The camera tracks slowly along a production line in a factory at night: a machine with gleaming rollers, a conveyor carrying parts, grey electrical cabinets and yellow floor markings, under sodium and fluorescent lamps.

Industrial AI

Your plant's biggest cost isn't on any invoice. Industrial AI for Greek manufacturers, from an office inside the Almyros industrial area

Book a Plant X-Ray Play: 8 defects in 60 seconds

From €4,900 · credited in full against a pilot

Scroll. The scan line shows what a vision model sees on the same line.

AI-generated footage of a production line at night. As you scroll, a diagonal scan line opens the machine-vision view: boxes around every part, one defect with a confidence score, a parts counter.

Engineers who sit your site safety induction before they set foot on the line. And a second office in Heraklion for plants in Crete.

It's in the hours the shift stands waiting, the parts that come back, the power burned at peak, the paper written up twice. Nobody sees all of it. We measure it.

Eurostat 2025 · firms with 10+ staff

Europe has already started.

Share of firms with 10+ staff using at least one AI technology, in Greece and across the EU. Pick a sector.

Greece 7.77% · EU 17.27%

European manufacturers use AI more than twice as often.

Greece 3.84% · EU 12.47% (2025)

Mind the year: in 2024 Greece was ahead (9.75% vs 6.79%). A swing that big in one year needs the 2026 figures to confirm it.

Greece 4.72% · EU 14.33%

Europe is three times ahead.

Basic and fabricated metals (NACE C24–C25).

Greece 12.67% · EU 16.77%

The narrowest gap on the list. But the category is broad: alongside plastics it counts cement, glass and ceramics.

In the statistics: rubber and plastics, plus non-metallic mineral products (NACE C22–C23). It isn't a plastics-only figure.

Greece 1.88% · EU 4.14%

Fewer than 2 in 100 Greek firms use AI on the production process itself.

Large 24.27% → 37.58% · Small 8.20% → 7.33% (2024 → 2025)

The large firms sped up. The small ones took a step back. Small: 10–49 staff. Large: 250 and up.

The big players have started. The SMEs are standing still. That's your opening.

In your sector, whoever moves first won't find a queue ahead of them.

Source: Eurostat, tables isoc_eb_ain2 and isoc_eb_ai, 2025 data; firms with 10+ staff, financial sector excluded. isoc_eb_ain2 (opens in a new tab) isoc_eb_ai (opens in a new tab)

Shift log

Twelve notes from plants like yours.

None of them shows up on the balance sheet. All of them cost you every day. Tap one to see what's behind it.

When Nikos is on leave, line 2 stops.

When Nikos is on leave, line 2 stops.

Production

What you don't see

Time to repair (MTTR) · calls to one person

What AI does

A voice maintenance copilot over your manuals and past work orders

The operator copilot Pin this problem to the X-Ray form
We type every invoice in by hand. Every one.

Accounts

What you don't see

Minutes per document · exceptions

What AI does

Document AI straight into the ERP, and on to myDATA through your provider

Automations that give hours back Pin this problem to the X-Ray form
Our quote takes three days. The customer ordered elsewhere in three hours.

Sales

What you don't see

Time to quote · requests lost

What AI does

A quote copilot: the RFQ and the drawing become a draft quote

The quote copilot Pin this problem to the X-Ray form
Our customer finds the defect, not us.

Quality

What you don't see

Defects reaching the customer (ppm) · credit notes

What AI does

Machine-vision inspection on the line

Vision quality control Pin this problem to the X-Ray form
The machine always breaks on Friday night.

Maintenance

What you don't see

Unplanned downtime hours

What AI does

Predictive maintenance

Predictive maintenance Pin this problem to the X-Ray form
We rebuild the production plan in Excel every morning.

Planning

What you don't see

Planner hours · mould or line changeovers

What AI does

Production schedule optimisation

Production planning Pin this problem to the X-Ray form
We always have too much or too little. Never the right amount.

Supply chain

What you don't see

Waste % · stock-outs

What AI does

Demand forecasting

Data and decisions Pin this problem to the X-Ray form
Orders arrive by email, Viber and phone, and someone retypes them.

Customer service

What you don't see

Order-entry hours · wrong lines

What AI does

A digital assistant that gathers orders from every channel and posts them to the ERP

Connecting to your ERP Pin this problem to the X-Ray form
Electricity is our second-biggest cost and nobody knows where it goes.

Operations

What you don't see

kWh per unit · peak kW

What AI does

Energy optimisation

Plant energy Pin this problem to the X-Ray form
Every May we train 40 seasonal staff from zero.

People

What you don't see

Time to competence

What AI does

A multilingual staff assistant over your own procedures

Food and beverage Pin this problem to the X-Ray form
Every audit (ISO, HACCP, customer) costs us a week of paperwork.

Quality

What you don't see

Audit-prep hours · findings

What AI does

A compliance document assistant

Automations that give hours back Pin this problem to the X-Ray form
The phone rings. At 15:01 nobody answers.

Front office

What you don't see

Missed calls · lost orders

What AI does

A voice assistant that passes the call to a person when it needs to

AI Front Desk Pin this problem to the X-Ray form
We close the month on the 12th of the next one.

Management

What you don't see

Days to close the month

What AI does

Reports that build themselves

Data and decisions Pin this problem to the X-Ray form
Suppliers raise their prices and we notice months later.

Purchasing

What you don't see

Purchase-price drift

What AI does

Purchasing analytics that flags every rise the day it lands

Data and decisions Pin this problem to the X-Ray form
Trucks queue at the gate in the morning. By afternoon the docks are empty.

