Skip to main content Skip the demo: read how it works

Η σελίδα υπάρχει και στα ελληνικά. Διαβάστε τη στα ελληνικά

Book an X-RayX-Ray Book a Workflow X-Ray: a fixed-price diagnosis

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 overview

Services

Sectors

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

What we deliver, day by day

Automations that give hours back

Monday, 08:10. In accounts, a stack of invoices is waiting to be typed in again.

We type every invoice in by hand. Every one.AI process automation: your hours, back.

We read the documents your team types in today, check them and post them to your ERP and myDATA. Anything uncertain is approved by a person. The hours left over go back to work that needs one.

Book a Workflow X-RayTry it: watch it read an invoice

Try it in your browser. Nothing is uploaded anywhere.

A stack of paper delivery notes and invoices beside a keyboard under a desk lamp in a small business office at night.retyped by handThis image was created with an AI tool (AI Act, Article 50).

The demo didn't load. The text below says what you would have seen.

Give it a document

An example that shows how it works.

Drag a document into the frame, or pick one with the keyboard.

Packaging 47 P.C.

VAT number (ΑΦΜ) 123456783

Invoice A-4471

12/10/2026

  1. Stretch film 23 µm, 50 rolls × €18.40 = €920.00
  2. Wooden pallets 80×120, 20 × €6.50 = €130.00
Net amount
€1,050.00
VAT
VAT 24%: €252.00
Total
€1,302.00

Delivery note DN-1208

Sender
Right Angle Transport G.P.
Recipient
Cyan Diagonal Plastics S.A.
Loading point
Volos industrial area
Purpose of movement
Sale
  1. CODE LD-220, 24 sacks × 25 kg
  2. CODE MB-07, 4 sacks × 25 kg

Polymerwerk 47 GmbH

Certificate of analysis

Batch 26-0417

  1. MFI 2.0 g/10 min (limit 1.8–2.2)
  2. Density 0.922 g/cm³ (limit 0.920–0.924)
  3. Moisture 0.03% (limit ≤ 0.05%)

Ready for approval

ERP · new purchase entryHow sure the model is about this field. Below the threshold you set, the field goes to a person.

Supplier
Packaging 47 P.C.confidence 99%
VAT number (ΑΦΜ)
123456783confidence 98%
Document no.
Invoice A-4471confidence 99%
Date
12/10/2026confidence 99%
Lines
Stretch film 23 µm, 50 rolls × €18.40 = €920.00confidence 96%
Lines
Wooden pallets 80×120, 20 × €6.50 = €130.00confidence 94%
Net amount
€1,050.00confidence 99%
VAT
VAT 24%: €252.00confidence 99%
Total
€1,302.00confidence 99%
myDATA classification
Purchases of raw materials and suppliesconfidence 91%

VAT number valid: the check digit matches The last digit of a Greek VAT number is calculated from the first eight. If it doesn't match, something was mistyped. We check it before anything reaches the ERP.

The totals reconcile

Sender
Right Angle Transport G.P.confidence 98%
Recipient
Cyan Diagonal Plastics S.A.confidence 97%
Document no.
Delivery note DN-1208confidence 99%
Loading point
Volos industrial areaconfidence 95%
Purpose of movement
Saleconfidence 99%
Lines
CODE LD-220, 24 sacks × 25 kgconfidence 95%
Lines
CODE MB-07, 4 sacks × 25 kgconfidence 93%
Supplier
Polymerwerk 47 GmbHconfidence 98%
Batch
Batch 26-0417confidence 99%
MFI
2.0 g/10 min (limit 1.8–2.2)confidence 97%Within limits
Density
0.922 g/cm³ (limit 0.920–0.924)confidence 96%Within limits
Moisture
0.03% (limit ≤ 0.05%)confidence 95%Within limits
Check result
Within limits on all three measurements

By hand: minutes · Here: 0:06The time here is an example. We time the by-hand version on your own workflow, in the X-Ray.

