The flavor changeover does not start without evidence

Cola in the morning, grapefruit at midday, water in the afternoon. Every changeover means a rinse or clean-in-place, a label change, a code change and sometimes a container format change. The flavor changeover is the highest-risk moment of the day and simultaneously the one under most time pressure: the worst possible combination.

‹ See all cases of beverage bottling

Edge cameras verifying the finished syrup tank and the pipework after clean-in-place during a flavor changeover at a bottling plant
The problem

The riskiest moment of the day coincides with the biggest hurry.

The first batch of the new flavor can come out with carry-over from the previous one — color, aroma, off taste — or with the label of the product no longer being filled. Carry-over is not always visible, and a crossed label is only caught once pallets are already built. Either one turns a full production run into unsellable product. Someone signs that the line is ready, under the pressure of recovering the downtime. This is not a question of professionalism: there simply is no objective evidence to sign against. The whole responsibility sits on human judgement against the clock.

  • Cola in the morning, grapefruit at midday, water in the afternoon. Every change requires a rinse or clean-in-place, a label change, a code change and sometimes a container format change.
  • The first batch of the new flavor can come out with carryover from the previous one — color, aroma, off taste — or with the label of the product no longer being filled.
  • Carryover is not always visible. The crossed label is only caught once pallets are already built.
  • Either of the two turns a whole production run into unsellable product.
  • And somebody signs that it is ready with the pressure of recovering the downtime on top. It is not a professionalism problem: there is simply no objective evidence to sign against.
How it fits the IRIS system

Edge with JIDOKA AI — and SMED AI pulling the other way.

Edge compares against the reference state, point by point. iLEAN places fixed cameras at the critical points of the changeover: finished syrup tank, pipework and filters, filling head, label hopper and coder panel. The models distinguish rinsed from residue present, correct label from previous flavor's label, new code loaded from old code still running. JIDOKA AI: the line will not start until everything is clear and someone signs. No camera decides alone. The system blocks start-up until the critical points report clear and the quality lead signs against that visual evidence. Humans in command: the person still decides, but now decides while seeing. And the changeover record is created with no extra work.

JIDOKA AI and SMED AI offset each other on purpose: one prevents starting too early and the other orders the sequence. The changeover does not get longer — it gets ordered, and comes out shorter and correct.

See the full IRIS architecture →

Before and after

Today's flavor changeover versus the validated changeover

AspectTodayWith iLEAN Edge
What authorizes start-upA signature under pressurePoint-by-point conformity, then a signature on it
Carryover from the previous flavorNot always visibleBlocked before start-up
Crossed labelCaught with pallets builtCaught before the first pallet
Cleaning criterionVaries between shiftsSingle and reproducible
Changeover durationPadded with a marginGuided and measured by SMED AI
Evidence for the licensorThe signatureBefore and after photo of each point

Impact estimate

Estimated impact — to validate with your own numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

  • Estimated payback 6-12 months, depending on how many flavor changeovers you run per week.
  • Whole unsellable production runs from carryover or a crossed label: avoided.
  • Protection of the relationship with the licensor of franchised brands.
  • And the changeover does not get longer: combined with SMED AI, it gets ordered.

It avoids the cost of entire unsellable runs and protects the relationship with the licensor of the franchised brands. Combined with SMED AI the changeover does not get longer, it gets ordered. Estimated payback between six and twelve months depending on how many flavor changes happen per day. Estimate to be validated.

And the fair question from the production manager

«What if the camera blocks the line for no reason?» — that is the right objection, because a false stop during a flavor changeover costs real money. Conformity is checked against the reference state for that specific point and, in case of doubt, the system does not block silently: it escalates to the quality supervisor, who signs on the evidence. It stops on a verifiable discrepancy, not on model uncertainty.

[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.

Frequently asked questions

What people ask about validating the flavor changeover

Which points are watched during the change?

The critical ones for that specific change, defined at commissioning: the finished syrup tank, pipework and filters, and the label material loaded on the machine.

Does it catch carryover that is not visible to the eye?

That is what is trained: the model learns to distinguish relevant residue from the rest, using that plant's real points and flavors.

Doesn't it make the changeover longer?

The check takes seconds. What lengthens the changeover today is the margin added for lack of evidence; with it that margin can be trimmed, and SMED AI orders the sequence.

What about the wrong-flavor label?

It verifies that the loaded material matches the product on the order. That is the failure that today only appears once pallets are built.

Can validation be skipped when in a hurry?

An exception can be defined with a supervisor signature, and it is recorded as such. What cannot happen is starting up with no trace of who decided to skip it.

Let's talk

Tell us how many flavor changeovers you run per week and how long verification takes.

We work on your plant's real data, not ours. Assessment with no commitment.

Request estimated ROI within 48h ‹ See all cases of beverage bottling See food