Logistics

What you don't see

Truck turnaround minutes · gate waiting

What AI does

Dock scheduling with arrival forecasts

Logistics and distribution Pin this problem to the X-Ray form
The mixer truck waits for the contractor. The contractor waits for the mixer truck.

Building materials

What you don't see

Waiting minutes per load

What AI does

Dispatch planning that changes with the day

Plant X-Ray Pin this problem to the X-Ray form
A buyer in Munich asked for a spec sheet. We sent it three days later.

Export

What you don't see

Time to answer buyers · shipments held up

What AI does

Export paperwork and buyer replies, in three languages

Food and beverage Pin this problem to the X-Ray form

The notes promise no numbers. Yours come out of your own X-Ray; here we only show what we measure.

Six senses

AI on the plant floor isn't a chat window.

It reads, sees, hears, predicts, optimises and decides. Here's where, on the plant.

Isometric drawing of a plant: a production line with a camera, a compressor with a sensor, a dispatch office with paperwork, a sales desk with a phone, a schedule board and a reject gate. Six spots light up, one per sense.

Hover over, or tap, a spot on the plant.

Reads

The dispatch office

Delivery notes, 3.1 mill certificates, CMRs: read and posted to the ERP.

Use case Automation that gives you hours back
Sees

The camera over the line

Looks at every part, not one sample an hour.

Use case Vision quality control
Hears

The sales phone and the technician's mobile

Takes the order at 15:01 and writes the fault report from a voice note.

Use case AI Front Desk
Predicts

The sensor on the compressor motor

Spots the vibration anomaly before it becomes a breakdown.

Use case Predictive maintenance
Optimises

The energy meter and the schedule board

Moves flexible loads out of the peak and sequences orders with fewer changeovers.

Use case Plant energy optimisation
Decides

The reject gate at the end of the line

Pulls the faulty part off the line, within limits you set. Anything uncertain is judged by a person.

Use case Vision quality control

Process X-Ray

This is how an order travels through your company.

Pick your trade, mark your guess, then run the X-ray. See where the flow runs hot, and what changes once AI goes in.

Plastics: from RFQ to delivery

An injection-moulding plant, 60 people, 120 requests for quotation a month.

This is how a quote request travels through your plant.

Steps, minutes, volume and what changes with AI · use case
Step	Who	Minutes	Volume/month	With AI	Minutes on the new path
RFQ arrives by email: PDF drawing and spec sheet	Sales	4 min	120	Reads Pulls material, quantity, tolerances and dimensions from the email and the drawing	0.5 min
Hunting for a similar past part in old quotes and folders	Sales	15 min	120	Reads Finds the closest past parts and what they were quoted at	1 min
Part weight and cycle time estimated in the engineer's spreadsheet	Production engineer	25 min	120	Predicts Estimates weight and cycle from the plant's own history; the engineer checks it	5 min
Phone call to the supplier for today's resin price	Purchasing	10 min	120	Reads Takes the price from recent invoices and the supplier's list	0.5 min
Quote written up in Word	Sales	20 min	120	Acts Drafts the quote with every figure filled in	3 min
Waiting for the owner to sign off	Management	10 min	120	Stays human A person still approves the price, with everything on one screen	3 min
The customer orders; the order is retyped into the ERP	Sales	12 min	45	Acts Opens the order in the ERP straight from the accepted quote	1 min
Resin stock checked in the warehouse spreadsheet	Warehouse	8 min	45	Acts Checks stock and reserves the quantity	0.2 min
Call to production: when can we fit it in?	Sales	10 min	45	Predicts Proposes a slot and a delivery date; the planner confirms	2 min
Email to the customer with the delivery date	Sales	6 min	45	Acts Sends the confirmation the moment the date is locked	0.5 min
Sample inspection at the end of the batch	Quality	30 min	45	Sees A camera checks every part on the line; the inspector sees only the doubtful ones	10 min
Delivery note and invoice keyed in again	Accounts	10 min	45	Acts Issues the documents from the order and transmits them through your e-invoicing provider	1 min
Calls: where's my order?	Sales	5 min	90	Acts Answers with the order status; anything it doesn't know goes to a person	0.5 min

13 steps → 5. 233 hours → 135–174 hours a month.
≈ €50,800 → ≈ €29,500–38,000 a year in labour, at €18.20/hour (Eurostat 2025, manufacturing).
≈ 0.4–0.6 people back on real work (hours freed ÷ 160 hours a month)
Time to quote, for requests on the new path: 3 days → same day
The hours are redeployed, not cut.

Nobody: what you can't see

Not slow work. Work that isn't happening at all.

Peak-power charges on the electricity bill: nobody links them to press start-ups.
Press 3 runs 6 °C hotter than in May. Nobody is watching it.
Lost quotes: nobody asks why the customer ordered elsewhere.

Hours = Σ (minutes × monthly volume) ÷ 60 · With AI = hours today − coverage × (hours today − hours on the new path), coverage 30–50% · € = hours × 12 × €18.20/hour (Eurostat 2025, manufacturing) · People = hours freed ÷ 160

We only count labour hours here, and conservatively: we assume that in year one only 3 to 5 cases in 10 take the new path, while the rest run as they do today. Lost output, waste and orders that went elsewhere aren't in the sum. That's what the meter further down is for.

Metals: breakdown → repair → certificate

A steel processor supplying larger manufacturers: a cut-to-length line, 45 people, 160 certificates a month.

This is how a breakdown travels through your plant, all the way to the paper that ships with the load.

