Pilot in progress
UAE’s first Dubai chocolate manufacturer to pilot A-Eye
A Dubai chocolate manufacturer is working with Rubicon to test A-Eye on its live production line. This page explains what we are testing, what we are trying to achieve together, and how the workflows are being shaped around the way the business actually runs.
Pilot in progress. This is not a finished case study and no outcomes are being claimed yet. What follows is the intent of the pilot and the work underway, published while it happens rather than written up afterwards.
What we are testing
The question behind this pilot is straightforward: can continuous visual monitoring give a manufacturer a reliable picture of stock without adding work to the people already running the floor.
Whether the picture holds up
Production and storage areas are busy, and conditions change through the day. We are testing whether A-Eye keeps an accurate view of what is where as normal operations carry on around it.
Whether it fits the day
A system that produces good information at the wrong moment gets ignored. We are testing when information is genuinely useful during a working shift, and routing it to arrive then.
Whether it stays out of the way
The pilot succeeds only if the team does not have to service it. Anything that asks staff to feed the system defeats the purpose of installing it.
Whether it earns trust
A system becomes genuinely useful at the point where the team relies on it without checking behind it. That confidence comes from weeks of ordinary use, which is what a pilot period is for.
What we are trying to achieve together
The aim is not to replace how the business works. It is to remove the parts of the day that exist only because nobody can see the current position.
Less manual counting
Counting stock by hand is necessary today because there is no continuous view. The goal is to make routine counting the exception rather than the routine.
Knowing sooner
Most stock problems are manageable when they are found early and expensive when they are found late. Shortening that gap is the practical value.
A cleaner handover
Production, storage and dispatch each hold part of the picture. The goal is one shared view all three can work from.
Fewer surprises at close
Reconciling at period end is painful when discrepancies have been accumulating quietly. Catching them as they happen changes the character of that work.
Confidence in the numbers
Decisions get delayed when nobody fully trusts the stock position. Trustworthy information is what makes faster decisions possible.
Something the team wants to use
Adoption is the real test. If the floor finds it useful without being told to, the pilot has worked.
Polishing the workflows
Most of the effort in a pilot like this is not the detection itself. It is fitting the information into how people already work, which is the part that decides whether a system survives past the trial.
| What we are shaping | The question being answered |
|---|---|
| Who gets told | Which role is genuinely able to act on each type of information, so nothing lands with someone who cannot do anything about it. |
| When it arrives | Whether something needs attention immediately, at a shift change, or in a daily summary. Not everything is urgent, and treating it as though it is causes alerts to be ignored. |
| How much detail | Enough context to act without a person having to go and look for the rest of the story. |
| What counts as normal | Every facility has its own rhythm. Tuning the system to the site’s normal is what stops routine activity being reported as a problem. |
| Where it goes next | How the information reaches the systems the business already runs on, so it becomes part of the existing process rather than another screen to check. |
| What happens on a quiet day | A system is only trusted if the team knows the difference between nothing wrong and nothing working. |
Why a chocolate facility is a demanding place to prove this
Chocolate production is a genuinely challenging setting for visual monitoring, which is part of why this pilot is worth running and worth documenting.
Products that look alike
Ranges often share a single look across several variants. Telling them apart consistently is a much finer task than distinguishing obviously different goods.
Packaging that plays with light
Wrapping is designed to catch the eye, which makes it harder for a camera to read than plain outer packaging.
A system that performs well in this setting has been asked harder questions than one that only has to cope with a conventional warehouse. That is the reason for starting here rather than somewhere easier.
What has happened so far
Facts from the deployment as it runs. No performance claims are made below, because the validation work that would justify them has not been completed yet.
Figures as at 26 August 2026, while the pilot is ongoing.
What the deployment shows today
Setting performance aside, the pilot has already established several things about how this kind of system fits a working factory.
It works with the cameras already there
The facility’s existing camera was enough to begin. Nothing was fitted to the line and nothing about how the line is operated had to change. The only equipment we added was a small device on site to run the processing.
It learns the customer’s own products
The system was taught from photographs of the real items running on that line, rather than matching against a generic product catalogue.
Nothing leaves the building
Video is processed on site as it happens and is not recorded. Only reference snapshots are kept. That answers the question most manufacturers ask first, which is where their video ends up.
It started producing output quickly
First detections were recorded within three days of connecting, so the team could see real output early rather than waiting weeks for something to look at.
The team can see it live
A browser-based dashboard at the site shows the live view, current counts and product history, without anyone needing to request a report.
Every count keeps a picture
Each detection stores a reference snapshot of what the system saw at that moment, so counts can be checked against real images rather than taken on trust.
That last point matters more than any figure we could publish today. A count you can check against a picture of the moment it happened is a different proposition from a number a system simply asserts, and it is what makes the verification work ahead meaningful.
The log
Scope agreed and pilot underway
Areas to be covered agreed with the team, objectives set, and the pilot started. Published at the beginning deliberately, so what we set out to do is on record rather than described after the fact.
Working on the floor
What the production environment taught us in the first weeks, and what it changed about the original plan.
Fitting the routine
How the information found its place in the working day, and what had to be adjusted for the team to actually use it.
What we learned
What worked, what did not, and what we would do differently in the next deployment of this kind.
About this pilot
Is A-Eye a finished product or is this a trial?
A-Eye is deployed and running elsewhere. What is being trialled here is how it performs in a chocolate manufacturing environment specifically, and how the workflows around it should be shaped for this kind of operation.
Why publish a pilot before it has finished?
Because a page written only after a good outcome tells you very little. Publishing during the work means the intent is on record from the start, and whatever we learn can be reported honestly rather than shaped to fit a conclusion.
What happens if it does not work as hoped?
It gets written up here, with the reasons. A section that only ever reports success would not be worth reading.
Does this require new equipment?
A-Eye is built to work with camera coverage that is already in place wherever that is practical. What a specific site needs is assessed before anything is committed.
Will the client be named?
The pilot is running with a Dubai chocolate manufacturer, described as they have asked to be described. Anything further is theirs to share rather than ours.
Can we run a pilot in our facility?
Yes. The approach is the same: agree what we are trying to achieve, shape the workflows around how your team already works, and report honestly. Get in touch and we will tell you whether your environment suits it.
Considering a pilot in your facility?
If you run production or storage and your stock position is only as current as the last manual count, the same approach applies. We will tell you what is realistic in your environment before anyone commits to anything.