It costs at least $100,000 to train a law enforcement officer. The most recent nationwide academy graduation analysis pegs average graduation rates at 86%. That means that every year, law enforcement agencies across the country spend at least (do you think that rate is still 86%?) $837M to train candidates that never become law enforcement officers. 

Early exits are an even bigger, harder to measure, problem. In a 2024 IACP survey, 68% of respondents indicated turnover is most common in the first five years. What’s the impact on your agency when officers who have just reached full productivity, head for the door? 

Some of that attrition is okay. Nobody bats a thousand. And the folks who are ill-equipped for the role should not be on the job. But these academy misses and early exits amount to more than a cost-hit: they erode an agency’s ability to deliver results. 

Early exits impact your organization’s productivity, not just its costs. 

In the private sector, poor candidate quality and early exits impact a company’s operating costs and profitability. And things that impact profitability, get measured. That’s why we know that low-turnover teams have 14% higher sales performance than high-turnover ones. We know that restaurants with high turnover teams see service times (and customer frustration) spike 22% during peak hours. 

Low-turnover teams don’t just drive revenue up, they drive costs down. Empty seats equal lost revenue. Early exits mean higher recruiting and training costs. High-turnover teams lack the expertise needed to make tough judgment calls, quickly. 

Pattern and keyword matching solutions create risk. 

AI-powered recruiting and hiring tools have existed in the private sector for some time. The simplest – and most prevalent – are resume screening tools that look for words or phrases that match a job description. These tools assume the job description is a carefully crafted outline of the responsibilities and skills required for success, rather than an artifact from three hiring managers ago or a cheery attempt to cast a wide net. 

More sophisticated tools pattern-match profiles based on who you have historically hired or industry benchmarks. Without understanding what traits in your existing employee pool actually impact outcomes like future tenure or performance, these tools can end up entrenching sub-optimal outcomes. Even worse, they can create hiring bias risk by inadvertently using proxies for race, gender or age that end up punishing applicants from protected classes. At Sigma Squared, we mitigate these risks in part by linking experience to outcomes, but more importantly by automatically running every trait through bias analysis algorithms to ensure it is not a proxy for a protected class. 

Outcome-oriented models help you understand what works, and why. 

If your organization is suffering from early exits or low productivity, throwing AI at resumes will not solve the problem. Poorly calibrated pattern matching will exacerbate it. Improving candidate quality and reducing early exits requires an organization to understand what things drive retention in your current employee base and then consistently parse and weigh those traits across your applicant pool. 

Just because the trait exists, doesn’t mean it predicts. A Sigma Squared client with below industry-average tenure was convinced that a specific “culture-fit” interview question was critical. Those with strong “culture fit” would stay longer. But as we analyzed thousands and thousands of current and former employees and looked at how they answered that question, we found it had no bearing on future performance or the length of time they stayed. 

What did predict strong, high-tenure employees for this company? Role velocity. Likely commute times. Specific remits within a brand, regardless of what the brand was. Experience combinations that showed versatility. These insights weren’t about up-ending the client’s hiring process, but rather providing guidance about how to weigh and consider these inputs in a way that better aligned with the firm's goal of increasing average tenure and store-level performance. 

Don’t just analyze, embed the insights into relevant workflows. 

Any single factor may be intuitive. Everyone knows that an applicant with a history of job-hopping is a risk. What’s very difficult for the human brain to do is to weigh a few dozen factors, on the fly, and apply those factors across hundreds of applicants consistently. 

In practice, this means pushing model intelligence into hiring workflows so recruiters, assessors, and decision makers can access consistent, outcome-oriented scoring on-demand.  

For example, a national restaurant chain bumps applicant background and interview data against models developed using Sigma Squared, generating readiness and retention scores in real-time. Those scores are pushed directly into applicant tracking software. They fast-track high-readiness candidates to the interview phase and provide store-managers with critical performance insights. In less than a year, they’ve cut early exits by 43%, recouped $4.8M in annual turnover related costs, and are generating an incremental $13.1M in revenue.  

Start small, move fast, measure results. 

The firm above has big-picture goals they’re looking to hit, but they started with something they could quickly measure: early exits. Tenure timelines in law enforcement are longer than in the private sector. While you could model out the predicted impact on tenure immediately, you will not reap the benefits for years. 

But you could very quickly impact academy outcomes. Or map out average readiness across an applicant pool and set a target to drive that up on a twelve month timeline. Or, consider the ways in which outcome-oriented models may help you better evaluate non-traditional applicants fairly, without lowering the bar. When expanding the applicant pool is an important piece of the puzzle, outcome-oriented models become a particularly useful tool. Let these measurable impacts provide a bridge to the longer-term outcomes that we all want for the profession.