Steps, minutes, volume and what changes with AI · use case
Step	Who	Minutes	Volume/month	With AI	Minutes on the new path
The line stops without warning; the operator goes looking for the shift lead	Operator	10 min	8	Predicts Motor vibration and current show the anomaly days ahead; the stop goes into the schedule	3 min
The shift waits: 4 people for 90 minutes	Shift crew	360 min	8	Predicts The fault shows up early and becomes a planned stop: 4 people for 40 minutes instead of 90. At this scenario's coverage that happens for 2 to 4 of the 8 breakdowns; the rest still come unannounced	160 min
A call to Nikos, the only one who knows this fault	Maintenance tech	15 min	8	Reads The maintenance copilot answers by voice, in Greek, from the manuals and past work orders	3 min
Digging out the manual and the spare-part drawing	Maintenance tech	25 min	8	Reads Shows the right page and the part number	2 min
Is the spare in stock? Shelf, then spreadsheet	Stores	15 min	8	Acts Shows stock and shelf location; if it's missing, raises a purchase request	1 min
Ordering the spare by phone and email	Purchasing	30 min	3	Acts Sends a ready request for quotation to three suppliers; the buyer decides	5 min
The repair itself	Maintenance tech	120 min	8	Stays human The repair stays a technician's job. It simply happens, more often, before something breaks	120 min
Handwritten fault report, typed up later	Tech and office	15 min	8	Hears The tech says it into a phone; the work order writes itself	2 min
Mill certificates (3.1) arrive as PDFs from the steelworks	Quality	6 min	160	Reads Extracts heat number, grade, dimensions and chemistry	0.5 min
Cross-checking against the customer's order spec	Quality	10 min	160	Reads Checks every value against the spec and flags only the mismatches	1 min
Writing the certificate for the customer	Quality	12 min	160	Acts Drafts the certificate from the checked data	1 min
Quality manager signs	Quality manager	3 min	160	Stays human A person always signs, with any mismatches already flagged	2 min

12 steps → 5. 159 hours → 105–126 hours a month.
≈ €34,700 → ≈ €22,900–27,600 a year in labour, at €18.20/hour (Eurostat 2025, manufacturing).
≈ 0.2–0.3 people back on real work (hours freed ÷ 160 hours a month)
Line downtime per month: 12 h → 8.7–10 h The cost of lost output isn't counted here. You add it in the meter, with your own € per downtime hour.
The hours are redeployed, not cut.

Nobody: what you can't see

Not slow work. Work that isn't happening at all.

The decoiler motor draws a little more current every week. Nobody looks until it stops.
The same spare part is rush-ordered again and again: nobody counts how often a year.
Customers asking for the certificate a second time: nobody records how many, or why.

Hours = Σ (minutes × monthly volume) ÷ 60 · With AI = hours today − coverage × (hours today − hours on the new path), coverage 30–50% · € = hours × 12 × €18.20/hour (Eurostat 2025, manufacturing) · People = hours freed ÷ 160

We only count labour hours here, and conservatively: we assume that in year one only 3 to 5 cases in 10 take the new path, while the rest run as they do today. Lost output, waste and orders that went elsewhere aren't in the sum. That's what the meter further down is for.

Food: forecast → production → waste

A dairy, 70 people, fresh products for supermarkets and hotels.

This is how a week of demand travels through your plant, all the way to the waste bin.

Steps, minutes, volume and what changes with AI · use case
Step	Who	Minutes	Volume/month	With AI	Minutes on the new path
Milk intake from farmers: weighbridge slip filled in by hand	Intake	5 min	400	Reads Weight and slip go straight to the farmer's account	0.5 min
Supermarket orders by email, portals and phone	Sales	5 min	300	Reads Pulls every order into one list, by SKU	0.5 min
Next week's demand, guessed in a spreadsheet	Sales manager	180 min	4	Predicts Forecasts each SKU from weather, holidays, retailer promotions and the tourist season; the manager signs it off	30 min
Production plan rebuilt in Excel every morning	Planner	60 min	22	Predicts Proposes a plan with the fewest changeovers; the planner adjusts it	15 min
Rush order: the plan is redone mid-shift	Planner and supervisor	40 min	10	Predicts Re-sequences in seconds and shows what moves	10 min
Milk and packaging: calls to stores and suppliers	Purchasing	30 min	22	Acts Matches needs against stock and deliveries; alerts only on shortfalls	5 min
HACCP logs on paper: temperatures, pasteurisation	Operator	15 min	44	Reads Readings come from the sensors; the operator confirms and signs	3 min
Spot checks of labels and use-by dates	Quality	20 min	44	Sees A camera at the end of the line reads every date and lot code	3 min
End-of-day waste weighed and written on paper	Warehouse	20 min	22	Acts The scale logs waste by SKU straight into the report	3 min
Expired returns from retailers: counting and credit notes	Accounts	15 min	40	Reads Reads the return notes and prepares credit notes for approval	2 min
Monday report for the owner, stitched from four files	Accounts	120 min	4	Acts Writes itself every Monday, with forecast error by SKU	10 min

11 steps → 5. 161 hours → 93–120 hours a month.
≈ €35,200 → ≈ €20,200–26,200 a year in labour, at €18.20/hour (Eurostat 2025, manufacturing).
≈ 0.3–0.4 people back on real work (hours freed ÷ 160 hours a month)
Waste: measured daily by SKU, not once a month
The hours are redeployed, not cut.

Nobody: what you can't see

Not slow work. Work that isn't happening at all.

Waste climbs every Monday after a long weekend. Nobody has linked it to Friday's order.
SKUs that always come back from the same store: nobody asks why.
The cold-room temperature is logged on paper. Nobody sees the night-time peaks.

