---
title: "Incident AI | ICAM & Incident Investigation Software"
description: "Structure incident evidence, timelines, interviews and actions into audit-ready reports with source citations and human review. ICAM-ready."
url: https://mineguard-ai.com/solutions/incident-ai
updated: 2026-11-20
language: en
publisher: MineGuard AI
---

[Safety investigations](https://mineguard-ai.com/solutions/incident-ai) [HR investigations](https://mineguard-ai.com/solutions/incident-ai/hr)

New: Agent Mode is rolling out to customers

# Faster, higher-quality incident investigations

Human-in-the-loop AI that helps safety teams run evidence-anchored investigations, with professional outputs built in. Without compromising judgement.

[Book a demo](https://mineguard-ai.com/demo)

Animation of four Incident AI screens: in Agent Mode an investigator asks for a draft to be shortened and the changes are shown, then a 5 Whys analysis with a source for every step, interview questions drafted for each witness, and approved findings becoming a briefing deck.

80%

Faster investigations

100%

Evidence traceability

3x

More consistent outputs

Encrypted

In transit & at rest

The Challenge

## Investigations fail for predictable reasons

Incident AI tackles these directly by making the investigation process structured, guided, evidence-centric, and output-ready.

### Time pressure

Investigations must be done quickly, often while running operations.

### Admin overload

Formatting, collating evidence, version control, and building reports is heavy work.

### Inconsistent quality

Different investigators produce very different outputs; teams can't compare apples to apples.

### Weak evidence linking

Findings and contributing factors are often not clearly tied back to evidence.

### Gaps in interviews

Key questions get missed; follow-ups happen late or never.

### Blame drift

Investigations can drift toward individual error instead of system conditions.

### Fragmented evidence

Emails, photos, PDFs, statements, and logs are scattered across systems.

### Weak corrective actions

Actions can be vague, not risk-based, and not measurable.

Workflow

## Watch an investigation take shape

Six steps, shown the way they happen in Incident AI. Each one plays as you reach it.

### See what’s missing on day one

Upload what you have. Incident AI compares it with what an investigation like this usually needs and flags the gaps while there is still time to collect them.

- Suggestions come from your own evidence set
- Close a gap, or record why it isn’t needed

Animation: a radar sweeps the evidence set and lists five pieces of missing evidence, each marked open.

### Conflicting accounts, side by side

When two sources can’t both be right, the conflict is listed with both versions and who said what. Nothing is quietly averaged away.

- Both accounts shown with their source
- You decide how each one is settled

Animation: a discrepancy register shows two statements that give different times for the same event.

### A timeline where every event has a source

Events are placed in order with the file each one came from. Gaps in the story are marked, not smoothed over.

- Each event links to its evidence
- Unexplained periods are flagged for follow-up

Animation: a timeline builds from four sourced events, then two gaps in the story are highlighted.

### Controls tested against the evidence

Each control from your register is checked against what the evidence shows: what held, what failed and what was missing.

- Every call comes with its reason and source file
- What worked is recorded, not just what failed

Animation: five controls between a hazard and an event are tested one by one, then listed with the evidence behind each result.

### In your team’s language

The interface, the analysis and the exports work in English, Spanish, French and Turkish, so crews read findings in the language they work in.

- Switch language without redoing the work
- Exports follow the language you choose

Animation: an incident report switches from English to Spanish, French and Turkish.

### Reports in your own template

Approved findings go straight into your Word and PowerPoint templates, keeping your branding and layout.

- Your template, not ours
- Report and briefing deck from the same findings

Animation: a branded investigation report fills in section by section, followed by a briefing deck.

New: Agent Mode

## It does the legwork. You make the calls.

Agent Mode brings the whole investigation into one guided conversation. It reviews the evidence, proposes the next step and prepares each analysis. You direct the work and approve every result.

| Classic mode | Agent Mode |
| --- | --- |
| You open each analysis and run it yourself | It proposes a plan and a depth, with its reasons. Nothing runs until you start it |
| You edit each result line by line | You tell it what to change, and it shows you every change |
| You cross-check conflicting statements by hand | It sets out each conflict with its sources and asks you to decide |
| You check drafts against earlier decisions yourself | It checks every draft against the decisions you have recorded |

Classic mode remains available and unchanged.

### Change any output by instruction

The biggest change from classic mode. Instead of editing line by line, tell Agent Mode what you want: shorten it, make it to the point, remove the inferences, rewrite it without an assumption. It revises the output and shows you exactly what changed.

- Every change is shown, and earlier versions are kept
- Ask why it says something, and it quotes the source

Animation: an investigator types “Shorten it and take out the inferences.” Agent Mode removes two inferences, shortens a sentence, shows the comparison, and the investigator approves the new version.

### It asks before it plans

Agent Mode reads every statement, report, photo and procedure first. Where accounts disagree, it won’t plan around a guess. It shows each conflict with its sources and asks you to decide.

