AI traceability agent: a first in a charcuterie maker's meat cutting room

An AI traceability agent lets operators in a meat cutting room photograph the supplier label and dictate the cuts they produce with their actual weight, instead of retyping everything. The agent structures this information, has a supervisor check it, then records it in the ERP. At a charcuterie client, it runs on all three stations in the cutting room.

This article tells the story of that rollout from the inside: how the cutting room worked before, what the agent does at each station, what it pays back, and what you need to do the same in a meat or charcuterie SME.

Why meat traceability is hard for cutting rooms in SMEs

Traceability is not optional: EU Regulation 178/2002 requires businesses to identify, at any time, where their raw materials come from and where their products go (Article 18, EUR-Lex). In meat cutting, this is the weakest link, because a whole bovine carcass becomes around thirty cuts, and each cut has to stay linked to its original lot.

Large groups industrialise this with connected scales, automated lines and, increasingly, AI. Danone, for example, describes a control tower that combines a digital twin with agentic AI to run a machine from a tablet (McKinsey). An SME with a few dozen employees has neither that budget nor that equipment. It has operators, knives, unconnected scales and an ERP.

That is exactly our client's situation. And that is where an AI traceability agent makes a difference.

Before the agent: a cutting room run by hand

At our client, meat arrives in several forms: whole carcasses of beef, veal, game and lamb, but also pork, poultry and some beef already cut by the supplier. The cutting room receives 5 to 6 deliveries a day, each with 5 to 6 product references, and around 5 whole bovine carcasses a week.

The work is split across three stations. At the first, goods-in, the operator checks the goods against the delivery note, creates the internal lot (three letters for the supplier, the receipt date, the item code) and enters all the additional traceability data into the ERP, Euragro: category, breed, type, origin, country of birth and rearing, slaughterhouse number, cutting plant number, use-by date, supplier. This information is on the supplier label, but its layout changes from one supplier to the next.

At the second station, the carcass is cut into forequarters and hindquarters. At the third, it is boned and cut into ready-to-sell cuts. At both stations, weights have to be recorded.

In practice, weights were largely allocated using theoretical yields rather than systematic actual weighing. Not out of negligence: the scales are not connected to the ERP and cannot be, and typing some thirty lines with gloves on, between two cuts, is not realistic.

The result: time lost retyping at goods-in, and a possible gap between the theoretical weight recorded and the meat that actually left the cutting room. Yet that meat yield is precisely what drives the margins of a food and drink SME.

3 steps where the AI traceability agent works in the cutting room

1. Reading the supplier label instead of copying it

At goods-in, the operator takes a photo of the label with the cutting room's phone. The agent extracts the traceability data (origin, slaughterhouse and cutting plant numbers, use-by date, supplier...) and puts it in the right fields, whatever the supplier's label template.

The operator only has to check what was read. Labels from two different slaughterhouses do not share the same layout: it is up to the agent to adapt, not the operator.

Retyping disappears, and with it the typos in a slaughterhouse number.

2. Dictating cuts and weights during cutting

At the cutting stations, the operator weighs as usual, then dictates into the phone what they have just produced: the muscle or cut, and the weight read on the scale. The agent understands cutting room vocabulary, links each cut to the right lot and prepares the corresponding entry.

The same principle applies to meat that arrives already cut by the supplier: scan the label, then dictate the muscle and the weight.

The recorded weight becomes the actual weight, not a calculated one.

3. Checking before writing to the ERP

The agent writes nothing on its own. Every prepared entry goes through a check by a supervisor before it is finally recorded in Euragro, the client's ERP, published by Cetra Informatique.

If an error is spotted later, it is corrected in Euragro as before. The agent fits into the cutting room's existing way of working, it does not replace it.

People stay in control, the machine does the data entry.

After these three steps, the cutting room also gains accuracy on its stock. Each cut is linked to the raw material lot it comes from, with its actual weight: the stock of cuts in the ERP reflects what is really in the cold room, not a theoretical allocation. Finding the origin of a cut, or every cut from the same carcass, becomes a simple search in the ERP.

Before and after the agent: what changes at each station

"This is going to change our lives," says one of the cutting room operators.

What the agent pays back: the gain by number of lots traced

The gain per lot does not depend on the size of the cutting room: each lot traced with the agent saves about 2 min 45 of data entry, or about €1.05 (based on a fully loaded cost of €23 per hour for a skilled operator). The agent costs €149 excl. VAT per month, whatever the volume. It therefore pays for itself from around 7 lots traced per day, and the gain grows with volume.

Calculated on 21.7 working days per month, Monday to Friday. Our client traces 30 to 36 lots a day. This calculation does not include the accuracy gained on actual weights and stock, which has a direct impact on meat yield.

What the agent doesn't do

It does not replace the ERP's traceability. Euragro remains the reference record for lots, upstream and downstream. The agent prepares the entries and records them there after approval: the ERP carries regulatory traceability, not the agent.

It does not weigh. The recorded weight is the one the operator reads on the scale and dictates. A misreading is caught when the supervisor checks the entry, or corrected afterwards in the ERP as before.

It does not replace the goods-in check. The operator still checks the goods against the delivery note. The agent saves them the retyping, not the check.

Where to start in your cutting room: 3 steps

Map your cutting room station by station. Who receives, who cuts, who enters what, and when. At our client, this description of the three stations was the basis of the specification.

Collect photos of your supplier labels. A few examples per supplier are enough to check that the useful information is readable and to anticipate different formats.

Check access to your ERP. The agent needs to read your item codes, bills of materials and lots, and to write approved entries. Talking to your ERP vendor early saves weeks.

Frequently asked questions about the AI traceability agent

Do we need to replace the cutting room scales?

No. At our client, the scales are not connected and cannot be. The operator reads the displayed weight and dictates it into the phone. The agent therefore works with existing equipment, with no hardware investment in the cutting room. If your scales can be connected, they can complement the set-up, but that is not a prerequisite.

Does the agent work with Euragro and other ERPs?

At our client, the agent writes to Euragro, thanks to an integration set up as part of our partnership with Cetra Informatique. Agrolytics also works with other food industry ERPs. The requirement is to be able to read item codes, bills of materials and lots, and to record approved entries, either through an API or through file exchange.

How long does it take to set up?

At our client, about six weeks passed between signing at the end of August and on-site fine-tuning in mid-October, including the ERP integration, the specification and the development. The timeline depends mainly on the state of your ERP data and on your vendor's availability.

Do operators need to be comfortable with IT?

No. The operator takes a photo and speaks. There is no complex screen and no menu to navigate. Checking and approval are done by a supervisor, who stays in control of what goes into the ERP. A short hands-on session at the station is enough to get started.

How much does an AI traceability agent cost?

The initial development is billed once, at an amount that depends on how hard it is to integrate with your ERP and to fit into your business process. Once this scoping is done, the agent costs €149 excl. VAT per month. The payback calculation above is based on this monthly subscription.

Going further

To understand what an AI agent can take on beyond the cutting room, our guide to AI agents in the food industry covers the main uses, and our article on the integration between Euragro and Agrolytics details the ERP connection.

In 30 minutes, we look together at how your cutting room traces its lots today and how much retyping an agent could remove. Book a call.