Hours = Σ (minutes × monthly volume) ÷ 60 · With AI = hours today − coverage × (hours today − hours on the new path), coverage 30–50% · € = hours × 12 × €18.20/hour (Eurostat 2025, manufacturing) · People = hours freed ÷ 160

We only count labour hours here, and conservatively: we assume that in year one only 3 to 5 cases in 10 take the new path, while the rest run as they do today. Lost output, waste and orders that went elsewhere aren't in the sum. That's what the meter further down is for.

Wholesale: Viber order → invoice

A food distributor in Crete, 35 people, 900 orders a month from shops and tavernas.

This is how an order travels, from a Viber voice note to money in the bank.

Steps, minutes, volume and what changes with AI · use case
Step	Who	Minutes	Volume/month	With AI	Minutes on the new path
The order lands on Viber: voice note, photo of a list, or text	Sales desk	3 min	900	Reads Turns voice, photos and text into order lines	0.5 min
"The usual cheese": which product code is that?	Sales desk	4 min	900	Reads Matches descriptions to codes from the customer's history; asks only when unsure	1 min
Retyped into the ERP	Sales desk	6 min	900	Acts Creates the order in the ERP	0.5 min
Balance and credit-limit check	Accounts	3 min	900	Acts Checks the limit and holds only the orders that exceed it	0.3 min
Call to the warehouse: do we have it?	Sales desk	3 min	300	Acts Shows availability and suggests a substitute	0.3 min
Confirmation back to the customer on Viber	Sales desk	2 min	900	Acts Sends the lines and the delivery window	0.2 min
Calls: when is my order coming?	Sales desk	3 min	400	Acts Replies on Viber with the delivery status; a person takes the exceptions	0.3 min
Delivery note and invoice issued, lines checked	Accounts	3 min	900	Acts Issues the documents from what was actually loaded and transmits them through your provider	0.3 min
"You brought me two short": credit note	Accounts	15 min	60	Reads Reads the message and the signed proof of delivery; prepares a credit note for approval	3 min
Matching payments: bank transfers and drivers' cash	Accounts	4 min	700	Reads Matches payments to invoices; the accountant approves the suggestions	0.5 min
Chasing overdue invoices by phone	Accounts	5 min	120	Acts Sends a polite reminder with the balance; the difficult conversations stay human	1 min
Monthly sales report by customer	Management	240 min	1	Acts Builds itself on the 1st of every month	15 min

12 steps → 5. 426 hours → 241–315 hours a month.
≈ €81,700 → ≈ €46,300–60,500 a year in labour, at €16.00/hour (Eurostat 2025, trade).
≈ 0.7–1.2 people back on real work (hours freed ÷ 160 hours a month)
The hours are redeployed, not cut.

Nobody: what you can't see

Not slow work. Work that isn't happening at all.

Customers who ordered every week and then stopped: nobody notices for months.
Lines that run out every Friday by 11:00: nobody counts the orders they cost.
Discounts agreed on the phone and written down nowhere: nobody checks them against the invoice.

Hours = Σ (minutes × monthly volume) ÷ 60 · With AI = hours today − coverage × (hours today − hours on the new path), coverage 30–50% · € = hours × 12 × €16.00/hour (Eurostat 2025, trade) · People = hours freed ÷ 160

We only count labour hours here, and conservatively: we assume that in year one only 3 to 5 cases in 10 take the new path, while the rest run as they do today. Lost output, waste and orders that went elsewhere aren't in the sum. That's what the meter further down is for.

Logistics: truck arrival → POD → billing

A third-party warehouse near the port of Volos, 50 people, 600 trucks a month.

This is how a truck travels through your warehouse, all the way to the customer's bill.

Steps, minutes, volume and what changes with AI · use case
Step	Who	Minutes	Volume/month	With AI	Minutes on the new path
Dock slot booked by phone and email	Traffic office	5 min	600	Predicts Offers a dock slot based on expected arrivals and workload	0.5 min
Gate: the guard writes plate and time in a logbook	Gate	3 min	600	Sees A camera reads the plate and ties it to the booking	0.3 min
Receiving: damage photographed when someone remembers	Warehouse	3 min	600	Sees Every pallet photographed on arrival, possible damage flagged	0.5 min
CMRs and delivery notes typed into the WMS	Office	8 min	600	Reads Reads the CMR and delivery note and opens the receipt in the WMS	1 min
Quantity mismatch: email to the customer with whatever evidence turns up	Office	20 min	40	Acts Drafts a discrepancy report with photos and documents	4 min
Shift rota set by gut feel	Warehouse lead	60 min	22	Predicts Forecasts tomorrow's volume and suggests headcount per shift; the lead decides	10 min
Calls: where's my shipment?	Customer service	4 min	800	Acts Answers on Viber and email from the WMS; a person takes the exceptions	0.5 min
POD: the driver sends a photo, the office files it	Office	3 min	1,200	Reads Reads signature, date and remarks, and closes the delivery	0.3 min
POD remarks ("short") discovered at month-end	Office	25 min	30	Reads Flags the remark the moment the photo arrives	5 min
Customer billing: WMS movements pasted into Excel	Accounts	90 min	25	Acts Calculates billing from movements and contract rates; the accountant approves	10 min
Billing dispute: digging for proof	Accounts	40 min	8	Reads Finds PODs, photos and timestamps in seconds	5 min

11 steps → 5. 394 hours → 222–291 hours a month.
≈ €133,800 → ≈ €75,300–98,700 a year in labour, at €28.30/hour (Eurostat 2025, transport and storage).
≈ 0.6–1.1 people back on real work (hours freed ÷ 160 hours a month)
The hours are redeployed, not cut.

Nobody: what you can't see

Not slow work. Work that isn't happening at all.

