Client
A UK port operator's freight and customs processing arm, delivering primary checks and risk-based vehicle selection under contract at a major roll-on roll-off ferry terminal
Sector
Border Security & Customs
Engagement
Freight risk targeting platform linking pre-lodged manifest data, automatic number plate recognition and intelligence-led referral alerts into one shared operational view - multi-quarter programme.
What the client needed
Deciding which freight vehicle to pull aside for a physical check depends on three sources of information arriving on three different clocks: the manifest and customs declaration lodged before the vehicle ever reaches the port, the automatic number plate recognition match logged as it rolls off the ferry, and an intelligence-led referral that can land from a national system at any point, sometimes minutes after the vehicle has already cleared ANPR. Before this engagement, those three feeds sat in three separate systems that nobody was cross-checking in real time, so a referral arriving a few minutes late - which happened more often than the client wanted to admit - meant the matching vehicle had already been waved through primary check and was on the road before anyone realised there had been a live match at all. The client's own incident reviews showed this happening often enough to be treated as a routine operational risk rather than a rare failure, and there was no shared view that could tell a controller, in the moment, whether a vehicle rolling toward the barrier right now had an outstanding referral against it.
How we worked
- Integrated with the port's existing ANPR camera feed and the customs pre-lodgement system, both of which already held the data needed but had never been queried against each other in real time.
- Built a live matching layer that checked every ANPR read against outstanding intelligence-led referrals the instant the read occurred, rather than on the periodic batch reconciliation the client had been relying on.
- Added a hard-stop barrier trigger for a confirmed match, holding the vehicle at the physical barrier automatically rather than depending on a controller noticing an alert on a screen among many other open tasks.
- Built a controller-facing queue showing every vehicle currently between ANPR read and barrier exit, so a referral landing in that window could still be actioned before the vehicle left rather than logged as a miss afterwards.
- Ran the matching layer in shadow mode alongside the existing batch reconciliation process for a full quarter, comparing every case where the two approaches would have produced a different outcome before the hard-stop trigger went live.
- Trained primary check controllers and customs officers together on the shared queue, since the barrier trigger only reduces risk if the officer stopping the vehicle understands why it was held.
Measured results
All figures verified with the client. Specific site, personnel and case detail withheld in line with our standard confidentiality terms and border security data-handling requirements.
- Vehicles waved through primary check despite an outstanding intelligence match dropped from a routine occurrence to a rare, individually reviewed exception once the shadow-mode comparison period ended and the hard-stop trigger went live.
- The controller-facing queue meant a referral landing after ANPR but before barrier exit could be actioned in the moment in the large majority of cases, rather than being logged afterwards as a missed interception.
- The shadow-mode quarter itself surfaced a systemic timing gap between the national referral feed and the port's own batch reconciliation cycle that nobody had previously quantified, informing a separate fix to the referral feed's delivery timing.
- Customs officers report substantially more confidence acting on a barrier hold, since the shared queue shows them the underlying manifest and referral detail rather than an alert with no context.
- The client is extending the same real-time matching approach to a second RoRo berth at the same port, where vehicle volumes and the referral timing problem are similar.
"We had all three pieces of information somewhere in our systems - the manifest, the ANPR read, the referral - and every single one of those near-misses came down to the same thing: they just didn't reach the same person at the same moment. Nobody was cutting corners, and no officer ever ignored a referral they could see. The referral simply wasn't visible yet when the vehicle was still in a position to be stopped. Closing that gap was the entire job."
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.
Context and constraints
A freight risk targeting decision at a busy RoRo terminal has to be made inside a window measured in seconds once a vehicle has cleared ANPR and is moving toward the barrier, not the minutes a controller might have when reviewing a case at a desk. The brief was explicit from the outset that the platform had to work with the port's existing ANPR camera estate and the customs pre-lodgement system already in use rather than replace either - both were established, audited systems, and the actual gap was that the data inside them was never being cross-checked against each other at the speed the physical operation actually moved.
Border security data-handling requirements shaped the build throughout. Manifest, referral and vehicle movement data all carry their own sensitivity and access rules, and the platform was built so a primary check controller saw a barrier hold and enough context to act on it, while the underlying intelligence detail behind a referral stayed restricted to customs officers cleared to see it. Getting that access boundary right, and agreeing it with the client's own security accreditation process, took as long as building the real-time matching logic itself.
Choosing a hard stop over a screen alert
The hardest design decision wasn't the matching logic - once ANPR reads and referrals were flowing into the same system, identifying a match was comparatively simple - it was deciding what should happen automatically at the moment of a match. An alert on a screen depends on a controller looking at the right moment among many other open tasks during a busy sailing; a vehicle that has already passed the barrier can't be un-waved-through. We worked with the client's operations and legal teams to define a hard-stop trigger at the physical barrier itself for a confirmed match, accepting the operational cost of occasionally holding a vehicle for a few extra seconds against the alternative of a missed interception that couldn't be recovered.
A full quarter in shadow mode before anything could stop a vehicle
We ran the real-time matching layer in shadow mode, logging what it would have flagged without actually holding any vehicle, for a full quarter before the hard-stop trigger went live. That comparison period, run against the client's existing batch reconciliation process rather than a shorter and more convenient sample, is what surfaced a systemic timing gap between the national intelligence referral feed and the port's own reconciliation cycle - a delay nobody had previously quantified because the batch process simply absorbed it as normal variation. That finding fed directly into a separate fix to how quickly referrals reached the port at all, independent of anything the platform itself could do once a referral arrived.
Lessons learned
The first lesson was that having all the necessary data somewhere in an organisation's systems is not the same as having it in front of the person who needs to act on it at the moment they need to act - the manifest, the ANPR read and the referral all existed well before this engagement, and none of them were the problem.
The second lesson was that an automated hard stop is only defensible once you have run the underlying logic in shadow mode for long enough to trust its false-positive rate, and a full quarter, not a shorter pilot chosen for convenience, was what it took to build that confidence with the client's own operations team.
The third lesson was that a shadow-mode comparison period is worth running for what it reveals about the wider system, not just the platform itself - the referral feed timing gap it surfaced was arguably as valuable an outcome as the hard-stop trigger the engagement was originally scoped to deliver.
If your organisation is weighing whether a time-critical decision depends on data that already exists but isn't reaching the right person fast enough to act on it, we would be glad to discuss what a programme like this might look like for you. Email sales@halfteck.com.