Sectors · Metals & steel
“The crane stopped at 23:40. The one person who knows why is asleep.”
AI for steel and metal fabrication
In metals, the share of firms using AI is three times higher in Europe than in Greece.
Firms using AI
source (opens in a new tab)Eurostat, AI use by sector (isoc_eb_ain2), 2025 · official statistics · firms with 10+ staff · basic metals and fabricated metal products (NACE C24–C25)
If you make metal, or work it for the big mills, a stoppage costs more than hours. It threatens your standing as an on-time, in-full supplier and pushes you into expensive buffer stock. source (opens in a new tab)Siemens, True Cost of Downtime 2024 · vendor report
Our office is inside the Almyros industrial area, next door to the steelworks.
The invisible
Four sentences from every shop floor.
We hear them from shift leads, quality managers and owners. None of them shows in cost per tonne until someone measures it.
-
“The crane stopped. We found out when it stopped.”
- The hidden metric
- Unplanned downtime hours, MTBF
- What AI does
- Predictive maintenance: vibration, current and temperature on 3–5 critical assets The service
-
“When Nikos is away, nobody knows how to restart it.”
- The hidden metric
- MTTR, calls to one person
- What AI does
- A maintenance copilot that answers from manuals, past work orders and shift logs, by voice too The service
-
“We type the 3.1 certificate by hand from the lab results.”
- The hidden metric
- Minutes per certificate, deviations from the customer’s spec
- What AI does
- Document AI: writes the EN 10204 3.1 certificate and checks it against the order The service
-
“We add alloy ‘just in case’ to every heat.”
- The hidden metric
- Alloy kg per tonne, kWh per tonne, off-spec heats
- What AI does
- Recipe optimisation with explainable machine learning The service
What we build
Fewer stoppages. More proof.
In metals, the customer is also buying reliability: delivery on time and a certificate without errors. The left column cuts stoppages and paperwork. The right makes your reliability visible to the buyer.
AI & Custom Solutions
Predictive maintenance
Vibration, current and temperature sensors on mills, cranes, pumps and fans. The alert comes on Wednesday; the stop happens when you plan it.
- What we measure
- Unplanned downtime hours a month, MTBF, maintenance €/tonne
- Pilot or estimate
- 8–12 week pilot on 3–5 critical assets A planning range for a pilot in a Greek SME. Your own figure is measured in the pilot.
- Evidence
- BlueScope: about 2,000 hours of downtime and 53 interruptions avoided in 3 years (vendor data). source (opens in a new tab)Siemens on BlueScope · vendor data
Recipe and alloy optimisation
Scrap mix, ferroalloys, power-on time: the model says when “just in case” isn’t needed, and explains why.
- What we measure
- Alloy kg/tonne, kWh/tonne, % off-spec heats
- Pilot or estimate
- 10–14 week pilot, with heat-level history A planning range for a pilot in a Greek SME. Your own figure is measured in the pilot.
- Evidence
- Gerdau: about 9% lower production cost through process optimisation (company statement, 2021). source (opens in a new tab)TechTarget on Gerdau (2021) · company statement
EN 10204 3.1 certificates
The lab produces results. The system writes the certificate and checks every value against the customer’s spec before it leaves.
- What we measure
- Minutes per certificate, deviations caught
- Pilot or estimate
- 4–6 week pilot A planning range for a pilot in a Greek SME. Your own figure is measured in the pilot.
- Evidence
- A benchmark from another process, invoices: 19.9% exceptions on average, 11.8% for the best. source (opens in a new tab)Ardent Partners via Medius · vendor-published research
Maintenance copilot
Ask in Greek, by voice, from the shift tablet. It answers with the manual page and the past work order.
- What we measure
- MTTR, calls to senior staff, time to competence
- Pilot or estimate
- 6–8 week pilot A planning range for a pilot in a Greek SME. Your own figure is measured in the pilot.