Trucks waiting over an hour at the dock: nobody bills the waiting time.
Damage visible in the receiving photos: nobody matches it to the claims that arrive months later.
One customer disputes the same charge every month: nobody has fixed the clause in the contract.

Hours = Σ (minutes × monthly volume) ÷ 60 · With AI = hours today − coverage × (hours today − hours on the new path), coverage 30–50% · € = hours × 12 × €28.30/hour (Eurostat 2025, transport and storage) · People = hours freed ÷ 160

We only count labour hours here, and conservatively: we assume that in year one only 3 to 5 cases in 10 take the new path, while the rest run as they do today. Lost output, waste and orders that went elsewhere aren't in the sum. That's what the meter further down is for.

Olive oil: from olive intake to export

An olive mill and bottler in Crete, 25 people, 300 growers, exporting to Germany and the US. Peak-month volumes (November to January).

This is how a load of olives travels, from the weighbridge to the container.

Steps, minutes, volume and what changes with AI · use case
Step	Who	Minutes	Volume/month	With AI	Minutes on the new path
Olive intake: weighbridge slip written by hand	Intake	3 min	900	Reads Weight and grower details go straight into the system	0.3 min
Acidity and lab results typed into Excel, grower by grower	Lab	4 min	900	Reads Reads the analyser output and links it to the load	0.5 min
Grower settlements: kilos, yield, quality, price	Accounts	10 min	300	Acts Produces each grower's settlement from the same data; the accountant approves	1 min
Growers ringing: how much oil did I get?	Office	4 min	600	Acts Replies on Viber with yield and acidity; complaints go to a person	0.3 min
A buyer abroad asks for price, sample and spec	Export desk	30 min	40	Reads Drafts a reply in English or German from the spec database; the export manager approves	6 min
A product spec sheet for every buyer	Export desk	45 min	10	Acts Builds the sheet from that batch's lab analyses	8 min
Quote including freight: calls to forwarders	Export desk	40 min	10	Acts Requests freight rates in parallel and drafts the quote	10 min
Shipping documents: origin, analyses, packing list, invoice	Export desk	120 min	8	Acts Fills in the documents from the order and the batch	20 min
Checking against the buyer's requirements; one error holds the load at customs	Export desk	40 min	8	Reads Checks every document against the buyer's list and flags mismatches	5 min
Keeping the buyer posted on the shipment	Export desk	10 min	16	Acts Sends an update at every status change	1 min
Traceability for a buyer audit: which growers, which tank	Quality	180 min	2	Reads Shows the batch's path from weighbridge slips to bottle	15 min

11 steps → 5. 259 hours → 145–191 hours a month.
≈ €56,600 → ≈ €31,700–41,700 a year in labour, at €18.20/hour (Eurostat 2025, manufacturing).
≈ 0.4–0.7 people back on real work (hours freed ÷ 160 hours a month)
Reply to a buyer, on the new path: 2–3 days → same day
The hours are redeployed, not cut.

Nobody: what you can't see

Not slow work. Work that isn't happening at all.

Growers who brought fewer olives this year: nobody rang to ask why.
Buyer enquiries from Germany left unanswered over the weekend: nobody counts how many are lost.
Every batch's acidity is on record, but nobody links it to the grove and the harvest day.

Hours = Σ (minutes × monthly volume) ÷ 60 · With AI = hours today − coverage × (hours today − hours on the new path), coverage 30–50% · € = hours × 12 × €18.20/hour (Eurostat 2025, manufacturing) · People = hours freed ÷ 160

We only count labour hours here, and conservatively: we assume that in year one only 3 to 5 cases in 10 take the new path, while the rest run as they do today. Lost output, waste and orders that went elsewhere aren't in the sum. That's what the meter further down is for.

Ready-mix: order → dispatch → delivery ticket

A ready-mix concrete plant in Heraklion, 14 mixer trucks, 1,100 loads a month.

This is how a concrete order travels, from the contractor's phone call to the invoice.

Steps, minutes, volume and what changes with AI · use case
Step	Who	Minutes	Volume/month	With AI	Minutes on the new path
The contractor rings: "tomorrow 7:00, 24 cubic metres of C25/30"	Dispatch	6 min	500	Hears The voice assistant takes the order and reads back quantity, grade and time; anything unclear goes to a person	1 min
Tomorrow's schedule on a whiteboard, in marker	Dispatch	45 min	22	Predicts Plans runs so trucks wait neither at the plant nor on site; the dispatcher has the final say	10 min
Same-day changes: rain, a site that isn't ready	Dispatch	15 min	120	Predicts Re-plans the runs and tells the drivers	3 min
Drivers calling: where to, which gate?	Drivers	3 min	1,100	Acts Sends route, pour point and site contact to the driver's phone	0.3 min
The truck waits on site	Driver	15 min	1,100	Predicts Departures are timed to the site being ready; in this scenario waiting drops, it doesn't vanish	10 min
Handwritten ticket, signed on site	Driver	3 min	1,100	Acts Digital ticket from the batch weights, signed on the phone	1 min
Ticket photos typed into the ERP	Accounts	3 min	1,100	Acts Ticket data arrives in the ERP already structured	0.3 min
The contractor disputes the quantity	Accounts	30 min	15	Reads Pulls up batch weight, departure time, route and signature	5 min
Strength-test results typed into Excel and sent to the customer	Lab	10 min	120	Reads Logs results from the compression tester and sends the report	2 min
Monthly invoicing per site	Accounts	30 min	80	Acts Consolidated invoice from the month's tickets; the accountant approves	3 min

10 steps → 5. 604 hours → 422–495 hours a month.
≈ €131,900 → ≈ €92,200–108,100 a year in labour, at €18.20/hour (Eurostat 2025, manufacturing).
≈ 0.7–1.1 people back on real work (hours freed ÷ 160 hours a month)
The hours are redeployed, not cut.