- Settle a conflict, or leave it open on the record
- Every later analysis reads your decision

Animation: twelve evidence files are read and three pairs of statements are found to disagree. Agent Mode holds back the plan and asks whether the operator radioed a warning, showing the source for each answer. The investigator’s answer is added to their decisions.

### Push back, and it answers with evidence

Before each analysis runs, Agent Mode lists the facts it will rest on. Challenge one, and it responds with the supporting evidence and asks whether the fact is accurate as written.

- Confirm or correct the facts before a step runs
- It doesn’t simply agree. It shows its sources

Animation: Agent Mode lists five facts a Contributing Factors analysis will rest on. The investigator comments that a geotechnical engineer wasn’t needed, and Agent Mode replies with two pieces of evidence and asks whether the fact is accurate as written.

### Every draft checked against your decisions

When a draft is ready, Agent Mode checks it against the decisions you have recorded. Anything not yet reflected is flagged and can be added in one step, with the changes shown.

- Nothing in a draft contradicts what you have settled
- You see the changes before you approve

Animation: a Brief is checked against the investigator’s decisions and two are found missing. The investigator adds them, the changed sentences are shown, and the Brief is approved.

### Nothing reaches the report until you approve it

Every result waits for an investigator: approve it, edit it, run it again or leave it as a draft. The PDF report and the shared learning deck are built from approved work only.

- Agent Mode never approves anything for you
- Only approved work is saved to the case record

Animation: the options for a result are shown in turn: approve, edit, run again and leave as draft. The eight steps of the investigation plan are approved one by one, then the PDF report and the shared learning deck are built from approved work only.

Agent Mode is rolling out to Incident AI customers now. See it on your own kind of incident.

[Book a demo](https://mineguard-ai.com/demo)

Capabilities

## See every capability in action

From evidence ingestion to professional reports: watch how Incident AI accelerates each step of your investigation.

### Evidence Ingestion

Bulk upload documents, emails, PDFs, and statements. Evidence is structured and indexed automatically to eliminate manual collation.

### Video transcript

Bulk upload documents, emails, PDFs, and statements. Evidence is structured and indexed automatically to eliminate manual collation.

[Download Example](https://mineguard-ai.com/IncidentAIExamples/EvidenceingestionExample.pdf)

### Evidence Review

Quickly surface key facts across large evidence sets. Trace every finding back to its source for full defensibility and confidence.

### Video transcript

Quickly surface key facts across large evidence sets. Trace every finding back to its source for full defensibility and confidence.

### Case Management

Centralised control for all investigation artifacts. Manage timelines, evidence, and outcomes in a single, structured environment.

### Video transcript

Centralised control for all investigation artifacts. Manage timelines, evidence, and outcomes in a single, structured environment.

### Timeline Builder

Structure events chronologically with 5-Whys style thinking. Automatically link timeline events to supporting evidence.

### Video transcript

Structure events chronologically with 5-Whys style thinking. Automatically link timeline events to supporting evidence.

### Table Editing

All tables can be edited now: add and remove columns and rows to finalise your incident within Incident AI.

### Video transcript

All tables can be edited now: add and remove columns and rows to finalise your incident within Incident AI.

### Interview Support

Generate structured interview questions based on the current evidence set and investigation context to ensure no detail is missed.

### Video transcript

Generate structured interview questions based on the current evidence set and investigation context to ensure no detail is missed.

### Incident Wise

Query your entire evidence set using natural language. Get immediate answers to complex investigation questions with source citations.

### Video transcript

Query your entire evidence set using natural language. Get immediate answers to complex investigation questions with source citations.

### Investigator Input

New

Investigators can now save their conversations at incident wise as evidence to steer with full auditability (saves time against manual edit).

### Video transcript

Investigators can now save their conversations at incident wise as evidence to steer with full auditability (saves time against manual edit).

### Report Generation

Overview tools developing: initial details and investigation team information can now be included on the professional PDF report.

### Video transcript

Overview tools developing: initial details and investigation team information can now be included on the professional PDF report.

[Download Example](https://mineguard-ai.com/IncidentAIExamples/PDFReportExample.pdf)

### Photo Library

Organise site photos and visual evidence with AI-assisted tagging and descriptions. Export professional visual annexures instantly.

### Video transcript

Organise site photos and visual evidence with AI-assisted tagging and descriptions. Export professional visual annexures instantly.

### Learnings Share

Distill investigation findings into shareable safety alerts and bulletins. Ensure critical insights reach the frontline faster.

### Video transcript

Distill investigation findings into shareable safety alerts and bulletins. Ensure critical insights reach the frontline faster.

[Download Example](https://mineguard-ai.com/IncidentAIExamples/LearningShareExample.pdf)

### Incident Overview

New

Continuous investment in tools that make investigations faster and more defensible, with new capabilities added monthly.

### Video transcript

Continuous investment in tools that make investigations faster and more defensible, with new capabilities added monthly.

Evidence and examples

## Review the outputs teams can audit

Download representative investigation artifacts, report examples, and security documents before a walkthrough.

