The hard part of AI on your ERP is the ERP.
Strategy firms will not touch your Soft1 database. Data science shops do not know what your tables mean. We built the customisations — we already have both the access and the context.
Four applications that make sense in a Greek mid-market company.
All of them assume an ERP with clean data. That is the first piece of work.
NATURAL-LANGUAGE REPORTING
"How much did we sell this customer last quarter and what do they owe us?" without waiting for someone to build a report. The question becomes a query over Soft1 data, respecting that user's permissions.
DOCUMENT EXTRACTION
Supplier invoices, as PDFs or on paper, read and posted into the ERP — with a person confirming rather than typing.
DEMAND & STOCK FORECASTING
Order suggestions from history, seasonality and supplier lead times, instead of the buyer's instinct.
ANOMALY DETECTION
Transactions outside the pattern: duplicates, prices off policy, unusual discounts. A control that runs continuously rather than once a year.
And where the ERP does not reach.
USER SUPPORT
Answers to repetitive internal requests, escalating to a person when the question is not routine.
INTERNAL KNOWLEDGE
Procedures, manuals and contracts made searchable by the people who need them.
TOOL INTEGRATIONS
Connecting models to real systems through MCP and APIs, so they can act rather than only answer.
What we promise here, and what we do not.
ERP, hosting and support are services we have run in production for twenty years. AI is our newest pillar, and we present it as capability and method rather than as a list of delivered projects.
What we bring is access to your data and an understanding of how the ERP underneath works — which is usually where AI projects stall. We do not sell pilots that never reach production.
If an idea has no measurable benefit, or the data does not support it, we will say so during the assessment — before you pay for a build.
Paid assessment, then pilot, then production.
Each phase has its own deliverable and its own go/no-go. You do not commit to the next before seeing the last.
-
01
Readiness assessment
2–3 weeks [VERIFY]
Which tasks are worth it, what state the data is in, and what GDPR and the EU AI Act require. Deliverable: a ranked list with benefit and cost estimates.
-
02
Pilot
4–8 weeks [VERIFY]
One use case, real data, and a success criterion agreed before it starts.
-
03
Production
Per project
Integration into the workflow, permissions, action logging and user training.
-
04
Operation
Ongoing
Monitoring output quality, cost and drift. A model in production needs maintenance like any other system.
The questions your auditor will ask.
GDPR & THE EU AI ACT
Where the data goes, which risk category the application falls into, and what documentation is required.
DATA QUALITY
Most failures are not technical. They are duplicate customer records and product codes with no rule behind them.
BUILD OR BUY
When an off-the-shelf tool is enough and when development is warranted. Usually off-the-shelf is enough.
Before you start an AI project.
Does our data leave to train someone's model?
No. We choose providers and settings where your data is not used for training, and we document that in writing. Where sensitivity demands it, we look at options that run inside your own infrastructure.
What happens when the model gets something wrong?
It will, eventually. That is why we design these applications so a person confirms before any action with financial consequence, and so every action is logged.
Have you delivered projects like this?
AI is our newest pillar and we do not present case studies that do not exist. What we do have is twenty years of access to customers’ ERP data and the engineering ability to connect it — which is exactly where most AI projects stall.
What does it cost?
The readiness assessment is a fixed-price engagement. Whatever follows is quoted once we know what is worth doing. [VERIFY pricing]
Which task eats the most hours?
Start there. If it is not worth automating, we will tell you.