- Evidence
- In a call centre, new agents with an AI assistant resolved 34% more issues an hour. A different job, the same problem: senior people’s know-how. source (opens in a new tab)NBER w31161 · academic
Energy loads
Moves whatever can move out of the peak windows.
- What we measure
- Blended €/MWh, peak kW
- Pilot or estimate
- 6–10 week pilot A planning range for a pilot in a Greek SME. Your own figure is measured in the pilot.
- Evidence
- Siemens names energy prices as the main reason downtime got more expensive. source (opens in a new tab)Siemens, True Cost of Downtime 2024 · vendor report
No savings percentage before the X-Ray. We measure it in your own business.
Marketing & Growth
-
Batch Passport
A QR or NFC tag on every bundle, coil or pallet: heat number, certificate, test values, shift. Buyers can attach it to their tenders and reports.
- What we measure
- Scans, tenders where transparency is scored, quality disputes
-
The Glass Factory
A live page with production counters for buyers. The same plumbing as predictive maintenance.
- What we measure
- Buyer portal logins, RFQs from abroad
-
AI visibility for export buyers
Grades, dimensions and tolerances in HTML, in English and German, so AI answers name you.
- What we measure
- Mentions across 30 prompts, RFQs from AI answers
Web & Apps: where the two meet
A customer portal with order progress and each batch’s certificates, connected to the ERP.
Web & AppsX-Ray
This is how a breakdown travels through your plant.
One stoppage, one call to Nikos, one binder, one warehouse spreadsheet. And every day, certificates typed by hand.
From breakdown to certificate
How this mechanism runs for a business in this sector, step by step.A scan line passes over the steps and marks the hot ones.
-
2 h
The machine stops. The operator looks for the shift lead.
Predicts The vibration anomaly shows days earlier. The stop is planned.
10′2′ Minutes on the new path -
5 h
The shift lead calls Nikos
Reads The copilot answers from manuals and past work orders
25′5′ Minutes on the new path -
8 h
Searching binders of manuals and old work orders
Reads Merged into the previous step
40′0′ Minutes on the new path -
6 h
Spare part: in stock or not? Warehouse spreadsheet
Acts Checks stock in the ERP and raises a request
30′3′ Minutes on the new path -
3 h
Repair report by hand, then into the system
Hears The technician dictates it and it files itself
15′3′ Minutes on the new path -
80 h
3.1 certificate from the lab results
Reads Writes the certificate from the results and the order. The manager signs.
12′3′ Minutes on the new path -
40 h
Checking the certificate against the customer’s spec
Reads The check happens as it is written. Deviations are flagged.
6′0′ Minutes on the new path -
8 h
Weekly OTIF report for the key customer, in a spreadsheet
Acts Builds itself from the ERP. The manager reads it.
120′15′ Minutes on the new path -
29 h
Orders from customer portals and email are retyped
Reads Reads the order and opens a draft in the ERP
7′1′ Minutes on the new path
| Step | Who | Minutes | Times a month | With AI | Minutes on the new path |
|---|---|---|---|---|---|
| The machine stops. The operator looks for the shift lead. | shift | 10 | 12 | Predicts: The vibration anomaly shows days earlier. The stop is planned. | 2 |
| The shift lead calls Nikos | shift | 25 | 12 | Reads: The copilot answers from manuals and past work orders | 5 |
| Searching binders of manuals and old work orders | maintenance | 40 | 12 | Reads: Merged into the previous step | 0 |
| Spare part: in stock or not? Warehouse spreadsheet | maintenance | 30 | 12 | Acts: Checks stock in the ERP and raises a request | 3 |
| Repair report by hand, then into the system | maintenance | 15 | 12 | Hears: The technician dictates it and it files itself | 3 |
| 3.1 certificate from the lab results | quality | 12 | 400 | Reads: Writes the certificate from the results and the order. The manager signs. | 3 |
| Checking the certificate against the customer’s spec | quality | 6 | 400 | Reads: The check happens as it is written. Deviations are flagged. | 0 |
| Weekly OTIF report for the key customer, in a spreadsheet | sales | 120 | 4 | Acts: Builds itself from the ERP. The manager reads it. | 15 |
| Orders from customer portals and email are retyped | sales | 7 | 250 | Reads: Reads the order and opens a draft in the ERP | 1 |
Measure your hidden cost
The X-Ray above is someone else’s day. The meter takes yours: a few sliders, your own cost per hour, the formula on the page.
