Η σελίδα υπάρχει και στα ελληνικά. Διαβάστε τη στα ελληνικά
AI & Custom Solutions
Four ways to get hours back, a written standard for how we do it, and a diagnosis that tells you where to start.
What AI does inside a business: the overviewSectors
Not sure where to start?
Workflow X-Ray
In one to two weeks: a map of hours and euros, five opportunities with the payback worked out, and a demo running on your own data.
From €890 · credited 100% against the pilot
Proof
No logo wall. Just this.
Every claim on this site is either playable, measured or sourced. Here you'll find all three, each one labelled, so you know what you're looking at.
How to read it
If you see a number without a label, that's our mistake. Tell us at info@netex.gr.
Real projects
Two projects we can show today. One stays a story until its numbers are verified. The other is what you're reading.
Field Case 01 · e-shop
A car drives through an Instagram Reel. In a single frame, a discount code. People rewound and paused, again and again, to catch it.
The numbers aren't here yet. They go up once we've verified them from the client account's Instagram Insights.
Method note The numbers will come from the client account's Instagram Insights, with dates and screenshots.
this site
This site's speed, accessibility and security, measured on launch day, with the method next to every number.
Measured on launch day. The targets are already here; the measurements go next to them as soon as they're taken.
live reading Measured now, in your browser, on your connection. If anything is off, you'll be the first to see it.
Every file here has an incident, a hidden cost, an intervention, a result, a payback period, and what the AI didn't do. We're looking for the first client in each sector to be measured in public, with the method in the open.
Proof you can play
Every demo is yours to play. Where there's a model, you see its mistakes too.
The demos are tools we built so you can see the mechanism for yourself.
Open files
One reel for one e-shop is one story. Here are ten more, from the weighbridge of a scrap yard to the tsikoudia still. For each: what happens, what it costs, what we'd build, what we'd measure, and what the AI wouldn't do.
Ten use cases, written on the same Black Box form as our projects. Volumes and minutes are an example of the calculation, in plain view; the hourly cost is official Eurostat data. Client projects carry the "real project" label.
Drawings are drawings. Photographs on this site are of real places, or carry the "AI-generated image" label.
The grade gets called by eye, at the weighbridge, with the truck waiting. Then we write it down by hand.
What would disappear The handwritten ticket, and typing it all into the ERP in the afternoon.
A yard that buys scrap from garages, demolition jobs and small collectors and sells it to the steel mills nearby. Every load is weighed, the grader looks into the truck bed and calls a grade, and the ticket is written by hand. In the afternoon someone types it all into the ERP and prepares the transport paperwork. When a supplier disputes the grade, there's no photo for anyone to look at.
≈ €13,200–16,800 a year, in hours alone
The formula 660 loads a month × 5.5–7 fewer minutes of paperwork per load ÷ 60 ≈ 60–77 hours a month. × 12 × €18.2 an hour ≈ €13,200–16,800 a year.
Assumptions 30 loads a day, 22 working days. About 9 minutes of paperwork per load today: ticket, ERP, transport documents. After: 2 to 3.5 minutes, human check included.
Average labour cost €18.2/hour (industry, construction and services): Eurostat lc_lci_lev, 2025 · official statistics. Volumes and minutes: example calculation.
The bigger cost isn't the hours. It's the wrong grade at the weighbridge, whichever way it tips. That can only be measured on site.
In a pilot we'd measure: minutes per load, weighbridge to ERP · truck waiting time · grade disputes a month · how often the grader agrees with the model's suggestion
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't decide the grade or the price: it suggests, and the grader decides. It can't see below the top of the load; whatever's hidden underneath, the person at the tipping point sees.
The grader at the weighbridge. Instead of writing, they look: at the loads the model flagged, and at anything that mustn't reach the furnace.
Two weeks of baseline before anything else: a stopwatch from weighbridge to ERP, and a count of disputes. Then four weeks of pilot, the model suggesting without deciding, every suggestion logged next to the grader's decision.
Watch a vision model work in front of you, mistakes included.
The camera looks at the truck bed, not the driver. Signage at the gate and notice to staff and suppliers (GDPR). Photos are kept only as long as the retention policy says.
The architect wants to see that exact slab. We send a phone photo, taken in the sun.
What would disappear Phone photos with the sun glaring off them, and the slab somebody promised twice.
Four hundred slabs in the yard, each with its own veining. Clients choose from phone photos full of reflections, sizes are written in marker on the edge, and the same slab gets held for two clients because nobody wrote it down. A crack shows up once the slab is already on the truck.
≈ €7,000 a year in hours
The formula Today: 40 "send me a photo" requests a week × 15 minutes in the yard ≈ 43 hours a month. After: 2 minutes per request to send a link, plus 5 minutes of photography for each of the month's 60 new slabs ≈ 11 hours. That gives back ≈ 32 hours a month: × 12 × €18.2 an hour ≈ €7,000 a year.