Nobody: what you can't see

Not slow work. Work that isn't happening at all.

Loads that come back with concrete left over: nobody measures how many cubic metres are thrown away each month.
The same contractor cancels at the last minute every rainy Friday: nobody has built it into the schedule.
Sites that always pay late: nobody sees it before the next load leaves.

Hours = Σ (minutes × monthly volume) ÷ 60 · With AI = hours today − coverage × (hours today − hours on the new path), coverage 30–50% · € = hours × 12 × €18.20/hour (Eurostat 2025, manufacturing) · People = hours freed ÷ 160

We only count labour hours here, and conservatively: we assume that in year one only 3 to 5 cases in 10 take the new path, while the rest run as they do today. Lost output, waste and orders that went elsewhere aren't in the sum. That's what the meter further down is for.

X-ray your own flow45 min · free · no slides

Game · 60 seconds

Spot the defect.

An olive-oil bottling line, eight defects. You have 60 seconds. Then you see what a machine found in the same image.

Before you start: the image was made with AI and hides 8 defects.

Your line runs 24/7. Your inspectors blink.

Send us 20 photos of defects from your line

We look at them and tell you whether a camera can catch them. It's the first step of a Plant X-Ray.

AI-generated image: an olive-oil bottling line with glass bottles on a stainless-steel conveyor, with a 60-second timer in the corner.

Tap every defect you can see. Zoom in for the small ones.

Missing cap
Upside-down label
Torn label
Label applied too low
Underfilled bottle
Foreign particle
Wrong cap
Tipped-over bottle

When the clock stops, you see what the vision model flagged on this line, false alarms included.

The demo didn't load. The description below tells you what you'd have seen.

Keyboard: arrows to move, Enter to mark, + and − to zoom.

Text version: the list of the 8 defects

All eight, one by one

Missing cap The neck is bare, threads showing: the bottle leaves unsealed. Missing cap (open neck, threads visible) · 0.94
Upside-down label Text at the bottom, olive branch at the top: the label went on upside down. Label upside down · 0.90
Torn label Its left part is missing and the lower edge hangs loose. Torn / peeling label · 0.92
Label applied too low It sits almost at the base, far lower than on the bottles beside it. Label applied too low · 0.72
Underfilled bottle The oil sits well below the shoulder the others are filled to; same bottle as the low label. Low fill level (visible meniscus below shoulder) · 0.55
Foreign particle A dark fragment is floating in the oil. Foreign particle in the oil · 0.68
Wrong cap A red cap in a batch of green ones: the wrong closure reached the capper. Wrong cap colour (red instead of green) · 0.91
Tipped-over bottle It has fallen on its side on the conveyor and can jam the line. Bottle tipped over on conveyor · 0.95

What the machine found

The machine: 8/8 · missed: 0 · false alarms: 3.

It missed none of the 8.

It also saw 3 defects that aren't there: Label missing ×2 · Dark spot on glass, possible chip.

The machine found all eight, but it also saw three that aren't there. See where it went wrong.

The boxes are what the vision model flagged on this line.
Confidence is the model's own estimate for each finding. It ranks the findings; it isn't a probability.
A box counts as a find when it overlaps the defect by at least 0.3 (IoU).
On your own line the defects may be smaller and the frame moves. That's why the system is trained during the pilot on defect photos from your line.

Open the decode

Decode

Play → Decode → Deploy

Mechanism: A hidden-object game against a machine.
Psychology: A competence challenge, then a comparison. You feel the gap instead of reading about it.
Platform signal: Engineers share scores and challenge colleagues. On LinkedIn it reaches a pool of 3.1 million members in Greece (DataReportal 2026).
Honest range: 2–5% of players from the industry ask for a feasibility check.Typical range for a pilot in a Greek SME · Netex estimate
Lineage: Who did it before us, and what we kept.Frontify ran a B2B hidden-object game: over 40,000 plays in the first month and +96% leads in two weeks (self-reported, via Econsultancy). Econsultancy (opens in a new tab)

Want a game like this for your own brand?

What we build

Four steps, in order of risk.

We start where the risk is lowest. We earn trust with numbers, then go deeper.

Pilot lengths are typical ranges. The exact length goes into the pilot charter.

On your line

Who touches your line, and how.

Systems: ERP, SCADA, historian, PLC. We start read-only. We write back only where you sign off on it.
Protocols: OPC UA, Modbus TCP, MQTT; CSV exports where there's nothing else.
Cameras and light: Vision inspection stands or falls on lighting. We test it in the camera trial, before we propose any hardware.
Edge or cloud: Anything that makes decisions on the line runs locally, beside the line. Only reports and training data go to the cloud, with your agreement.
Safety: A safety induction before every visit. No work on a machine that hasn't been isolated and locked out.
Cybersecurity: We connect to the production network only through the route your network owner sets, and every access is logged. If your plant falls under NIS2 (Greek Law 5160/2024), we factor that in from the X-Ray onwards.

Bundle

Smart Plant Starter

Four pieces that talk to each other from day one.

Diagram: four blocks (energy, vision, plant dashboard, alerts) connected to one shared data spine. The hardware block is drawn in outline because it comes from partners.

Energy monitoring and optimisation
Vision quality control on one line
A plant dashboard: energy, quality and stops on one screen
Alerts on the right person's phone, not in an inbox nobody opens

Hardware (cameras, meters, sensors) comes from partners where it's needed. We take on the data, models, dashboards and alerts. The data stays yours.