Why it works

Mechanism
Vision plus rules. The model reads, the rules check (VAT number, VAT, totals), a person decides on the exceptions.
Why it holds
We don't trust the model on its own. Every field carries a confidence score and every number passes a rule.
What we measure
Minutes per document · share with no human touch · corrections per 100 documents
What to expect, honestly
We give you a range after the X-Ray, from your own documents. Not before.

Show me this on my own invoices

Static version: the end state, step by step.

The problem

Five minutes here, ten there. Across the whole team.

No line in your accounts says “retyping”. The cost is sliced thin: a few minutes per invoice, per order, per email, spread across many people. That's why nobody sees it, and why nobody fixes it.

Meanwhile, 69% of Greek firms struggle to find skilled staff. Every hour lost to copying is an hour missing from work that needs judgement. EY Barometer 2026 via OT.grsource EY Barometer 2026 via OT.gr (opens in a new tab)

  1. Copying from PDF into the ERP
  2. Checking VAT numbers and totals by eye
  3. Looking up a price for a quote
  4. Retyping an order from Viber
  5. The same report, every Monday
  6. Chasing paperwork
minutes nobody countsThe bar is an example. The X-Ray puts in your own minutes.Example values that show how the calculation works.

Measure your hidden cost

E-invoicing

Every invoice is data now. Let it do the work.

From 2 November 2026, B2B e-invoicing becomes mandatory for businesses with 2023 revenue up to €1 million too, with a transition period until 31 January 2027. Every invoice you receive now arrives as a structured file, not as paper.

And one change that affects your ERP: from 1 February 2027, invoices reach myDATA only through a certified provider or AADE's own apps, no longer directly from an ERP. If your ERP transmits directly today, that connection has to change.

Compliance is the minimum. If someone then opens the file and types it into the ERP again, the gain is lost. We make it post, check and classify itself, with the accountant approving.

AADE, press release on e-invoicing, 30/9/2026AADE, press release on e-invoicing, 30/9/2026

How ready are you? Check in 2 minutes

Where we start

Five automations that pay back first

We don't start from the tool. We start from the workflow that eats the most hours at the lowest risk.

Supplier invoices and delivery notes → ERP and myDATACopying from PDFs and email into the ERP.We measure: minutes per document · exceptions %

Before and after

Average cost to process an invoice: $9.90, against $2.67 for the best teams. We measure yours in the X-Ray.

Ardent Partners via Mediussource Ardent Partners via Medius (opens in a new tab)

A person approves here

Exceptions and low-confidence fields go to accounts.

Quotes from email or an RFQHunting for prices across three files for every quote.We measure: quote turnaround · win rate

Before and after

At a US logistics firm: from 30–40 minutes to 3 or less per quote.

Toptal, US logistics casesource Toptal, US logistics case (opens in a new tab)

A person approves here

The salesperson checks and sends. Prices come from the ERP, never from the model.

Orders from email, Viber or phone → ERPTyping the same order in a second time.We measure: minutes per order · wrong lines

Before and after

We measure it in the X-Ray: minutes per order and wrong lines.

Use case

A person approves here

Unknown item code or a price off the list? The order waits for approval.

Inbox triage with drafted repliesThe morning hunt for the one urgent email among all the rest.We measure: time to first reply · emails per person

Before and after

20–40 hours a month per busy shared inbox (info@, say). For comparison: in a study of 5,179 customer-support agents, an AI assistant raised average productivity by 14%.

Netex estimate from the SME automation research (track 03, §4) · NBER w31161source NBER w31161 (opens in a new tab)

A person approves here

A person sends every reply. Nothing sensitive goes out on its own.

Weekly management reportThe same report, every Monday, from scratch.We measure: hours a week · time the report lands

Before and after

We measure it in the X-Ray: hours a week to produce the report.

Use case Data

A person approves here

The numbers come from the data model. The AI writes the text; a person reads and signs it.

For accounting offices

How many more clients could the same team take on?

