Client
A UK territorial police force, covering digital evidence handling and disclosure preparation for several thousand live cases across CID and frontline policing
Sector
Policing & Criminal Justice
Engagement
Consolidation of body-worn video, CCTV and digital media ingestion into a single evidence platform with automated redaction workflow - phased delivery over four quarters.
What the client needed
Body-worn video adoption across the force had grown faster than the systems built to manage its output. Footage from body-worn cameras sat in one vendor platform, custody suite and interview recordings sat in another, and CCTV or third-party doorbell footage handed in during an investigation was typically saved to a shared network drive with no consistent naming convention, because no single system was built to ingest all three at once. Disclosure officers preparing a case file under the Criminal Procedure and Investigations Act had to manually track down every relevant piece of media across these locations, then redact each item by hand using desktop software before it could be reviewed by the Crown Prosecution Service. By the time we were engaged, the force's own case progression data showed disclosure preparation had become the single largest source of delay in charging decisions on cases involving video evidence, and several trials had been adjourned specifically because disclosure-ready evidence packs were not completed in time.
How we worked
- Built a single evidence ingestion layer that pulls body-worn video, custody recordings and externally submitted digital media into one case-linked repository, replacing manual copying between systems and shared drives.
- Introduced AI-assisted redaction that proposes faces, number plates and other identifying detail for review before a human redaction officer confirms or corrects each suggestion, rather than requiring every frame to be marked up from scratch.
- Built a disclosure schedule view that tracks every item of evidence against CPIA relevance and sensitivity categories automatically as it's added to a case, replacing a spreadsheet the disclosure team had been maintaining by hand.
- Designed role-based access so investigating officers, disclosure officers and reviewing lawyers see the same case timeline with permissions matched to what each role is legally allowed to view and export.
- Ran the rollout by division rather than force-wide in one go, starting with the division carrying the heaviest body-worn video volume so the redaction workflow was proven under real load before wider rollout.
- Trained disclosure officers on the redaction review workflow before decommissioning desktop redaction tools, keeping both routes available in parallel until confidence in the new workflow was established.
Measured results
All figures verified with the client. Specific division, case and individual detail withheld in line with our standard confidentiality terms and the sensitivity of live criminal proceedings.
- Average time to prepare a disclosure-ready case file involving video evidence fell by more than half once AI-assisted redaction review replaced fully manual frame-by-frame marking.
- Disclosure officers report substantially fewer instances of evidence being missed from a case file, now that body-worn video, custody recordings and third-party media sit in one searchable, case-linked location.
- The force's case progression data shows a measurable reduction in charging-decision delays attributed specifically to disclosure preparation on video-heavy cases since the platform's rollout.
- Redaction officers now review and confirm AI-proposed redactions rather than creating them from scratch, materially reducing the backlog that had built up across the division with the heaviest body-worn video volume.
- The force's professional standards function has cited the platform's audit trail - showing exactly who viewed, exported or redacted each item of evidence - as a direct improvement to disclosure integrity assurance.
"We knew our disclosure officers were drowning in footage from systems that had never been designed to talk to each other. What's changed isn't just the speed - it's that a case file going to the CPS now has a complete, auditable trail behind it, and that's exactly what a disclosure officer needs to stand behind their own work in court."
Working on something similar?
If this engagement looks like the kind of problem you are facing, we would be glad to compare notes by email.
Why redaction review, not redaction removal, was the right target
It would have been possible to pitch this project as replacing human redaction entirely with automated output. We didn't, and we'd make the same call again. Redaction decisions in a live criminal case carry real consequences if a face or detail that should have been withheld is missed, and an automated system confident enough to skip human review is a liability the moment it's wrong once on a case that matters. What we built instead treats the AI-generated redaction as a first draft a trained officer confirms, corrects or overrides, which is why the time saving came from removing the manual marking-up work, not from removing the officer's judgement from the process. That distinction mattered enormously to the force's own professional standards function during sign-off, and we'd treat it as a non-negotiable design constraint on any similar system, not a cautious compromise.
The shared drive wasn't careless. It was the only tool available.
As with much legacy infrastructure in public sector organisations, it would be easy to frame the previous shared-drive workaround as poor practice. It wasn't - it was disclosure officers doing the only thing available to them when three separate systems held evidence for the same case and none of the three could reference the others. That workaround had been quietly absorbing coordination risk for years, dependent on individual officers remembering where everything for a given case actually lived. Replacing it wasn't really the headline goal of this project. Removing the need for that memory to live in any one person's head was.
Division-by-division rollout cost time we'd spend again
Rolling out force-wide in one go would have looked faster on a programme plan. We deliberately started with the division carrying the heaviest body-worn video volume, on the basis that if the redaction workflow was going to fail under real operational load, we wanted to find that out with close support in place on one division rather than discover it force-wide with disclosure deadlines already running against several hundred live cases. That ordering meant a longer rollout timeline than a single cutover would have needed. Given what a missed disclosure deadline means for a live prosecution, we'd choose the slower, risk-led sequence again without hesitation.
Lessons learned
The first lesson was that AI-assisted redaction earns trust fastest when it's positioned explicitly as a proposal a human confirms, not a replacement for the officer's judgement - that framing shaped adoption more than any accuracy metric did.
The second lesson was that a long-standing manual workaround, like the shared drive here, is usually evidence of a structural gap worth taking seriously rather than a habit to simply switch off.
The third lesson was that sequencing a public sector rollout by operational risk, not by ease of delivery, is worth the extra time it costs when the downside of getting it wrong involves a live criminal case rather than an internal process.
If your organisation is managing evidence or case material across systems that don't talk to each other, and disclosure or case preparation deadlines are slipping as a result, we would be glad to discuss what a consolidation programme like this might look like for you. Email sales@halfteck.com.