Today
Packaging, the website, Instagram and retailer documents carry sentences like “high fiber”, “less fat” or “boosts your immune system”. Quality assurance checks by hand, with the nutrition table and the EU register, whether the recipe supports them and the wording is authorized, often just before printing. A single unsupported claim can stop a listing, trigger a warning letter or cost a whole print run of packaging.
An example
Tarvina is launching a lentil salad in a 250 g pot. The packaging says “high fiber” and “source of protein”, the Instagram draft “boosts your immune system” and the retailer sell sheet “30% less fat”. The agent calculates from the nutrition table: 4.1 g of fiber per 100 g is enough for “source of fiber” (from 3 g) but not for “high fiber” (from 6 g); 15% of energy from protein supports “source of protein” (from 12%). “Boosts your immune system” is not authorized wording, and with 9 mg of vitamin C per 100 g even the authorized immune system claim is not possible, as it needs 12 mg. The fat reduction compared with the previous recipe is only 24%, while such a claim needs at least 30%. The quality manager adopts the corrections before the artwork goes to the printer.
Your data, wherever it lives
Your claims sit today in packaging artwork and retailer documents, often as PDFs, on the website and in social media drafts, the recipes in PLM and ERP, lab values in the QMS and the authorized wordings in a quality assurance spreadsheet. demi reads, understands and links them, with your rights. So the AI checks every claim against recipe and rules in minutes, and quality assurance approves.
Where the data lives today
- PLM
- ERP
- QM system
- Office files
- PDFs
- Web & portals
- Social media
- Photos, video & audio
How demi solves it
Three layers in this order: connect the data first, then the rights, then the AI.
- 1 · ConnectData Layer
The case needs data from PLM, ERP and QMS, plus lab results, packaging artwork, the website and social media drafts. demi replaces nothing and links every claim to the article, the current recipe, nutrition values per 100 g from calculation or lab, and the authorized wording with its conditions from the EU register. So every claim shows the recipe version and value it rests on.
- 2 · PermitPermission Layer
Recipes and lab values are visible only to quality assurance, product development and the responsible person in marketing. Retail customers and agencies see only approved texts. No packaging goes to print and no post goes online before quality assurance has approved the claims, and every approval is recorded with the recipe version.
- 3 · ActIntelligence Layer
The agent finds nutrition and health claims in texts, on packaging and in images and recalculates: is there enough fiber for “high fiber”, is the fat really reduced as much as the text says, does the product contain enough vitamin C for a claim about the immune system? It compares every wording with the authorized claims and proposes a permitted version. The numbers come from recipe and lab, not from the language model.
How it runs
- 1
A packaging text, post, web page or retailer document goes into approval.
- 2
The agent finds every nutrition and health claim, including paraphrases and claims in images.
- 3
It recalculates the conditions from recipe and nutrition values and compares the wording with the authorized claims.
- 4
It marks what is not covered with value and threshold and proposes permitted wording.
- 5
Quality assurance checks and approves; if the recipe changes, the agent checks all texts again.
What changes
Every claim backed by recipe version, value and threshold
Errors are caught before printing, not on the shelf
Faster approvals for packaging, social media and retail
What does it bring you?
Check it with your own numbers
Every starting value is an assumption you can change. The calculation sits right below, step by step.
Your numbers
Result
+60 hfreed per year
Assumptions and calculation
demi checks every claim against recipe, nutrition values and authorized wording; quality assurance approves.
40 min → 28 min per case
How we calculate
25 packaging and promotional texts × 40 min × 30% × 1.0 (manual) × 12 months = 3,600 min = 60 h
Value per year, before demi costs
Freed per year+60 h
Assumption: labor cost of $48.62 per hour in US manufacturing in June 2026 (BLS), plus workplace, IT and overhead.
= Value per year, before demi costs
+$3,900
gained per year
An estimate, not a promise. The starting values are assumptions for an industrial company with 150 to 500 people.
Combine with more cases and get the packageQuestions about this use case
Which claims does demi check?
Nutrition claims such as “low fat”, “no added sugars”, “high fiber” or “source of protein”, health claims such as “contributes to the normal function of the immune system”, and paraphrases of them, also in images, hashtags and retailer documents. The basis is Regulation (EC) No 1924/2006 on nutrition and health claims and the authorized claims in the EU register, which you store as rules together with your quality assurance team.
Where do the nutrition values come from?
From your recipe in PLM or ERP and, where available, from lab results for that recipe version. The agent shows which value is calculated and which is measured. If the two differ widely, it marks that instead of picking one.
What happens when the recipe changes?
demi knows which packaging, web pages, posts and retailer documents rely on which recipe version. If product development changes the oil content, for example, the agent checks all claims again and reports which texts no longer hold before the next print run.
Does demi also write the permitted wording?
Yes. For every marked claim the agent proposes a version that recipe and authorized wording support, such as “source of fiber” instead of “high fiber”. Quality assurance and marketing decide whether and how it is used.
