An early intervention system is a database-plus-review process that flags officers whose recorded conduct crosses set thresholds — a set number of use-of-force reports, complaints, or other indicators within a window — so a supervisor can intervene with counseling or training before a pattern becomes a career of harm. Departments have adopted EIS widely since the 1990s, and federal reform agreements now require them: New Orleans's 2013 consent decree mandates one, as does Chicago's 2019 decree. But per the National Institute of Justice's research summaries, rigorous evidence that EIS flags reduce misconduct remains thin, and the systems' design choices — thresholds, indicators, and what happens after a flag — decide most of their real-world effect.
This article explains what the systems track, how a flag becomes an intervention, and what the documented record shows about outcomes. It publishes information, not legal advice.
What data feeds an early intervention system?
The inputs are records the department already keeps, plus some it often does not: use-of-force reports, civilian complaints, vehicle pursuits, resisting-arrest reports, firearms discharges, sick leave patterns, and in some systems, body camera audits and missed court dates. Scholars of the field, beginning with the criminologist Samuel Walker, whose 1990s work defined the modern EIS concept, described the design principle as non-disciplinary: the system is supposed to surface patterns that individual complaint files never show, because each incident looks defensible in isolation. The reliability of the output tracks the quality of the inputs — a lesson decree monitors have repeated, since complaint undercounting, documented in several large departments, starves the system of its raw material.
How are thresholds set, and what happens after a flag?
Thresholds vary by department and are often arbitrary in origin: three complaints in 90 days, a fixed count of force reports per quarter, a set number of pursuits. The National Institute of Justice's published guidance notes that threshold-setting has more tradition than evidence behind it. After a flag, the standard sequence is supervisor review, an assessment meeting with the officer, and a non-disciplinary intervention: counseling, training referral, a mentoring assignment, or a monitoring period. The distinction matters legally — because EIS is designed to be non-punitive, unions have generally accepted it, and officers' bills of rights in most states do not restrict it the way they restrict internal discipline. Whether the intervention happens at all depends on supervision: monitoring reports in decree cities have found flagged officers whose files show no recorded follow-up.
| Design choice | Common practice | Documented problem |
|---|---|---|
| Indicators | Force, complaints, pursuits, absenteeism | Quantity metrics catch busy officers, not necessarily risky ones |
| Thresholds | Fixed counts in fixed windows | Arbitrary; high and low performers flagged alike |
| Alert volume | Hundreds of flags per year in large departments | Supervisors triage; interventions go undocumented |
| Response | Counseling, training, monitoring | Follow-up completion rarely tracked or audited |
Peer comparison, in the monitoring record, changed the flag distribution more than any other adjustment: alerts that had clustered on officers in high-activity assignments spread more evenly once benchmarked against peers in the same work, and supervisors reported taking the alerts more seriously when the comparison was like-for-like. The finding, noted in decree monitoring reports and echoed in NIJ guidance, is a rare case where a design fix produced visible results in the alert data itself.
Related stories: What de-escalation policies require, and how compliance is measured · What use-of-force reporting requires, and where the gaps remain.
What does the evidence say the systems achieve?
The strongest documented findings are mixed. Studies of departments operating EIS, reviewed in the National Institute of Justice's summaries, have found that flagged officers did show higher prior rates of complaints and force — evidence the indicators track something real — while officers who received interventions did not consistently show lower subsequent complaint rates, with some studies finding no measurable change. A frequently cited structural finding across complaint research is that a small share of officers accumulate a large share of complaints, which supports the premise of targeted intervention. What no published study, per NIJ's summaries, has established is that a typical EIS, as implemented, reliably reduces serious misconduct. The honest characterization in the record is promising mechanism, unproven results.
Why do so many flags go nowhere?
The documented answer is supervisor capacity. Large departments generate hundreds of alerts a year, and intervention quality depends on a sergeant or lieutenant taking the meeting, documenting it, and following through — tasks monitoring reports have repeatedly found unrecorded in practice. In Seattle during its decree period, and in Chicago after 2019, monitors found alert backlogs and interventions closed with no outcome noted, patterns that turned the system into a logging exercise. Unions, for their part, have raised procedural fairness concerns about flags built on thin data, which some departments answered by adding a review step before any meeting occurs. Each of these friction points is documented in public reports, and together they explain the gap between the systems' premise and their results: the technology flags reliably; the human response is the part that never quite got engineered.
How do consent decrees use the systems?
Decrees treat EIS as a supervision instrument rather than a technology purchase. New Orleans's decree required a system with peer-group comparisons — an officer's force and complaint counts benchmarked against officers in similar assignments — precisely to fix the false-positive problem of raw counts. Seattle's decree-era system fed its broader data work, and Chicago's decree requires the city to publish system metrics, including flag and intervention counts, a transparency requirement most non-decree departments lack. Monitor reports in these cities have documented the recurring gaps: thresholds set too low or too high, alerts cleared without intervention, and no audit of whether interventions changed anything. The decree record's contribution is precision about implementation, which is where the academic literature had already located the decisive variable.
What would better systems look like, per the research?
The research record converges on four documented improvements: peer-relative rather than absolute thresholds; indicators of risk rather than volume, such as duplicate patterns in force narratives; audited completion of interventions, so a flag cannot simply expire; and outcome tracking — whether flagged officers generate fewer serious incidents afterward — published rather than buried. None is exotic, and departments under court supervision have implemented versions of each. The gap between best-documented practice and typical practice is the field's real measure: per the published record, a department running EIS has adopted the form of early intervention; whether it has the function depends on thresholds, supervision, and audit, not on the software.
For more context, read What de-escalation policies require, and how compliance is measured.
For more context, read consent decrees.
For more context, read internal affairs investigation.