emergency Management
I’ve spent a lot of time talking with emergency managers (from large state and city agencies to local emergency managers), researchers, technology companies, academics, and public and private sector leaders about artificial intelligence and emergency management.
Throughout these conversations it’s become clear that we don’t need to particularly plant a flag in either the “for” or “against” AI camp. The more important question raised has to do with where technology is a good solution to a particular problem.
It sounds straightforward on the surface, but when we dive in and think about putting it into practice, it drives us to ask some important questions:
What problem are we trying to solve?
What decision are we trying to improve?
Does the technology fit the operating environment?
What new risks does it introduce?
Can the organization support it?
And if we put it into an emergency operation, will people know when, why, and how to use it?
These questions matter far more than whether an organization can say it’s using AI.
AI is Broader than the Chatbot Conversation
One of the first things we ran into during the AI for Disasters + Emergencies (AIDE) Initiative was how differently people understand the term artificial intelligence.
For a lot of people, AI still means opening a box, typing in a question, and getting an answer back. I’ll admit that was part of how I thought about it going into this work too, but the applications are much broader.
We’re talking about predictive analytics, early detection, information processing, planning, situational awareness, evacuation intelligence, administrative support, and a wide range of other capabilities across preparedness, response, and recovery.
In one conversation around the AIDE Summit, I used the example of systems that use infrared imagery to identify fire starts earlier. That looks very different from the conversational tools driving so much of the public discussion around AI.
The AIDE research identified 1,179 AI-enabled products already serving the emergency management space. That number caught a lot of people’s attention, but the bigger takeaway is how little the term AI tells us on its own.
Two products may both use artificial intelligence and have completely different purposes, data requirements, operating environments, levels of risk, and demands on an organization.
AI literacy is understanding what a capability is before deciding to adopt it.
Start with the Operational Problem
Once you get past the terminology, the next question is whether the technology is solving something that needs to be solved.
Emergency managers consistently raised needs around better situational awareness, planning that can adapt as conditions change, and reducing administrative burden. At the same time, the research identified barriers including cost, capacity, procurement, and data readiness.
Both sides of that equation matter because you can have a legitimate operational problem, without the technology in front of you being the right solution for it.
One of the comments during the Summit that stuck with me was that expediting a flawed process doesn’t make that process more efficient: If nobody has clarity around who owns a decision, providing more information more quickly doesn’t fix the problem. If the data going into a system is unreliable, being able to process it faster will only get us so far.
The work has to start before technology enters the conversation, because we need to understand the process, the decision, and the outcome we’re trying to improve first.
Opportunities in Decision Support
Emergency management is fundamentally about making decisions under difficult conditions. Information changes, it’s incomplete, time matters, resources are limited, and the consequences of a decision are often significant. This is what makes decision support one of the more interesting opportunities in the AI space.
There’s a lot of value in helping emergency managers process more information than they could reasonably work through on their own, identify patterns, and build a clearer picture of what’s happening. That can put better information in front of decision-makers when time matters.
The AIDE research found significant interest around what it describes as “sensemaking,” helping emergency managers work through changing information and make informed decisions more quickly. However, that still leaves the decision with the person responsible for making it.
One point that came through in the Summit conversation was the importance of keeping human judgment central. Many of the products being discussed are designed to support decision-making rather than replace it, helping surface timely, actionable information that emergency managers still have to evaluate before deciding how to act.
Technology can improve the information available to emergency managers and help them act sooner, and human accountability still remains central to the decision.
The Harder Questions are Often Organizational
Once you move from talking about what a tool can do to whether an organization can rely on it, the conversation changes pretty quickly and a number of questions follow:
Can we procure it?
Does it need ongoing access to data?
Can that data be shared?
Does it require technical integration or specialized staff?
What are the cybersecurity implications?
Does it work if connectivity goes down?
Who maintains it?
What happens when something breaks?
These questions may feel less interesting than talking about the technology itself, but they’re often what determine whether adoption works.
The AIDE research found that 78% of the AI products evaluated require continuous internet connectivity. Only 10% can operate fully offline. 68% of vendors evaluated do not provide 24-hour technical support, and at the same time, more than 50% of local emergency management agencies operate with one or zero full-time employees. AI Magazine
Put those findings next to each other and you start to see the challenge: A technology can be incredibly capable and still be the wrong fit for a particular jurisdiction.