Your clients' expenses are already in myDATA. We take them from there, check them, suggest a classification and put them in an approval queue. The accountant approves what's certain in one click and only thinks about what isn't.

An office with 200 clients handles about 6,000 documents a month. A minute each is 100 hours. Netex estimate from the SME automation research (track 03, §4)

The demo didn't load. The text below says what you would have seen.

  1. Pull

    Incoming e-invoices and expenses from myDATA (REST API) or the e-invoicing provider.

    myDATA API v2.0.0

  2. Check

    VAT-number check digit, VAT per rate, totals, duplicates.

    VAT no. valid

  3. Suggested classification

    Expense category and classification with a confidence score, based on that client's history.

    confidence 96%

  4. Approval queue

    The accountant sees low-confidence items first. The rest are approved in bulk.

    1 waiting

  5. Post

    Into the office ERP (for example SoftOne via Web Services), with classifications sent to myDATA under the accountant's key.

    posted

Accountant's queue 1 waiting

  • Mobile operator€89.00Telecoms or a fixed asset?confidence 54%

Post

  • Fuel station€62.40Fuel and lubricantsconfidence 97%
  • Wholesaler€1,284.50Purchases of goods for resaleconfidence 95%
  • Restaurant€146.00Hospitality expensesconfidence 88%

We don't promise a percentage before the X-Ray. We measure your own documents and tell you how many hours come back.

An example that shows how it works.AADE, myDATA REST API v2.0.0 · SoftOne, Web ServicesAADE, myDATA REST API v2.0.0 · SoftOne, Web Servicessource AADE, myDATA REST API v2.0.0 (opens in a new tab)source SoftOne, Web Services (opens in a new tab)

See it on one of your clients' expensesAccounting offices: everything we do

The maths, in the open

1,500 documents a month. 1,200 go through on their own.

Example values that show how the calculation works.

The demo didn't load. The text below says what you would have seen.

The maths, in the openAutomation

%The rest are handled by a person, as today, and don't count towards the benefit. Default 80%: in Ardent Partners' research, 19.9% of invoices end up as exceptions on average. Ardent Partners via Mediussource Ardent Partners via Medius (opens in a new tab)
′
€Average hourly labour cost in Greece in 2025, across industry, construction and services (excluding public administration), firms with 10+ staff. Put in your own. Eurostat, hourly labour cost in enterprises with 10+ staff (lc_lci_lev), 2025source Eurostat, hourly labour cost in enterprises with 10+ staff (lc_lci_lev), 2025 (opens in a new tab)
Build cost
€
€
€
hours a month100 hours a month
labour value a year
€21,840
net benefit a month
€1,420
payback
11–18 months

about 0.7 full-time roles (at 1,800 productive hours a year)

= Hours/month = documents × % untouched × minutes ÷ 60 · Value/year = hours × 12 × hourly cost · Payback = build cost ÷ (value/month − running cost)

1,500 × 80% × 5 min ÷ 60 = 100 hours/month · 100 × 12 × €18.20 = €21,840/year · €1,820 − €400 = €1,420/month net · €15,000–25,000 ÷ €1,420 = 11 to 18 months' payback

100 hours a month is about two-thirds of a full-time role. Nobody is surplus: you send the hours where you're short of people.

An example of the calculation. We count only the documents that go through with no human touch, and take review time off their minutes.

Put your own numbers into the Meter

Other sectors, same method

Six businesses. Six workflows that eat hours.

From the port of Volos to a tutoring centre with 300 students: the same question every time. What gets typed in that already exists digitally, and who approves at the end.

Use cases built on how Greek businesses actually work. Wherever there's a number, its source or its maths sits right beside it.