Assumptions 40 requests a week, 60 new slabs a month. Photographing at intake is new work, and it's subtracted from the benefit.
Average labour cost €18.2/hour (industry, construction and services): Eurostat lc_lci_lev, 2025 · official statistics. Volumes and minutes: example calculation.
The costly part isn't the hours. It's the slab held for two clients, and the crack spotted on the truck.
In a pilot we'd measure: yard hours on photos and searches · slabs held twice · cancellations and returns over cracks · time from enquiry to hold
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't certify a slab as crack-free: it flags possible cracks, and a person checks them. It doesn't grade quality or set a price. It doesn't beautify photos: what the client sees is what the client gets.
The yard manager, who stops running between slabs with a phone and checks only what the model flagged.
Three weeks of baseline: photo requests, time in the yard, double holds, returns. Then every slab photographed once, and two months with the catalogue open to selected architects, measured the same way. Cracks the model flags are logged next to the ones the person finds.
In manufacturing, 7.77% of Greek firms with 10+ staff use AI. In the EU, 17.27% (2025).
Eurostat isoc_eb_ain2 Figures from an official statistics body, with the year shown.
The photos show slabs, not people. Holds and client details stay in your own system. The lighting at the photo spot doesn't change, or the model can't compare properly.
Feed is our biggest bill. And we log it in a wet notebook on the boat.
What would disappear The wet notebook, and copying it into Excel on Friday.
Every morning the feeding boat works its way round eighteen cages of sea bream and sea bass. How much feed went into which cage, only the skipper and the notebook know. The office finds out on Friday, if the handwriting is legible, and the feed conversion ratio comes out once per batch, when it's too late to change anything.
We won't put a number on it before we've seen your notebooks.
The formula Feed conversion ratio (FCR) = kg of feed ÷ kg of fish growth, per cage, per week.
Assumptions There's no reliable public figure for a farm like yours. We'll measure it.
use case
Here the hidden cost isn't hours. It's feed that ends up on the seabed instead of becoming fish, and nobody can say in which cage.
In a pilot we'd measure: FCR per cage per week · feed per cage against the estimate · office hours for the weekly report · mortality per cage
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't feed and doesn't decide how much. It can't see under water: we don't promise cameras that count fish. It doesn't diagnose disease; that's the job of the fish biologist and the vet.
The feeding boat's skipper, who writes less and, for the first time, sees what became of the cage they fed.
Four weeks with the app but no model, to set a baseline. Then six weeks with a per-cage estimate, comparing FCR between cages that followed it and cages that didn't.
Only 1.88% of Greek firms with 10+ staff use AI in production. In the EU, 4.14% (2025).
Eurostat isoc_eb_ai Figures from an official statistics body, with the year shown.
The farm's data stays the farm's. No feeding suggestion is applied without approval, and none concerns treatment.
The ship berths at 06:00 and sails at 22:00. Its list runs to 140 lines in English.
What would disappear Matching line by line, with the catalogue open in the next window.
Requests come from agents and masters by email, PDF and spreadsheet: ropes, filters, paint, provisions, with IMPA codes if we're lucky. Someone matches them line by line to our own codes and stock before the quote goes out. If it's late, the ship buys at the next port.
≈ €5,600 a year in hours. The expensive part is elsewhere.
The formula 25 requests a month × (90 minutes today − 20 minutes with a check) ÷ 60 ≈ 29 hours a month. × 12 × €16.0 an hour ≈ €5,600 a year.
Assumptions Roughly one request per working day, 80 to 200 lines each. The 20 minutes "after" are a person checking the flagged lines.
Average labour cost €16.0/hour in trade: Eurostat lc_lci_lev, 2025 · official statistics. Volumes and minutes: example calculation.
The hours are the small number. The big one is the quote that arrives second: the ship doesn't wait.
In a pilot we'd measure: time from email to quote · lines matched with no correction · quotes won
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't set prices, discounts or credit. It doesn't pick a substitute on its own. And nothing goes to the ship without approval.
The sales lead, who checks ten flagged lines instead of a hundred and forty.
The last ten requests, timed as they were handled, as the baseline. Then the next twenty through the system, with every correction the salesperson makes logged.
At a US logistics-tech firm, preparing a quote with AI went from more than 30 minutes to 3 or fewer.
Toptal A figure published by a company that sells the solution. We show it, but with this label.
Ship and agent data stays in your ERP. Anything involving dangerous goods or safety-critical spares is always checked by a person.
Saturday: 34 departures at 10:00, 34 arrivals at 16:00. I build the plan on Friday night, on a whiteboard.
What would disappear Friday's whiteboard, and Saturday's group chats.
