Inspection kit
One or more Lensnode cameras with the optics and strobe ring chosen in the Feasibility Sprint, mounting brackets and guards, a Gatepoint G2 or G4, and the cabling to connect them.
Proof of Value
Six weeks on your line, judged in writing
A $24,000 fixed-price pilot with loaner hardware, success criteria signed before we arrive and a measured result at the end. If the system misses the criteria, you'll know exactly where and why.
The pilot
An industrial AI pilot program built to reach a decision
Industrial AI pilots have a habit of never ending. The system runs for months, nobody wrote down what success looked like, and the decision drifts until the person who championed it moves on. Our field note on why pilots stall describes the pattern.
A Proof of Value is built the other way round. It covers one inspection station or up to 25 monitored assets, runs for 6 weeks on your site, costs $24,000 fixed with loaner hardware included, and ends on a date agreed in advance with a measured result against criteria we both signed.
For inspection, it normally follows a Feasibility Sprint that has already shown the defect can be imaged. For condition and facility monitoring, where the sensing physics is well established, it can be the first paid step.
The system runs in shadow mode. It sees every part or reads every asset and makes every decision, but it does not fire your reject gate or change your maintenance plan. Your current process stays in charge, and the pilot is measured against it.
Week by week
What happens in each of the six weeks
-
Agree the criteria
We agree the success criteria in writing, collect your baseline figures for recent months — scrap, escapes, false rejects or downtime — and send a site preparation checklist. Nothing ships until the criteria are signed.
-
Install
Field engineers mount the loaner kit, connect power and network, and wire read-only signals from your PLC. By the end of the week the system is capturing images or signals from production.
-
Learn your line
For inspection, we adapt the model to images from your line and your quality team labels the borderline cases. For monitoring, each asset gets a baseline signature at its normal operating states.
-
Shadow and tune
The system decides on every part or asset without acting. Each day we compare its decisions with your inspectors or your existing alerts, investigate every disagreement and tune.
-
Measure
The model and thresholds are frozen. No tuning is allowed in the measurement window, so the result reflects the system you would actually deploy.
-
Report and decide
We present the measured result against each criterion, the data behind it and a deployment proposal. Then you decide — go, no-go or extend.
Success criteria
Example success criteria
Illustrative examples of the kind of criteria we agree in writing. Yours are set from your own baseline and the cost of being wrong on your line.
- Detection: at least 98% of seeded defect samples rejected, on every shift in the measurement window
- False rejects: below 0.5% of good parts during the measurement window
- Timing: every verdict available to the PLC within the agreed timing budget at maximum line speed
- Agreement: model and your senior inspector agree on at least 95% of a blind review set of borderline parts
- Alert quality: at least 80% of condition alerts confirmed as worth acting on by your maintenance lead
- Availability: system running and deciding for at least 99% of production hours in the measurement window
Loaner kit
What arrives on site
Loaner hardware is included in the fixed price. It is the same hardware a deployment uses, so the result carries straight over.
Monitoring kit
Up to 25 Pulsenode W, C or U nodes with mounting pads and clamps, and a Gatepoint gateway placed after a radio coverage check.
Software and access
Perflens Console running on the Gatepoint, user accounts for your team and a remote-support link you can switch off. Images and data stay on your site.
Your part
What we need from your site
- A named owner who can make decisions and sign the success criteria
- Baseline figures for the last few months, from your own records
- A mounting location, a power feed at the station and access during a planned stop or changeover for installation
- A network port in your OT network, or approval to run the kit standalone
- Two to four hours of a controls engineer's time to connect read-only PLC signals
- Inspectors or a maintenance lead to label borderline cases and review alerts, about an hour a day in weeks 2–4
- A site safety induction for our field engineers
After week 6
Go, no-go or extend
| Outcome | What it means | What happens next |
|---|---|---|
| Go | The criteria were met, or the gap is understood and acceptable to you | A deployment proposal with fixed prices, and 50% of the $24,000 fee credited to your subscription |
| No-go | The criteria were not met and we don't see a credible path to meeting them | We collect the loaner kit. You keep the results report, the data and a plain explanation of what fell short |
| Extend | The result is close and the cause is identified — a lighting change, more labeled defects, an unusual production week | An extension of agreed length and price, written down before it starts, against the same criteria |
How the credit works
Half of the Proof of Value fee — $12,000 — is credited to a Perflens subscription that follows it. The measured baseline from the pilot also makes shared-savings pricing possible, where part of the fee is tied to the reduction in scrap, escapes or downtime you actually achieve.
FAQ
Questions we hear often
Can the pilot affect production?
Not in shadow mode. The system reads signals from your line but does not write to your PLC or fire the reject gate. If you want it to act during the pilot, we agree that in writing with your controls team first.
Do we need a Feasibility Sprint first?
For AI visual inspection, usually yes — the sprint proves the defect can be imaged and gives the pilot a starting model. For condition and facility monitoring, you can go straight to a Proof of Value.
What if no machine fails during six weeks of monitoring?
That is the likely case, so condition monitoring criteria never depend on a failure happening. We measure what six weeks can show — baselines on every asset, alert quality, data completeness, and detection of faults that already exist or can be shown safely, such as a known air leak.
Who owns the data and the model from the pilot?
You own the data. The pilot model and its results report are yours to keep, and if you deploy, the pilot model becomes the starting point for the production model.
Can a Proof of Value cover more than one station?
The fixed price covers one station or up to 25 assets, which keeps the pilot short and the result clear. Larger pilots are possible and are quoted separately.
Agree what success means before we start
Book a discovery call and we'll draft success criteria for your line with you, before you commit to anything.