Home / Audit
Two weeks. We tell you exactly why your pilot stalled, and what it takes to fix it on your footage, your hardware and your sites.
Week one we reproduce the failure ourselves. Week two we cost the fix and write the spec.
Fixed scope and fixed price. No hourly billing, no change orders.
Three documents at the end of the two weeks. They are yours; take them to your own team, to your existing vendor, or to anyone else.
1
Diagnosis
A scored diagnosis
Every failure we can reproduce, ranked by what is holding up the rollout rather than by what is easiest to fix. Each with the evidence behind it: the measurement, the footage, the device trace.
Written for engineers, with a summary your leadership can read.
2
Plan
A rollout plan
What has to change, in what order, and what it costs to run per site. Sequenced so the things that could still sink the rollout are settled first, not last.
Costed across your estate, not extrapolated from one pilot site.
3
Spec
An acceptance specification
The numbers that define working for your system, and where each one gets measured at the PLC rather than at the model, over a full shift rather than a clean run.
This makes the next phase provable instead of arguable.
Three we’ve found before, anonymised. None of them were the model.
Logistics · edge inventory monitoring
Three models, one device, no thermal headroom
Fine in testing. Twenty minutes into sustained load the device throttled and the detection rate quietly halved. Nothing alerted, because nothing was watching.
Rebalanced the models across the device and fixed the thermal envelope. Held through a full shift.
Manufacturing · defect detection on a line
The model was fast. The pipeline wasn’t.
Everyone was optimising inference. Decode and colour conversion were eating most of the frame budget before the model ever ran.
Reworked the pipeline instead of the model. Frame rate recovered on the hardware already installed.
Retail · store analytics, India
Tuned for one store, wrong in the rest
Thresholds set by hand against one store’s cameras and lighting. Every new store needed an engineer to make it work.
Replaced fixed thresholds with per-site calibration. New stores configure themselves.
Anonymised at our clients’ request. Work shipped by our engineers, at Cortal Insight and before it.
Four things. The audit is scoped so your team doesn’t stop working while we run it.
Representative footage from the sites that are failing
Specifications for the hardware already installed
Around an hour with the engineer or vendor who built the pilot
Access to one representative site, in person or remotely
We sign your NDA and data processing agreements before we access anything. Footage stays in your environment wherever that is technically possible.
And when it isn’t. We would rather tell you on the call than two weeks in.
Worth it if
It works at one site and not at the next ten
It works, and nobody can scale it without an engineer per site
It stopped working and nobody can tell you why
It detects correctly, and nothing downstream ever acts on it
The people who built it are gone
Not yet if
You don’t have a model that works anywhere yet
There is no footage from the sites you intend to run on
You’re still deciding what the system should do
The cameras aren’t installed
Those are earlier problems, and an audit won’t solve them. Tell us on the call and we’ll point you at what will. And sometimes the honest answer, two weeks in, is that the pilot shouldn’t continue, we’ll say so, and show you why.
Pricing
Fixed price, quoted after a 30-minute call and scoped to the sites and systems involved.
No hourly billing, no change orders. Most audits lead to a sprint; some don’t, and the audit is priced to stand alone either way.
A 30-minute call with an engineer, not a salesperson.