The bottleneck in most traffic studies is not the cameras or the budget. It is the queue.
Video goes in. Analysts work through it one hour at a time, one file at a time. A two-week survey with footage from six junctions can take three to four weeks to process before the first count lands on your desk. By that point, the client is already chasing you, and you are still waiting on the manual counting contractor to finish classifying HGVs from a Thursday afternoon.
That backlog is not a people problem. It is an architectural one. Traditional counting workflows are sequential by design, and sequential workflows do not scale.
Why Sequential Processing Breaks Down at Scale
Manual counting and most legacy AI tools share the same fundamental constraint: one file gets processed at a time. The analyst (or the algorithm) works through footage chronologically, classifying vehicles and recording movements as the video plays. A one-hour recording takes roughly one hour to process, sometimes more when reviewers quality-check the output.
That makes project duration a direct function of footage volume. Double the number of survey sites, double the processing time. Add a second camera angle, add a second queue. For a busy consultancy running concurrent projects, this creates a genuine throughput ceiling that no amount of additional headcount fully solves.
The downstream effects compound quickly. Counts that should take days take weeks. Client deliverables slip. Modellers waiting on turning movement counts cannot start their Vissim builds. The whole project schedule shifts right because the data is not ready.
What Parallel Cloud Processing Changes
Cloud architecture breaks the sequential dependency. Instead of processing footage in order, a parallel system distributes work across many computing instances simultaneously. A ten-camera survey does not take ten times as long as a single-camera one. It takes roughly the same time.
GoodVision Video Insights applies this directly to traffic survey work. Upload your footage and the platform processes every file in parallel across cloud infrastructure. Regardless of how many hours of video you submit, results are ready within one to two hours. A single-junction survey and a twelve-site study return data at the same speed.
That is a structural change, not an incremental improvement. It means project timelines are no longer constrained by footage volume. A survey that would have taken a counting contractor three weeks to process manually can be turned around overnight. You get 95%+ accuracy on counts and classifications across eight vehicle classes, including motorcycles, cyclists, and HGVs, without waiting for anyone to watch the footage in real time.
The platform also reduces data footprint by 1,000x compared to raw video. Instead of moving large video files between contractors and storage systems, you work with structured trajectory data and processed counts. That reduces storage costs, speeds up transfers, and simplifies data management across projects.
What This Means for a Consultancy Running Multiple Projects
For traffic engineering teams managing several concurrent studies, parallel processing changes the economics of what is feasible. You are not choosing between fast turnaround on one project and slow turnaround on three. You get fast turnaround on all of them.
The implications for resourcing are significant. Manual counting contracts require substantial coordination: briefing crews, managing field logistics, waiting for submissions, and reviewing outputs for errors. Parallel AI processing removes most of that overhead. You upload footage, set your classification requirements, and receive structured data that goes directly into your models.
A+S, a mobility planning consultancy, moved to this model for volume traffic studies and found that junior engineers could manage the full data pipeline without specialist training. Time previously spent on counting logistics now goes toward analysis and client deliverables.
For Ramboll, which processes drone survey footage at scale, the ability to run multiple video files in parallel means post-processing no longer adds days to drone study timelines. Footage that comes off the aircraft goes straight into the platform and comes back as trajectory data, classified by vehicle type, ready for microsimulation inputs.
How Mott MacDonald Applied This in Manila
Mott MacDonald used Video Insights on a complex urban traffic study in Manila, a context where manual counting is particularly difficult given traffic density and the number of vehicle classes in the mix.
The parallel processing capability meant that high volumes of junction footage from multiple simultaneous survey points were processed within hours rather than weeks. The output included classified turning movement counts across all eight vehicle classes, giving the modelling team the data needed to calibrate transport models to local conditions. The speed of processing allowed the team to iterate on survey coverage rather than committing to a single data collection pass and hoping the footage was usable.
That kind of flexibility is only possible when processing time is not the constraint.
The Broader Point
Traffic surveys have historically been designed around the slowest part of the workflow: the analyst watching the video. Everything else (camera placement, survey duration, the number of sites) was negotiated against how long it would take to extract the data.
Parallel cloud processing inverts that. Processing is no longer the constraint. You can survey as many sites as the project requires, for the duration the study demands, and the data will be ready within two hours of upload. Survey design can be driven by what the study actually needs, not by what the counting budget can absorb.
For consultancies bidding on competitive tenders, that matters. Lower data collection overhead and faster turnaround translate directly into competitive pricing and tighter delivery commitments, without sacrificing accuracy or classification depth.
The counting bottleneck is a legacy of sequential tooling. Parallel AI processing removes it.
Book a demo at goodvisionlive.com/request-demo/ and see how Video Insights handles your full survey footage load from the next project you have in the queue.