How this mechanism runs for a business in this sector, step by step.
Freight forwarder · Port of VolosBefore-and-after diagram: Freight forwarder · Port of Volos. Today: request by email, rates in Excel (25 min per request), quote typed in Word. With the automation: request by email, reads route, weight, pallets, rates from the ERP, sales lead checks and sends. We measure: quote turnaround. Use case. Today Freight forwarder · Port of Volos Request by email Rates in Excel 25 min per request Quote typed in Word With the automation Request by email Reads route,weight, pallets Rates fromthe ERP Sales lead checksand sends we measure quote turnaround use case
Today
The request lands in email. Someone opens the rates in Excel, looks up the fuel surcharge and types the quote in Word. The customer waits. Or doesn't, and goes elsewhere.
With the automation
The assistant reads the request (origin, destination, weight, pallets, ADR), pulls the rates from the ERP and drafts the quote.
The human
The sales lead checks it and presses send. Prices always come from the ERP, never from the model.
We measure
Quote turnaround · quotes per day · win rate
The maths
10 requests × 25 min × 22 working days ≈ 92 hours a month. About 75 of them come back. Netex estimate from the SME automation research (track 03, §4)
What others measured
At a US logistics firm, quote preparation fell from 30–40 minutes to 3 or less. Toptal, US logistics casesource Toptal, US logistics case (opens in a new tab)
We have something like this

Connects to Logistics & distribution

How this mechanism runs for a business in this sector, step by step.
Accounting office · HeraklionBefore-and-after diagram: Accounting office · Heraklion. Today: Viber, email, myDATA, classified by hand (1 min × 6,000), posted line by line. With the automation: pulled from myDATA, VAT no., VAT, totals checked, suggested classification, accountant approves. We measure: % approved unchanged. Use case. Today Accounting office · Heraklion Viber, email, myDATA Classified by hand 1 min × 6,000 Posted line by line With the automation Pulled frommyDATA VAT no., VAT,totals checked Suggestedclassification Accountantapproves we measure % approvedunchanged use case
Today
Client expenses arrive from myDATA, email and Viber. Each one is classified and posted by hand.
With the automation
Expenses come straight from myDATA, get checked (VAT number, VAT, totals, duplicates) and receive a suggested classification with a confidence score, based on that client's history.
The human
The accountant approves. Anything low-confidence goes to them, not to myDATA.
We measure
Classifications approved unchanged · corrections · days to close the month
The maths
200 clients × 30 documents × 1 min ≈ 100 hours a month. With 60–70% classified automatically at high confidence, 60–70 hours come back, plus about 12 from chasing paperwork. Netex estimate from the SME automation research (track 03, §4)
We have something like this

Connects to Accounting & law firms

How this mechanism runs for a business in this sector, step by step.
Law firm · 5 lawyersBefore-and-after diagram: Law firm · 5 lawyers. Today: a 40-line email, filing, deadlines typed by hand (15 min per enquiry). With the automation: a 40-line email, matter type, parties, deadlines, draft calendar entry, a lawyer confirms. We measure: time to first reply. Use case. Today Law firm · 5 lawyers A 40-line email Filing Deadlines typedby hand 15 min per enquiry With the automation A 40-line email Matter type,parties, deadlines Draft calendarentry A lawyer confirms we measure time to first reply use case
Today
The office manager reads it, files it and types the deadlines into the calendar. One wrong digit in a date costs more than everything else put together.
With the automation
The assistant recognises the type of matter, extracts parties, dates and deadlines, and drafts the calendar entry. A private search assistant looks through your own past documents, respecting the firm's access rights.
The human
A lawyer confirms every deadline. The assistant gives legal advice to no one.
We measure
Time to first reply · deadline errors · hours to first draft
The maths
150 enquiries × 15 min × 65% ≈ 25 hours a month, on new-matter intake alone. Netex estimate from the SME automation research (track 03, §4)

Documents stay on EU infrastructure or on your own server.

