Nothing reaches the ERP without two human taps

The problem

The balance point is not in capturing: it is in validating.

The real reason the dock is not automated is not feasibility: it is that if the AI miscounts and that number enters inventory, the site manager ends up paying for the variance. But the dock supervisor is not going to type in every load either. The balance point is not capture: it is validation.

• The real reason the dock is not automated is not that it cannot be: it is that if the AI miscounts and that number reaches inventory, the depot manager ends up paying for the variance.
• But the dock supervisor is not going to type every load either: at bay cadence, any manual capture happens late, happens badly or does not happen.
• Hence the dilemma that blocks the case: either unvalidated data contaminating the master everyone consults, or data never captured out of distrust.
• And it is an expensive master to fix: product and returnable container inventory is consulted across the whole network.

How it fits the IRIS system

Connect with early human verification — a two-second summary.

A tablet on the bay and a two-second summary. Connect in early human verification mode puts an industrial tablet on the loading bay and at the route settlement point. Everything iLEAN captures by any route — the photo of the load sheet, the vision count of the pallet, the supervisor's dictation, the depot alert — appears there as a visual summary designed to be validated in two seconds. Discrepancies are shown, not silently resolved. If iLEAN counted thirty-eight cases and the sheet says forty, the tablet displays both figures and the photo behind them. The supervisor confirms with one tap or corrects with another. Only then does the data cross into the ERP. Latency of seconds, no training required and no change of habit for anyone.

The interface is designed for validating, not for entering data. Asking a dock supervisor to fill in a form with a truck waiting is asking them not to do it.

See the full IRIS architecture →

Before and after

The three ways of resolving it, compared

Aspect	Autonomous AI	The supervisor types
Inventory protected	No	Yes
Actually done at cadence	—	No: late, badly or never
Supervisor's time	Zero	Minutes per load
Who answers for the variance	Nobody, so it is not approved	The depot
Signature per movement	No	Yes
Approvable by the board	No	Yes, but unused

Impact estimate

An enabling piece — no payback of its own, and stated as such.

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.

• This case has no payback of its own: its value is that without it none of the other pieces gets approved.
• With it, inventory stays clean and every movement has a human signature behind it.
• That is what both an audit and the manager answering for their depot's variance require.
• And it removes both unvalidated data reaching the ERP and data never captured out of distrust.

Its value is not a direct saving but the fact that without it none of the other pieces get signed off. With it, inventory stays clean and every movement carries a human signature, which is what an audit and plain common sense both demand from whoever answers for the variance. Strategic value, not monetisable in isolation.

And the fair question from the production manager

«If a person signs in the end, what is the AI saving?» — the work is not in deciding, it is in counting and assembling. Today somebody would have to count product cases and empty containers by format and type it in. With the verification gate it all arrives counted in a two-second summary and they only confirm or correct. Every correction is stored, and the system learns where its counting fails.

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