The Training Gap: Every Agency Is Buying AI. Almost Nobody Is Training for It.

The Boulder Police Department spent a full year piloting Axon’s Draft One before rolling it out to all sworn personnel. Officers wrote reports by hand, then compared them side-by-side with AI-generated versions. Supervisors conducted monthly audits, cross-referencing AI reports against body camera footage. The department met with its Police Oversight Panel and with community members before going live.

The Oklahoma City Police Department did something similar. Before deployment, officers acted out mock scenes and compared their handwritten reports to Draft One’s output. According to COPS Office reporting, about half the time, the AI report needed no changes. The other half required officers to fill in missing context, a skill that, in turn, requires training.

These are the exceptions. Most agencies are not doing this.

The Onboarding Problem

When a department buys an AI tool, the vendor provides onboarding, which is product orientation. To ensure officers can effectively use the tool, departments must implement comprehensive training. That training should cover the tool’s capabilities and limitations, along with a solid base in general AI use.

Training means officers understand what the tool does and what it does not do. It means supervisors know how to audit AI outputs/artifacts against source material. It means records staff can identify when an AI-generated report has hallucinated a detail that was never spoken aloud. And it means prosecutors have been briefed on what they will see as the COPS Office documented, prosecutors are split on whether they even want AI-generated reports on their desks.

Vendor onboarding usually covers none of that.

The Readiness Deficit

Dr. Joseph Lestrange emphasizes that AI adoption in policing requires four integrated pillars: leadership commitment, co-design with vendors, governance and accountability, and ongoing investment in people through continuous training and development. The fourth pillar, the human element, is the most often neglected by law enforcement agencies.

Tell me if this sounds familiar.

The pattern is predictable. A department allocates a budget for the tool. The procurement process absorbs months. By the time the contract is signed and the check is cut, the pressure is to deploy, not to build a training infrastructure around it. Officers get a one-hour (possibly mandatory) walkthrough and are expected to figure it out on shift.

The result is uneven adoption. Tech-comfortable officers use the tool frequently. Others avoid it entirely. Supervisors lack the knowledge to evaluate AI-assisted work product. The department paid for a capability it cannot consistently use.

Across industries, 50% of organizations (Statista, subscription required) cite a lack of skilled professionals as their top barrier to AI adoption. Law enforcement faces a compounded version: chronic understaffing means the officers who most need time-saving AI tools have the least time available for training.

What Good Looks Like

Boulder’s year-long pilot is not realistic for every agency. But the principles scale down.

Start with side-by-side exercises. Let officers compare their output to the AI’s output before the tool goes live. Build in supervisor audit protocols from day one, not as an afterthought, after the first complaint. Brief your prosecutors and deputy attorneys general early; the COPS Office reporting shows that agencies that engaged their prosecutors’ office before deployment had smoother adoption than those who waited for court challenges.

Conduct what Lestrange calls an AI readiness audit: assess your agency’s adoption culture, existing policy infrastructure, training capacity, and long-term funding before you sign the contract. Addressing these gaps beforehand will help officers feel prepared and reduce uncertainty during deployment.

The agencies that treat AI training as a line item alongside procurement will use these tools effectively. The agencies that treat vendor onboarding as training will end up with expensive software that half their officers ignore and the other half misuse.