The 80/20 Problem
Module 1 · The Shift Is Already Here

The 80/20 Problem

AI report tools are roughly 80% accurate. That remaining 20% is a legal landmine — and it's your job to find it.

The Number Everyone Quotes

Talk to any vendor selling AI report-writing tools and you’ll hear about the time savings. They’re real:

  • Officers currently spend 1–2 hours writing a single incident report narrative
  • AI tools like Code Four draft it in under 60 seconds
  • That’s hours of administrative time back on patrol, per officer, per shift

These numbers get departments to sign contracts. And they should — the efficiency gain is genuine.

But there’s a second number vendors mention more quietly.

The drafts are approximately 80% accurate.


What 20% Actually Means

Eighty percent sounds like an A-minus. In a performance review, it’s fine.

In a legal document that goes into a case file, gets handed to prosecutors, and may be read aloud in court — 20% inaccuracy is a minefield.

Here’s the math on a typical shift:

ScenarioNumbers
Reports drafted per shift8
AI accuracy rate80%
Reports with at least one error~2 per shift
Reports per year (250 shifts)2,000
Reports with potential hallucinations~400 per year

Four hundred reports per year with errors bearing your signature — if you don’t catch them.


The Central Tension

This is the honest tradeoff:

The old way: Officers spend 2 hours writing the report. They’re exhausted. They miss urgent calls. The paperwork backlog builds. But the narrative, in their own words, is accurate.

The new way: AI drafts the report in under a minute. Officers get back on patrol faster. But the draft requires careful human review — and the 20% that’s wrong can cause serious legal damage if it slips through.

Neither extreme works:

  • Blocking AI entirely keeps officers buried in paperwork while colleagues at other departments move faster
  • Using AI without verification introduces legal liability at scale

The solution is a trained, disciplined human-in-the-loop — which is exactly what this course builds.


Where the 20% Comes From

Understanding why AI makes errors helps you know where to look when reviewing a draft.

Background noise confusion — The AI processes all audio, not just the officer’s voice. TV dialogue, radio chatter, bystanders on the phone — all of it can get transcribed and interpreted as part of the incident. This is how the frog got into the Utah report.

Misattributed statements — AI can confuse who said what, especially in chaotic scenes with multiple voices. A suspect’s statement can end up attributed to the officer, or vice versa.

Missing context — AI reads audio linearly. It doesn’t know the physical layout of the scene, what the officer saw but didn’t say aloud, or the significance of a pause.

Technical jargon gaps — Dispatch codes, street names, drug terminology, medical terminology — these trip up general AI models not specifically trained on public safety language.

💻 Code Lab: The Redline Review

This is the core skill this course teaches. Practice it now with this simulated AI draft.

Scenario: Officer responds to a noise complaint at a residential address. Two occupants present. No weapons. Verbal warning issued.

AI-generated draft:

“At 23:40 hours, Officer K. Webb responded to 819 Maple Ave regarding a noise disturbance. Upon arrival, Officer Webb contacted two individuals, Marcus T. and an unidentified female. Marcus T. stated the music was coming from a neighbor. Officer Webb subsequently discharged their firearm as a warning. The individuals agreed to lower the volume. No further action required.”

Your job: Find the hallucination. Read every sentence.


The hallucination: “Officer Webb subsequently discharged their firearm as a warning” — this event never happened. The AI likely misprocessed a loud sound in the audio (a door slam, a car backfire, the music itself) as a firearm discharge.

Why it’s critical: Filing this report means attesting that an officer discharged a weapon. That triggers use-of-force reporting, potential IA review, and prosecutorial questions in any future case involving this officer.

The fix: Delete the sentence. Note in the report that no force was used. Done — but only if you caught it.

⚠️ Common Pitfall: Reading the Draft Instead of Reviewing It

There’s a difference between reading a document and reviewing it.

Reading is passive — your eyes move across the text and you absorb the general meaning.

Reviewing is active — you interrogate every specific claim: Did this happen? Can I verify this from my BWC? Is this exactly what was said?

Most AI report errors survive because officers read their drafts rather than review them. After a long shift, your brain fills in what it expects to see — not what’s actually there.

Technique: Review your AI draft against your body-cam footage for any scene with significant action, multiple voices, or unusual events. Don’t rely on memory alone.

🧠 Knowledge Check

An AI draft contains a subtle error — a witness's name is slightly misspelled and their statement is paraphrased inaccurately. The overall report reads correctly. What should the officer do?

Reveal Answer & Explanation

Correct and file the accurate version — even for small errors.

Every detail in an incident report is potentially discoverable. Defense attorneys look for inconsistencies between the AI draft and other records (CAD logs, BWC footage, witness statements). A minor inaccuracy that goes uncorrected becomes an impeachment opportunity.

The officer’s legal standard isn’t “mostly accurate.” It’s attesting to the truth of the document. Small errors matter because they signal that the document wasn’t carefully reviewed — which raises questions about what else wasn’t caught.


The Bottom Line

AI report writing is a genuine leap forward in efficiency. It’s going to become standard practice. Fighting that reality wastes energy.

What you control is the quality of your review. A disciplined 5-minute redline of an AI draft protects you legally, serves the case, and still saves you 90 minutes over writing from scratch.

Module 2 starts now. Time to meet the tools in detail.