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 hours per month
- 57
- What could be automated
- 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).
The form will already hold your numbers.
Funding
Which programme covers what.
We don’t sell grants. We show you which programme fits, if one does.
| Programme | Status | What it can cover |
|---|---|---|
| Development Law, “Manufacturing – Supply Chain”, cycle 4 | 2026 call | Sensors, automation, production equipment source (opens in a new tab)Ministry of Development, Development Law calls · official |
| €150M AI scheme for SMEs | Announced. Terms and date not yet published. | According to the announcements: AI applications, training, integration consultants. The terms will confirm it. source (opens in a new tab)ictplus, €150M AI scheme · press |
| Social Climate Fund | Announced, being rolled out. Part of the budget goes to SME infrastructure and energy costs. | The measurement layer behind energy-saving projects source (opens in a new tab)taxrevenue.gr, new SME schemes and Social Climate Fund · press |
| Thessaly, “Anaptyssomai exypna” (70%) | Closed to new applications. Approved firms implement within 24 months. | Software, licences and SaaS within the call’s limits source (opens in a new tab)Anaptyxiaki, Thessaly «Anaptyssomai exypna» · consultancy |
Subject to the text of the current call. Every X-Ray includes a funding annex.
Proof
Become our first measured project in your sector.
We ask one thing: permission to publish the numbers, with the method and the name of the person who got their hours back.
We’d like to be firstWhat other people’s measurements say
- BlueScope: about 2,000 downtime hours avoided in 3 years (source (opens in a new tab)Siemens on BlueScope · vendor data, vendor data)
- Gerdau: about 9% lower production cost (source (opens in a new tab)TechTarget on Gerdau (2021) · company statement, company statement, 2021)
- Large plants: 27 hours of unplanned downtime a month (source (opens in a new tab)Siemens, True Cost of Downtime 2024 · vendor report, vendor report)
These are other people’s numbers, with sources. We measure yours.
Questions
What does AI do in a steelworks or a fabrication shop?
It predicts failures on cranes, mills and pumps before they stop, writes and checks EN 10204 3.1 certificates, makes senior technicians’ know-how available to every shift, and optimises recipes and energy loads. We start where the risk is lowest and the proof comes fastest.
Do we need new sensors?
For predictive maintenance, usually yes, on 3–5 critical assets: vibration, current, temperature. First we look at what your PLC or SCADA already records. Hardware goes in only where the data doesn’t exist, and we say so before the pilot, not after.
What does an hour of downtime cost?
It depends on your plant, which is why we measure it in the X-Ray. For reference, the large plants Siemens studied had 27 hours of unplanned downtime a month (vendor report). We work out the hourly cost from your own numbers: lost output, overtime, restart energy, late-delivery penalties.
Does our production data leave the plant?
Only what’s needed, processed in the EU, under a data processing agreement, and with no model training on your data. For sensitive lines we discuss processing inside the plant. All of it is written down in our AI Trust Standard.
Is there funding?
Possibly. The Development Law (“Manufacturing – Supply Chain”) covers equipment and automation, and a €150M AI scheme for SMEs has been announced, with terms not yet published. Every X-Ray includes an annex saying which programme fits, if one does.
X-Ray
Let us X-ray your plant.
Ten working days on site: downtime and work-order history, PLC and energy data, a census of your certificates, conversations with “the people who know”. You leave with a map of hours and euros, and a demo on your own certificates.
from €4,900
The form opens with your sector selected and whatever you pinned. Nothing goes into the page address.
- Credited in full against a pilot signed within 60 days.
- A free 45-minute call first, to see whether it’s worth it.