A large organization with dedicated IT staff, redundant connectivity, established procurement capacity, and the ability to manage a long-term software contract is operating in a very different environment from a one-person emergency management office. It’s important we account for both.
The question can’t only be, Does the technology work?
We also have to ask, Does it work here?
IT Needs to be Part of the Conversation
One of the conversations we need to get better at involves our Chief Information Officer (CIO) and IT teams. As an industry, we can be pretty quick to say, “IT won’t let us do this.” I understand where that frustration comes from: you find a technology that looks useful, you see a clear operational application, and then the conversation runs into questions about security, networks, data, integrations, or organizational policy.
But it’s worth thinking about the job IT is responsible for doing.
Those teams are there to protect the organization’s networks, systems, and data. They have risks they’re responsible for managing and they may be looking at the same tool from a very different perspective than the emergency manager who sees the operational opportunity.
We need each perspective in the conversation as an important part of adoption. Emergency management needs to be able to articulate the operational problem and why the capability matters. IT needs to understand what the technology touches, how it integrates, and what risks it creates. Procurement needs to understand what we’re trying to acquire and whether our existing processes can accommodate it. Depending on the use case, it may require bringing additional organizational stakeholders into the conversation.
Buying a Tool isn’t the same as Integrating a Capability
Another theme that came through clearly in the research was time. Emergency managers told us they don’t necessarily have the time to research hundreds of available tools, compare technologies, figure out which questions they should be asking, then build an adoption process from scratch.
That’s exactly one of the reasons we developed the AIDE Practitioner Playbook. The goal was to give organizations a process they could work through as they make those decisions.
The Playbook separates general organizational readiness from the process of adopting a specific use case. On the readiness side, the organization prepares and scans for opportunities. Once there’s a particular solution under consideration, the process moves through planning, commitment, exercising, expanding, and eventually integration.
This progression matters because there’s a significant gap between purchasing a product and having an operational capability. A pilot can tell us something, but an exercise can tell us more. Put a tool into an environment that starts to resemble how people will use it during an incident and different questions begin to surface:
What happens if the information it produces conflicts with another source?
Who validates the output?
What happens when connectivity drops?
Who has access to the system?
What information can staff put into it?
What is the backup process?
Does it save enough time or improve the decision enough to justify the complexity it introduces?
These are difficult questions to answer from a product demonstration. We start getting answers when we exercise the capability, learn from it, and see how it interacts with the rest of the operation.
Procurement, Literacy, and Governance
During the Summit I was asked what I’d look for several months from now to determine whether we’d moved from a good conversation into meaningful progress and three areas stood out to me: procurement, AI literacy, and governance.
Procurement is probably the least attractive part of the discussion, but it can quickly become one of the biggest barriers to moving forward. There’s tremendous capability in the technology sector, but public sector procurement processes were not designed around technology that changes this quickly. If organizations can’t responsibly acquire, evaluate, contract for, or move away from these tools, the availability of the technology only gets us so far.
AI literacy comes before a lot of those decisions.Emergency managers need enough understanding to evaluate what they’re being offered, distinguish one type of capability from another, understand the implications for their organization, and ask better questions.
Then there’s governance. There are still open questions around who establishes standards for AI use, what responsible use looks like, and how accountability should work. Emergency management isn’t going to answer every one of those questions on its own, but we do have work to do inside our own house. We need to determine how these tools fit within our operations, what risks we’re willing to accept, where human oversight is essential, and what standards we expect before relying on them.
Learning As We Go
Emergency management has always been a field that learns, from disasters, exercises, research, what worked and what failed. We have to bring that same mindset to AI. Waiting until every question is answered before trying anything isn’t realistic, but neither is adopting technology quickly then figuring out the implications afterward.
There’s a space in between those two approaches where we can test capabilities, exercise them, evaluate the results, understand the risks, make changes, and try again. This learning process is what gives us the opportunity to move from a technology that looks interesting to a capability that is useful.
There’s a lot already available to emergency managers, and there will be more. The measure of progress won’t be how much AI we can say we’ve adopted, but whether the technology helps us make better decisions, strengthens the way our organizations operate, and ultimately helps us serve communities better when it matters.