Priced after the Plant X-Ray, once we know the line.

Start with the Plant X-Ray

Hidden Cost Meter

Put your own numbers in.

The Production tab is already open. Four sliders, one result, the formula underneath.

Hidden Cost MeterWork sheet

Department Production & maintenance

Minutes of manual logging per shift: 45 min
Shifts per month: 44
Unplanned downtime hours per month: 6 h
People idle during each stop: 4
€ per downtime hour (your figure): €1,500
Hourly labour cost: €18.20/hour

Eurostat lc_lci_lev, 2025 · official statistics

Hidden cost per year€120,449

Hidden hours per month: 57
What could be automated: €13,290conservative€37,380ambitious
People back on real work: 0.08–0.15 full-time equivalentsWe don't count jobs cut. We count time given back.
Payback against a pilot: 5–6 months Against a typical pilot of €6,000 to €18,000.

What we'd build

Predictive maintenance on 3–5 critical assets
Sensor logging instead of paper
A voice maintenance copilot

Funding. In the X-Ray we check which programme fits, if any does, against the current call text.

The formula

Hidden cost/year = [(logging minutes × shifts ÷ 60) + (downtime hours × idle people)] × 12 × €/hour + (downtime hours × 12 × € per downtime hour)

Cost/year = hidden hours/month × 12 × hourly cost (+ external cost, where there is one). Automatable = labour cost × department coefficient (0.20–0.40) + external cost × a second coefficient. People = hidden hours/month × coefficient ÷ 150 hours. Payback = pilot cost ÷ (automatable ÷ 12).

The coefficients are our own conservative estimate, shown here so you can challenge them. 150 hours is one person's productive time in a month (≈1,800 a year).

Book the Plant X-Ray with these numbers

The form will already hold your numbers.

The method, for plants

Ten working days inside your plant.

We don't ask how your process works. We read it from your data.

Day 1

Safety and a walk

Safety induction, meeting the shift, a walk down the line with a stopwatch and a camera.
Days 2–3

Shadowing the shift

Photos of the paper on the desks, a document census, conversations with the people who "just know".
Day 4

The flow, from the data

Lightweight process mining: ERP timestamps (order → delivery note → invoice) plus email and Viber metadata.
Day 5

Energy, downtime, quality

Bills or meter data, breakdown logs, scrap records.
Days 6–7

Camera test

On one line: what the camera sees, in what light, at what speed.
Day 8

Opportunity map

Every opportunity on a Value × Feasibility grid, with hours, euros and risk.
Day 9

A demo on your own material

For example, defects marked on photos of your own line.
Day 10

Readout and charter

To the owner and the CFO. A pilot charter with success thresholds, signed before any code.

What we need from you

Power bills or meter data
Breakdown, downtime and quality logs
An export from the ERP
Access to one line for the camera test

Whatever we do on site, we fit around your shift pattern.

The numbers, with their formulas

Hours freed = Σ (touchless volume × minutes saved) ÷ 60, minus review time
Labour value = hours × 12 × €/hour
Downtime avoided = hours avoided × € per downtime hour
OEE = Availability × Performance × Quality
Payback (months) = one-off cost ÷ net monthly benefit

Pilot targets are signed before any code is written.

The Plant X-Ray, step by step

Honest AI

Every model makes mistakes. Here's where.

The AI-generated game image with three dashed model boxes tagged "to a person": two read "label missing" (0.45 and 0.40) over bottles in the right-hand lane, which runs unlabelled, and one reads "dark spot on glass" (0.38) over the steel guide rail.

Show the model's mistakes on the image

Three false alarms

Here the machine saw two bottles "missing a label". The whole right-hand lane runs unlabelled, in every row: that's how it works, not a defect. The model doesn't know the line's spec.

And here it saw a "dark spot on the glass". It's scuffing and reflection on the steel guide rail, not on the bottle.

Confidence 0.45, 0.40 and 0.38: the three lowest in the run. In a production system, anything below the reject threshold isn't binned; it goes to a person.

On this image, a 0.5 threshold would drop all three false alarms and keep all 8 finds. At 0.6 it would lose the underfill, which the model found at only 0.55. A threshold is a trade-off; we set it together, on your own parts.

And what did it miss? On this image, nothing. Its defects are large and clear; on your own line they may be smaller, and the frame moves. That's why the inspector stays on the line, and why the system is trained during the pilot on defect photos from your own line.

How we contain it

A confidence threshold. Below the line we agree together, the machine rejects nothing on its own.
A human review queue. The grey zone goes to an inspector, with the photo and the reason.
Logs. Every decision is recorded: what it saw, how sure it was, what happened next.
Retraining. Every correction the inspector makes becomes an example for the next version of your model.

In the pilot we measure accuracy and false alarms on your own parts, and hand them to you in writing.

What we don't promise

A factory that runs itself.
A percentage saving before the X-Ray.
Fewer people.

The hours we free up go where you need people.

The Netex AI Trust Standard

Proof

Other people's numbers, until ours are in.

Our measured industrial cases will go here, with a method note, a baseline period and what the AI didn't do. Until then, this is what others have measured, with the source and what kind of source it is.

What others have measured

≈ 2,000 hours

of unplanned downtime and 53 interruptions avoided over 3 years with predictive maintenance (BlueScope, 2022)

Siemens (opens in a new tab)
≈ 9%

lower production cost from process optimisation at a steelmaker (Gerdau, 2021)

TechTarget (opens in a new tab)
27 hours

a month lost to unplanned downtime at the average large plant

Siemens, True Cost of Downtime 2024 (opens in a new tab)
up to −40%

cooling energy with machine learning, in a data centre (2016): an analogue, not a factory

DeepMind (opens in a new tab)
69% · 3–4%

of Greek SMEs need predictive maintenance; 3–4% use it

AI Readiness Index 2025 (ENTERSOFTONE + AUEB/ELTRUN) (opens in a new tab)

Funding

Which programme covers what.