Eight cleaning teams, sixty villas across three municipalities, bookings from three platforms. Changes arrive in group chats, damage is reported with photos that get lost in the chat, and the owner abroad hears about the broken glass from the review.
≈ €5,200 a season in hours
The formula 4 hours of planning on Friday + 5 hours of calls on Saturday + 34 reports × 10 minutes ≈ 15 hours a week. With the system, about 4 remain. That gives back ≈ 48 hours a month in season: × 6 months × €18.2 an hour ≈ €5,200 a season.
Assumptions One big changeover a week, on Saturday, 34 villas, a six-month season. The 4 hours "after" are plan approval and exceptions.
Average labour cost €18.2/hour (industry, construction and services): Eurostat lc_lci_lev, 2025 · official statistics. Volumes and minutes: example calculation.
The costly part isn't the hours. It's the villa that isn't ready at 16:00, and the review that follows.
In a pilot we'd measure: villas ready before arrival time · coordination calls on Saturday · same-day damage reports with photos · cleanliness complaints
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't talk to guests and doesn't decide damage charges. It doesn't change the plan on its own: it proposes, you approve. Cleaning quality is judged by a person, with the photos in front of them.
The operations lead, whose Friday no longer ends at midnight.
Two Saturdays of baseline: which villa was ready when, how many calls were made, how many damages were reported. Then six Saturdays with the system, measured the same way.
Saturday changeovers are a scheduling problem with fixed rules: times, distances, teams. Optimisation algorithms handle that kind of problem well; the person keeps the exceptions.
our reasoning
Guest data stays on the platforms and in the booking system; teams see only what they need. Photos don't show faces or guests' belongings.
Every quote has six variables. I work it out on a calculator, between two presses.
What would disappear The calculator between the presses, and the "same as last time" phone call.
Labels for food producers, leaflets, boxes. Requests come by phone and email, and half are repeats of old jobs. Artwork often arrives without bleed, at low resolution or in RGB, and we find out once the press is already set up.
≈ €8,200–9,800 a year in hours
The formula 300 requests a month × 15 minutes = 75 hours. If 50 to 60% are standard jobs or repeats that clients complete themselves, that gives back 37–45 hours a month: × 12 × €18.2 an hour ≈ €8,200–9,800 a year.
Assumptions Roughly 14 requests per working day. The share that completes on its own is an assumption; it gets measured in the first month. Portal orders need no quote; the pre-press check already happens today and doesn't change.
Average labour cost €18.2/hour (industry, construction and services): Eurostat lc_lci_lev, 2025 · official statistics. Volumes and minutes: example calculation.
Not counting the printing plates remade because one file had no bleed.
In a pilot we'd measure: requests completed with no phone call · time to quote for non-standard jobs · files rejected at upload rather than at the press · repeat orders
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't set the price: the price comes from your formula, not from a model. The artwork check is rules, not guesswork. AI is barely needed here, and we say so.
The owner, who now only sees the interesting quotes: the new ones.
A month of baseline: each request with arrival time, quote time and whether it was won. Then three months with the portal open to the most frequent customers.
From 1 February 2027, documents reach myDATA only through a certified provider or AADE's own apps, no longer directly from an ERP. A portal that already talks to a provider is ready.
according to AADE (30/9/2026) An announcement by a public authority, such as AADE, with its date.
Each client sees only their own prices and jobs. Their files are used for nothing else.
The architect sends a window schedule as a PDF. We copy it out by hand, and the craftsman plans the cuts by eye.
What would disappear Copying out the schedule by hand, and the offcut bin nobody weighs.
A workshop making aluminium windows and doors for apartment blocks and hotels. Every project brings a schedule with dozens of openings: type, size, opening direction, glazing. The quote takes days, and in the workshop profiles get cut in whatever order they come, leaving offcuts that end up in the bin.
≈ €7,900 a year in hours, plus the offcuts
The formula Two parts. First, the hours: 12 quotes a month × 4 hours of copying and calculating = 48 hours; with the schedule read automatically and checked by a person, about 1 hour per quote remains, so ≈ 36 hours a month come back: × 12 × €18.2 an hour ≈ €7,900 a year. Second, the offcuts: kilos of aluminium in the bin each month, before and after cutting optimisation.
Assumptions 12 quotes a month, with 20 to 60 openings each. We don't estimate the offcuts: we weigh them.
Average labour cost €18.2/hour (industry, construction and services): Eurostat lc_lci_lev, 2025 · official statistics. Volumes and minutes: example calculation.
Offcuts are the cost that never shows in the quote but always shows in the bin.
In a pilot we'd measure: hours per quote · kilos of offcuts a month · quotes won · dimension errors that reached the site
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't measure on site or correct the architect's drawings. It doesn't change your prices. And the cutting optimisation isn't even AI: it's decades-old maths, and we use it because it works.
The craftsman at the saw, who cuts in the order the computer worked out and keeps the right to change it.
