Tarvina UK Limited · London·, since 1958·, 1,600 people·, 7 locations in 7 countries
Getting started with AI · Food & Beverage
How a food maker gets started with AI
Tarvina doesn’t exist. Everything else on this page will feel familiar.
demi turns your systems into one record, carries over your rights and lets AI agents prepare the routine work. First the data, then the rights, then the AI. A person always approves.
What five use cases free up
+366 h per yearEstimate, not a promise≈ $23,790 per year, calculated cautiously
SN-120 Pretzel & herb dip: The spice supplier reports traces of sesame in the seasoning blend RS-2231, specification V4 of Nov 10, 2026. Six snacks on line 3 are affected. The head of QA assesses the change, then the matrix becomes version 42.
Example
Tarvina’s allergen matrix: the 14 allergen groups under EU FIC, one row per product. Every value leads to its source.
The problem
Wednesday, 7:10 am: one oil changes, and 14 recipes depend on it
The supplier of the sauce base writes to purchasing: from December 1, its base RS-1187 will contain sunflower oil instead of rapeseed oil. Purchasing forwards the email to product development.
That is where the search begins. Which recipes use this base, what changes on which label, and which artwork has to reach the agency by when? The answers sit in three places:
In the PLM14 recipes with base RS-1187, each with its own nutrition calculation.
In the file share9 artworks for tubs and bottles, 5 of them private label. The retailer wants them on Friday.
In the inboxThe new specification as a PDF. What its fatty acids change on every label only shows up when you recalculate.
Two people in labeling spend three days on the calculations. The private labels miss the Friday deadline, and the print slot for tubs and bottles slips by a week.
FromSauce base supplier, L-0412
ToTarvina purchasing
SubjectSauce base RS-1187: sunflower oil instead of rapeseed oil from Dec 1, 2026
Mail #4471Good morning, because of the raw-material situation we are switching base RS-1187 to sunflower oil from December 1. The new specification is attached. All other ingredients stay the same.
Spezifikation RS-1187 V6.pdf4 pages
Sauce base RS-1187 · V6Page 2 of 4
Oil
Sunflower oil, previously rapeseed oil
Fat per 100 g
38 g
of which saturates
4.2 g, previously 2.7 g
Allergens
Egg, mustard
Valid from
Dec 1, 2026
1oil
1sauce base
14recipes
14nutrition tables
9artworks
5private labels by Friday
One ingredient, five rings
In deli foods, everything hangs on the recipe. When a raw material changes, so do the ingredient list, the nutrition values, the allergens and, in the end, the label.
In all 14 recipes, the ingredient list and the saturates change, most of all in the herb dressing SC-500: from 1.8 to 2.9 g per 100 ml.
Same pattern with allergens: one spice, six snacks.
What happens here every month
40specifications
30labels to recalculate
80artworks to check
10trace requests
20HACCP reviews
Together that is 1,520 hours a year, about 0.9 full-time roles. The hours are spread across QA, product development and logistics, for around 420 products, 60% of them private label, each with the retailer’s own specification.
What hurts there
01
Supplier questionnaires from retailers
Allergens, nutrition, origin, packaging, additives: every customer asks differently, and QA fills in every form by hand.
02
Certificates expire and nobody notices
Suppliers’ IFS and BRCGS certificates, declarations of compliance for tubs and films: scattered across portals and folders, each with its own expiry date.
03
Recall: which batch went where?
A spice is recalled. Which lots contain it, which pallets are still in the cold store, which are already at the retailer? Today, the logistics team works its way through ERP postings, weighing records and delivery notes, lot by lot.
What demi does · An example
14 recipes recalculated, and the private labels make the Friday deadline
The email
demi reads the email and the specification: new oil, saturates 4.2 instead of 2.7 g, valid from December 1.
Mail #4471RS-1187 V6
What depends on it
14 recipes use the base. Five of them only go to foodservice in buckets, and their labels are printed on the line. The other nine carry 9 artworks, 5 of them private label, due on Friday.
R-SC-500 V8+13 recipes
Recalculated
Nutrition values and ingredient list under EU FIC, with rounding and allergens in bold. The saturates change in all 14, most of all in SC-500.
SC-500+13 recipes
The drafts
New label data for all 9 artworks, every line with its recipe and specification. The 5 private labels are at the top.
DV-SC500-DE-05+8 artworks
The approval
Labeling and QA check, change one line and approve. On Friday all 5 private labels go to the agency.
The same label before and after the oil change. Two things change: the oil in the ingredient list and the saturates. demi finds both across 14 recipes and 9 artworks. A person approves.
