AI is already in probation and parole. The question is which jobs it should take from officers and which it should never touch. Our answer, and how we built to it.
Artificial intelligence is already in the criminal justice system: risk scores that inform bail and sentencing, predictive policing, facial recognition. The criticism is familiar and fair. Models trained on historical data can carry the biases of that history. Proprietary systems can be black boxes. And a recommendation from a machine can carry more weight than it deserves once it is on the screen in front of a decision-maker.
Community supervision has its own version of that debate, and it deserves a clearer answer than "AI should inform, not replace." Here is ours.
One corrections agency ran a time study on its officers and found 72.7% of their time going to administrative and system work. Writing the note. Entering the data. Chasing a phone number. Setting up the color-code voicemail for drug testing. None of that is supervision, and none of it needs a person's judgment. This is where AI belongs, and it is where RePath puts it. Visit Assist records the visit and writes the note; a note that took 30 to 45 minutes takes about five. Automated test scheduling gave one drug court coordinator 75 minutes a week back. Across our agencies, officers get 8 to 10 hours a week back, and the hours go to the part of the job that needs a human.
Impairment Risk is the closest thing in RePath to a prediction, and it is built as a prompt to look, not a verdict. It flags likely impairment; the officer decides what to do with the flag. In a Kansas juvenile district, a rising flag led to a phone call with the youth and his mother, a conversation about a new friend group, and an early course correction instead of a violation. In Alabama, a first alert turned out to be a medication change and ended with a doctor's visit. Three in four high flags that get tested come back positive. The one in four is why an officer, not the model, makes the call.
The officer stays the author. Every Visit Assist note is reviewed and posted by the officer whose name is on it, and the recording it came from is kept, word for word. That is what makes it defensible. A Kansas officer said she would not feel awkward reading one in testimony. A Texas juvenile chief answered a state audit finding on case-note quality with the recordings she already had. An officer in a specialized caseload noted the summary said what happened without recording clinical detail it should not.
The research is clear that officer training in core correctional practices reduces reoffending, and equally clear that it only works when the skills are actually used in real conversations. Almost nobody measures whether they are. RePath Intelligence looks at every recorded contact for practice quality and gives the officer coaching feedback. A twenty-year Kansas officer said it opened his eyes to different ways to have the conversation. A state deputy secretary of corrections, who has to audit one file per officer per month, said it would let supervisors coach instead of count.
We will not build a score that tells a judge whether to release someone. We will not build a model that recommends a sanction. We will not describe a flag as evidence. And we will not publish an accuracy number we did not measure across every agency we run in. The power in this system belongs to a person who knows the case. Our job is to give that person their week back and a better view of what happened between visits.
Book a 20-minute officer-hours call. We will show you the math on your numbers, not ours.
Book a Demo