### [Evidence ingestion example](https://mineguard-ai.com/IncidentAIExamples/EvidenceingestionExample.pdf)

See how imported statements, files, and source material are structured for traceable investigation work.

Open example PDF

### [Investigation report example](https://mineguard-ai.com/IncidentAIExamples/PDFReportExample.pdf)

Review a professional PDF report format with structured sections and evidence-ready outputs.

Open report PDF

### [Learning share example](https://mineguard-ai.com/IncidentAIExamples/LearningShareExample.pdf)

Preview how investigation findings can be packaged into a frontline learning share.

Open learning share

### [Security overview](https://mineguard-ai.com/docs/incident-ai-security-overview.pdf)

Understand the controls behind Incident AI, including data handling, privacy, and deployment posture.

Open security PDF

### [NIST AI RMF statement](https://mineguard-ai.com/docs/nist-ai-rmf-statement.pdf)

Read how MineGuard aligns AI risk management practices to governance and human oversight principles.

Open RMF statement

Hosting, database and AI inference can run in your region or your own cloud tenant. [How sovereign deployment works](https://mineguard-ai.com/trust#sovereign)

Why MineGuard AI

## Built different for safety-critical work

Incident AI isn't a generic AI tool adapted for safety. It's designed from the ground up for the realities of incident investigation in high-risk industries.

- **Evidence-anchored by design** Outputs are anchored to what's actually in the evidence set. No unsupported conclusions.
- **Human-in-the-loop** Supports judgement rather than replacing it. The investigator remains in control.
- **Built for safety-critical** Structured, defensible, consistent output orientation. Designed for high-stakes environments.
- **Consistency at scale** Standardise quality across sites, teams, and investigators. Compare apples to apples.
- **Reduces admin drag** Accelerates the unglamorous work (collation, formatting, drafting) so teams can focus on analysis.
- **Real investigation focus** Built around what teams actually do under time pressure. Not academic, practical.

Roadmap

## What we're building next

Continuous investment in capabilities that make investigations faster, smarter, and more defensible.

Planned

### Interview Recorder

Record witness interviews and turn them into accurate transcripts and structured statements, ready to use as evidence.

Planned

### Dashboards & Historical Analysis

Trend dashboards and analysis across your incident history, surfacing recurring causes and where to act next.

Common buyer questions

## Incident AI, ICAM and investigation governance

### What is Incident AI?

Incident AI is incident investigation software from MineGuard AI, built for mining and heavy industry. It structures evidence, timelines, interviews and corrective actions into ICAM-ready, audit-ready reports. Every statement cites the evidence it came from, and an investigator reviews and approves every result.

### Does Incident AI replace ICAM investigators?

No. Incident AI supports competent investigators by structuring evidence, surfacing gaps, drafting outputs and preserving source links. The investigation team remains responsible for findings, contributing factors, corrective actions and approvals.

### How does Incident AI support ICAM and 5 Whys investigations?

Incident AI helps teams organise evidence, build timelines, prepare interview questions, review contributing factors and check action quality. It does not force a single method; it supports human-led analysis with traceable material.

### What outputs can an investigation team generate?

Teams can generate evidence summaries, timelines, interview prompts, action reviews, learning shares and professional report material with links back to source evidence.

### What is Agent Mode in Incident AI?

Agent Mode brings the whole investigation into one guided conversation. It reviews the evidence, reports where the case stands, proposes a plan and prepares each analysis. Investigators direct the work, change any output by instruction and approve every result. Agent Mode is rolling out to customers now, and classic mode remains available.

### Does Agent Mode approve findings on its own?

No. Every result waits for an investigator, who can approve it, edit it, run it again or leave it as a draft. Only approved work is saved to the case record, and the report and shared learning deck are built from approved work only.

### How do investigators change an Agent Mode output?

By instruction instead of editing line by line. Investigators can ask it to shorten a result, make it more concise, remove inferences or rewrite it without an assumption. Agent Mode shows exactly what changed, keeps earlier versions and can quote the evidence behind each sentence.

## Prefer to have it done for you?

Our practitioners run investigations with Incident AI, in your method and your template.

[See the service](https://mineguard-ai.com/services/safety-investigations)

From the blog

## Related reading

- [When Hazards Combine: Why Mine Incident Investigations Must Reconstruct Control Interactions](https://mineguard-ai.com/blog/mine-incident-investigations-control-interactions)
- [Contractor Interfaces Are a Mine Safety System, Not a Procurement Checklist](https://mineguard-ai.com/blog/contractor-interfaces-mine-safety-system)
- [A Human in the Loop Is Not Yet a Safety Control](https://mineguard-ai.com/blog/human-in-loop-not-yet-safety-control)

[All articles](https://mineguard-ai.com/blog)

## Ready to transform your investigations?

See how Incident AI can help your team run faster, more consistent, evidence-anchored investigations with professional outputs built in.

[Book a demo](https://mineguard-ai.com/demo) See a sample report

No commitment required. See real outputs from real investigations.