Labeling recalculates around 30 labels a month, rarely 14 at once as with this oil. Of the 90 minutes of manual work per label, the calculation below counts only the roughly 22 that demi takes over on a cautious estimate.
TarvinaDelicatessenExample company
Fine foods from the Allgäu, family-owned since 1958.
Memmingen plant: five production lines for around 420 products, six in ten as private label
Tarvina, a family-owned company of 1,600 people, makes deli salads, dips, sauces and snacks for grocery retail, under its own brand and as private label. The story on this page happens at the Memmingen plant with 390 people: every batch needs its traceability, every retailer its questionnaire.
From salad kitchen to a brand in the chilled aisle
A small deli kitchen behind a butcher’s shop in Memmingen: potato salad, sausage salad and cheese salad made with Allgäu cheese, for butchers, inns and the first self-service shops. The founding family still runs the business today.
The sealed tub arrives in the chilled aisle, with Tarvina on the lid. Nine years later comes the first private-label contract.
The new plant in Memmingen-Nord, close to the A7 motorway: cold store, high-care area, three lines. In 2004, the first IFS certificate.
The sauce line, with cooking, hot holding and rapid chilling. In 2019 the snacks to go follow on line 3, then the sales companies in Vienna, Zurich, Strasbourg and London.
The Memmingen plant runs five lines, two shifts and the cleaning crew at night. This year Tarvina reformulated every barbecue sauce: since July 1, no smoke flavouring goes into any sauce.
Profile
Founded
1958 in Memmingen, family-owned to this day
People
Group 1,600; at the Memmingen plant 390 in eight departments, 220 of them in production. All departments in the org chart
Plant
Memmingen-Nord plant: sauce kitchen with three kettles, five lines, high-care area, cold store with FEFO
Customers
Private label 60%, foodservice 15%, exports to Austria, Switzerland and Poland 12%, own brand and wholesale 13%
Volume
Around 36,000 tonnes a year, €120 million revenue
Certificate
IFS Food Version 8
Per month
Around 40 new specifications, 30 recalculated labels, 80 artworks
Of 1,520 hours, we count 366: volume times time times twelve months, then only the cautious share of that, times 0.6 for partly digital processes. That is about 0.2 full-time roles. Calculated at $65 per hour: $48.62 labor cost in US manufacturing in June 2026 (BLS), plus workplace, IT and overhead. A number doesn’t fit your company? You can change every one of them in the configurator.
Office Floor <=> Production Floor
Where demi works at Tarvina: eight of 14 stations
demi can work at all 14 stations. Pick a use case and see which ones it touches.
Office Floor
O1Market & retail customers
O2Retailer requests & questionnairesnext
O3Specifications & allergens1
O4Quotes & listing
O5Orders & EDInext
O6Recipe & labeling2 3
O7Purchasing & suppliers1
O8Production planningnext
One record. Every fact runs both ways.
Production Floor
S1Goods in & raw-material store4
S2Kitchen & filling4 5
S3Label & checkweigher3
S4Cold store & shipping4
S5Complaints & recallnext
S6At the retailer’s warehouse4
demi works here
Use case touches the station
Next at Tarvina
demi can work here too
Path of lot L26314-1: from the supplier via kitchen and cold store to the retailer’s warehouse
The start
How Tarvina starts: the allergen matrix first, then the labels
Which systems and use cases join in the first 90 days
The first 90 days
Day 1
Week 1
Weeks 2 to 4
Months 2 and 3
Systems
Mailboxes, file share
QA and purchasing inboxes, specifications, read-only
PLM, ERP
Read-only, via a small connector in their own network
QMS, LIMS
Read-only
MES, WMS
Read-only
Use cases
1 Allergen matrix
With approval. Part of daily work from day 30
2 Nutrition
With approval
3 Label check
With approval
4 Traceability
With approval
5 HACCP
With approval
Writing back
Into the PLM
After approval
Why the allergen matrix comes first
The matrix and the specifications behind it carry everything else: nutrition values, labels and every answer to retailers. Reading alone gets demi everything the matrix needs: the specifications in the inbox and the file share from day 1, the PLM from week 1. Each new version of the matrix gets its number on approval by the head of QA. Under that number, the change goes to the retailer portals and onto the labels. Once the PLM is connected in week 1, the same specifications also supply the nutrition values. No use case in the calculation above is bigger: 130 hours a year.
What Tarvina does itself
Head of ITTwo days for the connections
Head of QAOne hour a week
Labeling & specificationsTwo weeks of reviewing drafts, alongside normal work
1Data
Eight systems become one record
1,227entries
1,309links
8systems
An excerpt from Tarvina’s record.