We have something like this

Connects to Accounting & law firms

How this mechanism runs for a business in this sector, step by step.
Fashion e-shop · WooCommerceBefore-and-after diagram: Fashion e-shop · WooCommerce. Today: product sheet, descriptions one by one (12 min per SKU), COD payouts in Excel. With the automation: product sheet, draft EL/EN and alt text, marketing approves, COD payouts matched. We measure: days to go live. Use case. Today Fashion e-shop · WooCommerce Product sheet Descriptions oneby one 12 min per SKU COD payouts in Excel With the automation Product sheet Draft EL/ENand alt text Marketingapproves COD payoutsmatched we measure days to go live use case
Today
Descriptions are written one at a time and the translations wait. Courier cash-on-delivery statements are checked line by line in Excel.
With the automation
The product sheet and photos produce a draft description in Greek and English, with alt text. COD payouts are matched to orders automatically.
The human
Marketing approves every description before it goes live. Accounts only sees the payouts that don't match.
We measure
Minutes per SKU · days to go live · unmatched COD payouts per month
The maths
300 SKUs × 12 min × 75% ≈ 45 hours on one season's descriptions alone. Netex estimate from the SME automation research (track 03, §4)
We have something like this

Connects to E-shops & retail

How this mechanism runs for a business in this sector, step by step.
Steel fabricator · Almyros industrial areaBefore-and-after diagram: Steel fabricator · Almyros industrial area. Today: 3.1 certificate, typed into Excel (never measured), quality file. With the automation: 3.1 certificate, reads heat no., grade, chemistry, checked against spec limits, deviations: quality decides. We measure: minutes per certificate. Use case. Today Steel fabricator · Almyros industrial area 3.1 certificate Typed into Excel never measured Quality file With the automation 3.1 certificate Reads heat no.,grade, chemistry Checked againstspec limits Deviations: qualitydecides we measure minutesper certificate use case
Today
The quality lead types the heat number, steel grade and chemistry from the EN 10204 3.1 certificate, matches it to the delivery note and files it. When a customer asks for traceability, the search starts from scratch.
With the automation
The certificate is read at goods-in. Values are checked against the specification limits and linked to the batch in the ERP.
The human
Anything out of spec is accepted or rejected by the quality lead.
We measure
Minutes per certificate · traceability lookup time · deviations caught at goods-in
The maths
We'll put a number on it only after timing your own goods-in. On the diagram the step reads “never measured”: nobody has ever timed it.
We have something like this

Connects to Metals & steel

How this mechanism runs for a business in this sector, step by step.
Tutoring centre · 300 studentsBefore-and-after diagram: Tutoring centre · 300 students. Today: attendance and scores, written from scratch (300 parents), message to the parent. With the automation: attendance and scores, draft per student, the teacher approves, Viber to the parent. We measure: teacher hours a month. Use case. Today Tutoring centre · 300 students Attendanceand scores Written from scratch 300 parents Messageto the parent With the automation Attendanceand scores Draft per student The teacherapproves Viberto the parent we measure teacher hoursa month use case
Today
Attendance, test scores and comments live in three different places. Every update is written from scratch.
With the automation
The assistant gathers the attendance and scores the teachers have already entered and drafts a note for each student. The update reaches the parent on Viber.
The human
The teacher edits and approves. The AI neither grades nor assesses any student.
We measure
Teacher hours a month on updates · updates sent on time · parent replies
The maths
30–50 hours a month for three automations together: parent updates, August–September enrolments and exercise variants. Netex estimate from the SME automation research (track 03, §4)

The AI Act classes AI that evaluates learning outcomes as high-risk. That's why, here, the AI only drafts. Jones Walker, AI Act: Article 50 and the Digital Omnibussource Jones Walker, AI Act: Article 50 and the Digital Omnibus (opens in a new tab)

We have something like this

Connects to Sectors

The form opens with your sector selected. Nothing goes into the page address.

What we measure

If it isn't measured, it didn't happen.

  • Minutes per document, quote or order, before and after
  • Share completed with no human touch
  • Corrections per 100 entries
  • Response time to customers or suppliers
  • Hours returned, and where they went

  1. We measure today first. No baseline, no “before and after”.

    applies

  2. We count only what finished with no human touch, and subtract review time.

    applies

  3. We give a range, not one impressive number.

    applies

  4. We say “hours redeployed”, not “jobs cut”.

    applies

  5. Costs saved and revenue recovered are counted separately.

    applies

= Hours/month = volume × minutes per task × share with no human touch ÷ 60 − review time

The PATH method, step by step

What the measurements say

The numbers that aren't ours, with sources.