Two months of baseline: hours per quote, and kilos of offcuts from the bin, weighed every Friday. Then three months measured the same way. Every correction to the schedule the AI read gets logged.
A window schedule is a document with a fixed structure, and cutting into bars is a classic optimisation problem. Both are solved well, as long as a person keeps the exceptions.
our reasoning
Dimensions are always confirmed by an on-site measurement before cutting. Architects' drawings are used only for their own project.
In October we fire up the stills. People find out in December, from the photos.
What would disappear December's photos of something that happened in October.
The year's strongest story happens over a few weeks in autumn, in front of very few people. Cretans abroad, the tavernas that buy from us and autumn visitors see it, if at all, once it's over.
We don't put a price on a story. We measure what it does.
The formula There's no formula in euros. There's a question: how many people would have wanted to be there, or to own a bottle from that night, and never found out?
Assumptions
use case
Here the hidden cost isn't hours. It's a story that happens once a year, and nobody sees it while it's happening.
In a pilot we'd measure: channel sign-ups · page visits on still nights · pre-orders of numbered bottles · tavern orders after the season
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
Nothing. This mechanism doesn't need AI. It needs a button, a channel and a numbered label. We say so, because diagnosis cuts both ways.
The distiller. The button is theirs to press, not an algorithm's.
We measure the season: sign-ups, visits on still nights, pre-orders and tavern orders, against the same period last year from your own records.
What's scarce, we want more: scarcity makes a thing more desirable.
Worchel et al., 1975 A published scientific study.
Adults only: an 18+ audience on every ad, age checks in the e-shop and on delivery. No message encourages excessive drinking. The rules on alcohol advertising, distilling licences and online sales are checked with a lawyer before any campaign.
One in ten comes out of the kiln with a mark. We hide it on the back shelf.
What would disappear The back shelf, and the discount said under your breath.
A bubble in the glaze, a drip that ran, a slight warp. The pieces are sound and useful, but they're sold quietly, at half price, to whoever happens to ask.
It's measured with one count of the back shelf.
The formula Second-choice stock = pieces per firing × share with a mark × firings a month.
Assumptions "One in ten" is how an owner would put it, not a statistic.
use case
What's hidden here is stock, not hours: pieces that don't sell because we don't show them.
In a pilot we'd measure: second-choice sales a month · "not as pictured" returns · saves and sends of the kiln reels · orders that also include a first-choice piece
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't judge which piece is second choice. It doesn't hide marks with image editing: the photos stay as shot. It publishes nothing the potter hasn't approved.
The potter, whose judgement becomes the content.
Count the back shelf on day one. Then three months with the pieces in the e-shop, comparing second-choice sales and returns with the previous three months.
When someone competent shows a small flaw, we like them more.
Aronson et al., 1966 A published scientific study.
The description of the mark must be exact: what the photo shows is what the customer gets. Reels show hands and pieces, no faces without consent.
The owners are in Hamburg, Lyon, Milan. Every Monday someone rings to ask whether their boat is all right.
What would disappear The Monday phone call, and the email that waits a week for a "yes".
Eighty boats on the hard for the winter, owners in five countries. Extra work found during cleaning (a worn anode, a crack in the gelcoat) waits for approval in emails that get answered a week later. And next year, the owner asks an AI assistant where to winter the boat.
≈ €2,300 a winter in hours. The expensive part is elsewhere.
The formula 80 boats × 2 calls a month × 8 minutes ≈ 21 hours a month, plus the approval emails. × 6 months × €18.2 an hour ≈ €2,300 a winter.
Assumptions Two calls a month per boat, eight minutes each, six months of storage.
Average labour cost €18.2/hour (industry, construction and services): Eurostat lc_lci_lev, 2025 · official statistics. Volumes and minutes: example calculation.
The hours are the small number. The big one is the work waiting for approval, and the boat that goes elsewhere next year.
In a pilot we'd measure: incoming owner calls a week · time from "we found something" to approval · share of extra work approved · bookings for next winter
Worked out after the X-Ray as pilot cost ÷ net monthly benefit, once human review time is subtracted. Pilots from €6,000.
It doesn't decide repairs or set prices. It writes nothing about a boat's condition that the yard manager hasn't seen and approved. It doesn't answer questions about seaworthiness.
The yard manager, whose Mondays go quiet while approvals come back the same day.
For one month we count calls and approval times as they are. Then the rest of the winter with Friday reports; and in spring, bookings for next winter against the previous year.
When we see the work being done for us, we value it more.
Buell & Norton, 2011 A published scientific study.
Every report says the text was drafted by AI and approved by a person. Owners' data stays in the yard's system (GDPR).
All the better. Tell us what's eating your hours, and the eleventh file will be yours.
The Workflow X-Ray starts from €890 for one process and is credited 100% against the pilot.