AConnected
ERP
Products, lots with best-before dates, retailer orders, prices and termsHere the lot is just a posting. Where its pallets are, only the WMS knows
PLM
Recipes, raw-material specifications, nutrition calculation, allergen matrixThe matrix is a spreadsheet, kept up by hand, version 41
MES
Production orders, kettle batches, weighing and dosing recordsThe batches are here, the core temperatures on paper
WMS
Cold store, pallets with SSCC, FEFO picking, shippingRemaining shelf life per pallet, but not per retailer requirement
QMS
HACCP plan, CCP sheets, complaints, audits, supplier approvalsThe shifts’ CCP sheets come in as scans
LIMS
In-house lab for pH and sensory, the external lab’s portal for microbiologyEvery external result a PDF, linked to its lot only by the file name
Suppliers, retailers, agenciesThat is where it says what really changed
Nothing gets replaced. Running a different ERP? demi connects to everything, from SAP and Oracle to the specialist software from the next town. At first, demi only reads. A person connects a system after seeing who will read what.
BHeld together
The same lot. Eight systems. Now one entry.
Value
Source
Lot L26314-1, best before Dec 22, 2026
Source: ERP
Barbecue sauce SC-250, 250 g
Source: ERP, product master
Recipe R-SC-250 V7
Source: PLM
Allergen matrix V41: celery, mustard
Source: PLM, supplier specifications from the mailboxes
PA-26-4182, batch A-2207, kettle K2
Source: MES
CCP 2 chilling: 3.6 °C after 95 min
Source: QMS, early-shift scan
P-26-5521: Listeria not detected, pH 4.2
Source: LIMS
Artwork DV-SC250-DE-04
Source: file share
12 pallets, delivery note LS-26-77310
Source: WMS
Distribution center of customer #1659
Source: ERP, EDI order
Target, HACCP plan V12max. 4 °C within 120 minCCP 2 chilling, sauce line
Actual, CCP sheet3.6 °C after 95 minBatch A-2207, early shift Nov 10
CLinked
Lot L26314-1 is barbecue sauce SC-250 to recipe R-SC-250 V7, cooked in kettle K2 and delivered to the distribution center of customer #1659.
Who?
Customers, suppliers, teams
Customer #1659Retail chain, central warehouse South
L-0277Tomato paste supplier
OE-QS-HACCPQA & HACCP
ERP, QMSdelivered to customer #1659
What kind?
Products and recipes
SC-250Barbecue sauce, 250 g
R-SC-250 V7Recipe, valid since Nov 3
RS-1043Tomato paste, specification V5
PLM, ERPis an SC-250 to recipe V7
Where?
Lines and warehouses
Line 1Sauce kitchen, kettle K2
KL-2Cold store, +2 to +4 °C
MES, WMScooked on line 1
What’s going on?
Orders and batches
PA-26-4182Production order
A-2207Batch in kettle K2
LS-26-77310Delivery note, 12 pallets
MES, ERPmade under this order
Which one exactly?
Lot L26314-1
Barbecue sauce, 250 g, best before Dec 22, 2026
14,400tubs from three batches, cooked on Nov 10, 2026
Tarvina packs around 6,000 lots like this a year. In a recall, each one has to be traced within four hours: what is in it and where it went.
ERPposted as a lot
What applies?
Specifications and certificates
KS-1659-SC-250 V3Retailer specification
V41Allergen matrix
IFS Foodvalid until Aug 14, 2027
PLM, file shareapplies to this lot
What’s in writing?
Records and results
CCP 2Chilling, early shift Nov 10
P-26-5521Lab result
DV-SC250-DE-04Artwork on the tub
QMS, LIMS, file sharedocuments this lot
What should it be?
Limits and requirements
HACCP V12Core at least 85 °C, chill within 120 min
KS-1659Remaining shelf life on delivery at least 30 days
QMS, file sharerequirement for this lot
What do we make of it?
Findings
F-26-044Chilling at kettle K2 has taken 9 min longer since October
demifinding on this lot
At the pump maker, the pump with its serial number sits in the middle; at the machine builder, the line at the customer’s site; at the space supplier, the flight unit in orbit. At Tarvina, it is the lot. The grid stays, the middle changes.
DWhat it can do
Recognize
“BBQ 250” on the retailer’s order, SC-250 in the ERP and “new grill sauce” in the agency’s email are the same product.
Unify
Nutrition per 100 g and per 100 ml, temperatures from the logger and from paper: one unit, one rounding.
Trace
Every value shows its source, exact location and version. Unclear rows are kept too.
Look back
The recipe as it stood on Nov 10, V7, not today’s.
Write back
After approval, the new allergen matrix lands in the PLM.