About 95% of generative-AI pilots show no measurable impact on the P&L. That's why pilot targets are signed before any code is written.

Become our first measured case in your sector.

We'd like to be first

Systems

We speak your systems' language.

System names, not partner logos. We don't claim partnerships we haven't signed.

  • SoftOneWeb Services (getBrowserData, setData)
  • Entersoftvia API or a partner
  • Epsilon (PYLON, Galaxy)via a partner
  • myDATAREST API v2.0.0
  • E-invoicing providersprovider API
  • WooCommerceREST API, webhooks
  • Gmail and Microsoft 365mail APIs
  • Viberbusiness messages
Every integration, system by systemSoftOne, Web Services · AADE, myDATA REST API v2.0.0SoftOne, Web Services · AADE, myDATA REST API v2.0.0source SoftOne, Web Services (opens in a new tab)source AADE, myDATA REST API v2.0.0 (opens in a new tab)

BundleAutomation

Back-Office Autopilot

Three automations, the connection to your systems, monitoring and an SLA.

  1. Three automations from the X-Ray, for example invoices → ERP, quotes, the management report
  2. Connection to the ERP, myDATA, email and Viber
  3. Monitoring: how many went through untouched, how many were corrected, where they got stuck
  4. An SLA and a quarterly value report with an “hours returned” counter
  5. AI training for your team (AI Act Article 4)

Price: after the X-Ray, once we know what's worth building.

A pilot on one workflow from €6,000, 4–8 weeks, with targets signed before it starts.

"From" prices, excluding VAT.

Start with the X-Ray

Questions

What can AI automate in a small or mid-sized business?

Anything repetitive that already exists digitally: posting invoices and delivery notes to the ERP and myDATA, quotes from email, orders from Viber or the phone, inbox triage and the weekly report. We start with the workflow that eats the most hours at the lowest risk, and a person approves the exceptions.

How much does an AI automation cost?

We start with a Workflow X-Ray, from €890, credited in full against a pilot. A pilot on one workflow starts from €6,000, runs 4 to 8 weeks and has targets signed before any code is written. We price the full rollout after that, with the payback worked out.

Will it replace staff?

No. It removes the copying, not the people. The hours it frees go to work that needs judgement: exceptions, customers, sales. In Greece, 69% of firms struggle to find skilled staff. Most teams are short of hands, not over-supplied. We measure the hours gained and show you where they went.

Does it work with our ERP (SoftOne, Entersoft, Epsilon)?

With most Greek ERPs, yes. We connect to SoftOne through its official Web Services and to myDATA through AADE's REST API. For Entersoft and Epsilon (PYLON, Galaxy) we work via API or a partner, depending on the installation. We check your setup in the X-Ray before promising anything.

What happens when the AI gets it wrong?

The system is built to expect it. Every field has a confidence score and every number is checked by a rule: VAT-number check digit, VAT, totals, duplicates. Anything below the threshold you set goes to a person. Everything is logged and corrections are counted every month.

Where does our data go?

Processing in the EU, a data processing agreement (DPA), no model training on your data and a log of every action. We keep the minimum data possible, and retention and deletion terms are written into the contract. All the limits are published in our AI Trust Standard. AI Trust Standard

How soon will we see results?

The X-Ray takes one to two weeks and ends with a demo on your own documents. The pilot runs 4 to 8 weeks, with a weekly measurement board and a go or no-go decision at the end. We work out the payback case by case, with the maths in view.

What is your team typing in that already exists digitally?

In one to two weeks you have a map of hours and euros, the top five opportunities, and a demo on your own documents.

Book a Workflow X-RayOr start with a free 45-minute call
From €890 · credited in full against a pilotThe form opens with whatever you picked on this page. Nothing goes into the page address.