We don't sell grants. We show you which programme fits, if any does.

Last reviewed: 13 October 2026

Which programme covers what.
Programme	Status	What it could cover	Where it fits
Development Law, "Manufacturing – Supply Chain" scheme, cycle 4 (2026) source (opens in a new tab)	Call decision dated 16 April 2026	Investment in equipment and automation: cameras, meters, systems	Energy and vision inside the plant's investment plan
New AI scheme for SMEs (€150m) source (opens in a new tab)	Announced · terms unpublished · checked 13 October 2026	According to the announcements: AI applications, staff training, integration consultants, infrastructure	X-Ray, pilot, team training
ESPA: Advanced Digital Transformation of SMEs source (opens in a new tab)	Closed to new applications; implementation extended to 31 months	For approved beneficiaries: AI-based data analysis and production optimisation	Delivering an approved project that has stalled
Thessaly: "Αναπτύσσομαι έξυπνα" ("Growing smart", 70% grant) source (opens in a new tab)	Closed on 31 March 2026; approved projects have 24 months to deliver	Projects of €50k–425k; software and SaaS within limits	Magnesia businesses with an approved project
Recovery Fund loans via the Hellenic Development Bank source (opens in a new tab)	Moving to the Development Bank from 1 September 2026	Low-interest financing for digital transformation projects	Scaling up after the pilot
Social Climate Fund source (opens in a new tab)	Announced, being rolled out	Part of the budget for SME infrastructure and energy-cost reduction	Energy measurement and optimisation

Subject to the current call text. We never build a project that only works if a grant comes through.

Every X-Ray includes a funding annex: which programme could cover which line of the project.

Questions

Before you call us.

What can AI actually do in a factory?

Four things, in order of risk: it measures and shifts energy use, checks every part with a camera, predicts breakdowns from vibration and current, and supports planners and operators. Alongside that, it reads the plant's paperwork: certificates, delivery notes, HACCP logs. Which of these makes sense for you is what the X-Ray shows.

Do we need to buy new equipment?

Often not. We start from what you have: IP cameras, energy meters, PLC and ERP data. Where something is missing, such as a vibration sensor on a critical motor, the X-Ray says so, with the cost and the reason. Partners supply the hardware; we take on the data, the models and the alerts.

How much does it cost?

The Plant X-Ray starts at €4,900 and is credited in full against a pilot signed within 60 days. Pilots start at €6,000; an industrial pilot is priced after the X-Ray, once we know the line, the data and the target.

How long does a pilot take?

Six to twelve weeks, depending on the step: energy 6–10, vision 6–8 on one line, predictive maintenance 8–12 on 3–5 critical assets. Success thresholds are signed before the build starts. If they aren't met, you see it in the same numbers we do.

Will it replace our people?

No, and we don't sell it that way. AI takes the retyping, the searching and the sample checks. The inspector judges the doubtful parts, the technician repairs, the planner decides. The hours freed go where you need people, and we measure them.

What happens to our production data and images?

We follow the Netex AI Trust Standard: processing in the EU, a data processing agreement, every decision logged. Your data doesn't train third-party models. The cameras look at the product and the line, not at people: we don't score employees.

Does it work with our ERP and machines?

That's the job: we connect what you already have. ERPs such as SoftOne, Entersoft or Epsilon, myDATA through your provider, PLC and meter data. If a connection can't be made, you find out in the X-Ray, before you pay for a pilot.

Is there funding?

Possibly. There's no open programme for everyone today. The new AI scheme for SMEs has been announced, but its terms aren't published yet. The Development Law covers equipment investment, and approved ESPA beneficiaries are still delivering. Every X-Ray has an annex listing what fits. We don't sell grants.

Where are you, and how soon can you come?

In Heraklion and inside the Almyros industrial area. We visit plants across Thessaly and Crete, and we sit your safety induction before we go near the line.

Plant X-Ray

Let us X-ray your plant.

Ten working days. An Hours-Leak Map, five opportunities with the payback worked out, a live demo on your own data and a funding annex.

From €4,900 · credited in full against the pilot

Book a Plant X-Ray Start with a free 45-minute Scan call

We reply within one working day. No commitment until you've seen the proposal.

Your plant's biggest cost isn't on any invoice. Industrial AI for Greek manufacturers, from an office inside the Almyros industrial area

Europe has already started.

Twelve notes from plants like yours.

AI on the plant floor isn't a chat window.

Reads

Sees

Hears

Predicts

Optimises

Decides

This is how an order travels through your company.

Plastics: from RFQ to delivery

Metals: breakdown → repair → certificate

Food: forecast → production → waste

Wholesale: Viber order → invoice

Logistics: truck arrival → POD → billing

Olive oil: from olive intake to export

Ready-mix: order → dispatch → delivery ticket

Spot the defect.

All eight, one by one

What the machine found

Decode

Four steps, in order of risk.

Who touches your line, and how.

Smart Plant Starter

Put your own numbers in.

What we'd build

The formula

Ten working days inside your plant.

Safety and a walk

Shadowing the shift

The flow, from the data

Energy, downtime, quality

Camera test

Opportunity map

A demo on your own material

Readout and charter

What we need from you

The numbers, with their formulas

Every model makes mistakes. Here's where.

Three false alarms

How we contain it

What we don't promise

Other people's numbers, until ours are in.

Which programme covers what.

Before you call us.

Let us X-ray your plant.