2Rights
The org chart becomes the permission system
What only product development may change in the PLM, only product development may change in demi. And a retailer’s terms are visible only to those who look after that retailer.
Memmingen plant: 390 people, eight departments
One to five: the path of a product through the company, from listing to shipping. Quality and engineering are involved at every step.
Management and staff functions 24Management 2 · Assistants 2 · Finance 8 · HR 5 · IT 5 · Health & safety 2
1Sales30
Key accounts, private label 10
Own brand & export 8
Sales office & EDI 6
Marketing & packaging 6
2Product development9
Recipes & sensory 5
Labeling & specifications 4
3Purchasing9
Raw-material purchasing 5
Supplier management 2
Material planning 2
4Production220
Sauce kitchen 34
Salads 70
Dips 28
Snacks 40
Hygiene & cleaning 22
Shift leads & planning 26
5Logistics48
Goods in 10
Cold store & picking 26
Shipping & routes 12
6Quality22
QA & HACCP 10
Laboratory 6
Complaints & audits 6
7Engineering28
Maintenance 22
Refrigeration & energy 6
24 + 30 + 9 + 9 + 220 + 48 + 22 + 28 = 390 people
Who opens which room
Swipe the matrix sideways →
Who
Recipes & specifications
Allergens & labels
Lots & traceability
HACCP & lab
Prices & terms
HR
Approves
Sales
reads
reads
own customers
–
own customers
–
Reply to the retailer
Product development
edits
edits
reads
reads
–
–
Recipe, ingredient list
Purchasing
own suppliers
reads
reads
–
purchase prices
–
Purchase order
Production
reads
reads
own line
own line
–
–
Shift CCP sheet
Logistics
–
–
edits
cold store
–
–
Shipping
Quality
reads
edits
edits
edits
–
–
Allergen matrix, label, hold
Engineering
–
–
–
own equipment
–
–
Calibration
HR (staff function)
–
–
–
–
–
edits
–
Customer #1659 in the portal
own products
own products
own deliveries
–
own
–
Retailer specification
Allergen agent, AI coworker
reads
draft
–
reads
–
–
Never
Label agent, on someone’s behalf
As much as the person who starts it. No more.
Never
Every room has keys. They come from your systems and never multiply on their own.
A retailer’s terms are visible only to those who look after it. Production sees the lots and CCP sheets of its own line, and nothing else.
An agent carries the keys of the person it works for. An AI coworker has its own, and always a person who is accountable for it. Every access is in the log.
3Intelligence
Five agents do the work, people approve
Allergen agent
AI coworker, accountable: head of QA
TriggerA new supplier specification in the inbox or the file share
ReadsSpecifications, recipes, line allocation, cleaning plans
DoesMaintains the allergen matrix and flags every change. If a specification is missing, it writes “unknown”, never “free from”
ApprovalQA assesses, the head of QA approves
Nutrition agent
Works for the person who starts it
TriggerSomeone from labeling starts it on a changed recipe or specification
DoesNarrows down which lots contain a raw material and where every pallet is, with a mass balance
ApprovalThe head of QA decides on any hold
HACCP agent
AI coworker, accountable: head of QA
TriggerEvery Monday and before every audit
ReadsCCP sheets, temperature loggers, cleaning records, lab results
DoesFinds gaps, such as two missing checks at the chilling CCP or a record without a signature, and drafts actions with deadlines
ApprovalShift lead, QA
Every nutrition table under EU FIC rests on the raw-material values in Tarvina’s PLM, ingredient by ingredient.
Calculations run on Tarvina’s own rules and data, not in the language model. Anything that goes out or is written back to a system is approved by a person.
The roadmap
Next up: questionnaires, complaints and remaining shelf life
Questionnaires and retailer portals
Every retailer asks about allergens, origin and packaging in its own way. demi pre-fills the questionnaire and the specification portal from the allergen matrix and the recipe, every answer with its source. QA approves.
O2
Complaints per lot
A complaint arrives with a photo of the lid. demi reads the lot and best-before date, finds the batch, CCP sheet, lab result and retained sample, and drafts the reply to the retailer.
S5
Planning with remaining shelf life
Every retailer requires a minimum remaining shelf life on delivery. demi plans production and shipping against these requirements, from the EDI orders and the stock in the cold store.
O5O8
Sources
Why data first, then rights, then AI
60%
Through 2026, Gartner expects organizations to abandon 60% of AI projects that lack AI-ready data.
Tarvina Feinkost GmbH & Co. KG is made up: a typical family-owned hidden champion in Food & Beverage, not a real one. An estimate from our assumptions, not a promise. You can change every number in the configurator.