Evidence intake
File classification, de-duplication, source registers, document families and first-pass extraction.
Technology doctrine
Human-derived decisions. Machine-accelerated analysis.
We are building and piloting internal agentic workflows to reduce the time spent organising, searching and checking evidence—so investigators can spend more time testing what it means.
Plain-language status
This capability is being developed and tested. We will not imply production maturity where it does not yet exist.
Why build it
Forensic work has a throughput problem.
Investigators frequently lose valuable time to mechanical tasks: normalising files, locating duplicates, building first-pass chronologies, finding references and checking whether a question has been covered.
Those tasks matter, but they are not the judgment. Carefully governed agents can handle defined portions of the mechanical work, maintain a structured trail and surface candidates for human testing. The resulting decision remains a professional act for which a named person is accountable.
Our objective is not “AI-generated findings.” It is a better investigative operating system.
The emerging stack
Four layers of controlled acceleration.
Each layer is bounded, reviewable and subordinate to the mandate. Automation is used where it improves coverage or speed—not where it obscures responsibility.
Research orchestration
Multi-source collection, jurisdictional monitoring, citation capture and question-directed retrieval.
Pattern surfacing
Chronology candidates, exceptions, named-entity relationships and transaction-review priorities.
Workflow control
Coverage checks, open-question registers, consistency flags, review queues and handoff discipline.
Investigator challenge
Alternative explanations, source reliability, corroboration, materiality and contradiction testing.
Partner sign-off
Human-owned findings, stated limitations, decision implications and defensible communication.
The control model
What the system may do—and what it may never own.
A premium technology position is defined by restraint as much as capability. These boundaries are part of the design.
May assist
- Organise and index supplied material
- Propose classifications, relationships and chronology candidates
- Surface exceptions and unanswered questions
- Retrieve cited source material for review
- Check coverage and internal consistency
- Prepare structured drafts for investigator revision
May not own
- Final findings of fact
- Legal or regulatory conclusions
- Source credibility determinations
- Unsupervised alteration of evidence
- Materiality, intent or culpability judgments
- Client advice without identified human accountability
The point is not to remove the investigator. It is to remove the investigator’s avoidable delay.Audit Corridor technology doctrine
Development discipline
The roadmap is a governed practice, not a marketing claim.
We will expand only where testing demonstrates better speed, coverage or consistency without degrading confidentiality, provenance or human review.
Prototype narrowly
Begin with repeatable, bounded tasks whose outputs can be compared against a human-reviewed baseline.
Test adversarially
Look for false confidence, omitted evidence, citation errors, over-broad inferences and failure under messy source conditions.
Log and review
Keep the instruction, source, intermediate output, reviewer intervention and disposition visible where appropriate.
Deploy selectively
Use a workflow in live work only when the mandate, information controls and reviewer competence support it.
Collaboration call
Help build the investigative stack.
We welcome serious conversations with AI engineers, data scientists, forensic accountants, lawyers, investigators, compliance leaders and domain specialists who can challenge or extend this model. Different experience is an asset when the operating problem is difficult.
We are open about the stage: this is active development and selective collaboration, not a claim of a completed autonomous platform.