Steffen Holzbau S.A. is a leading construction company based in Grevenmacher, Luxembourg. Known for its commitment to quality, sustainability and a human-centric approach, the company is now leveraging artificial intelligence to transform its tender-response process. This positions Steffen Holzbau S.A. as a forward-thinking leader in both construction and responsible innovation.
Photo: the team this story is about
At a glance
95%of time saved
≈43%fewer errors in material and labor
100%adoption from technical staff
Efficiency through automationOne of the most measurable improvements has been in tender evaluation. The time from receiving a tender to delivering a first draft has been reduced from six weeks to just two days.
Early finding of discrepanciesBy reducing the time spent on tender analysis, teams can focus on what matters most, thoroughly reviewing critical areas like fire safety and identifying potential discrepancies early.
Employee relief and focusA key outcome of the initiative is measurable employee relief. By reducing manual workload in the tender evaluation process, team members have gained back significant capacity.
01 · The challenge
Manual processes slow down growth
Preparing responses to tenders is a complex, labor-intensive process for Steffen Holzbau S.A. The workflow required:
Manual translation between French tender documents and the German ERP system
Cross-referencing tender requirements with product databases
Estimating materials, labor and costs, with a risk of human error
Compiling draft offers from scratch
This manual approach is time-consuming, error-prone and inconsistent, leading to delayed responses and potential financial risks due to miscalculation.
02 · What we built
A four-phase AI proof of concept (PoC)
The solution follows four phases. First, a multilingual NLP model extracts key data from French and German tender documents, structuring them for automated processing. An AI system then maps these requirements to the ERP database and previous tenders. Next, a predictive model estimates materials, labor and costs based on historical data. Finally, AI generates draft offers, which experts review and refine through a continuous feedback loop.
Connected sources
ERPProduct databases
Phase 1 · Natural language processing (NLP) information extractionA multilingual NLP model extracts key data from French and German tenders, including material specifications, deadlines and compliance criteria. The goal is to structure unstructured tender documents for automated processing.
Phase 2 · AI-enhanced ERP matchingAn AI-driven hybrid system maps extracted tender requirements to the German ERP database and previous successful tenders, ensuring accurate material and labor data retrieval. This phase validates compatibility with existing infrastructure.
Phase 3 · Predictive estimation modelBy analyzing historical tender data, AI predicts material quantities, labor needs and costs based on previous tenders. Human feedback refines accuracy, reducing miscalculations and improving reliability.
Phase 4 · Partially automated offer generationAI generates initial draft offers based on structured data. Experts review and refine these drafts, creating a feedback loop to continuously improve AI performance.
03 · The rollout
From the proof of concept to the whole company
ResultsThe PoC achieved a groundbreaking outcome: the valuation of a tender was reduced from 6 weeks until the first draft to just 2 days. This transformation delivers measurable efficiency gains and a scalable foundation for innovation, positioning Steffen Holzbau S.A. as a pioneer in intelligent automation.
ConclusionThe PoC demonstrates that AI can fundamentally transform Steffen Holzbau S.A.’s tender-response process, optimizing operations and reinforcing its role as a leader in construction.
Next stepsThe successful tender-response automation PoC has energized the Steffen Holzbau team, leading to an expanded partnership and the deployment of eokit, a model-agnostic, intuitive AI chat designed for everyday use across the entire organization. eokit integrates automation and agents, and ensures GDPR and EU AI Act compliance through a custom AI gateway developed by aigentcy. This gateway prevents any internal information from being shared with non-EU providers, aligning with Steffen Holzbau’s commitment to data security and responsible innovation.
Deploymenteokit has been successfully deployed to all 160 employees with strong adoption and highly positive feedback across the organization.
04 · The results
Key performance indicators and expected outcomes
Task
Before
With eokit
Tender evaluation, from receiving a tender to a first draft
6 weeks
2 days
Why this matters
By automating and streamlining tender-response processes, Steffen Holzbau S.A. aims to:
Increase competitiveness with optimized, more accurate bids.
Enhance profitability by reducing errors and optimizing resources.
Strengthen its reputation as an innovative, tech-forward leader in Luxembourg’s construction sector.
“When it comes to AI solutions, we could not be working with a better team than aigentcy, and we can not wait to have them work on a number of other valuable opportunities.” Stephan Hostert · CEO, Steffen Holzbau S.A.
KB
ContactKarina Bures Operations · aigentcy S.A.
FAQ
Asking the right questions
FAQ ›Compliance
5 questions
eokit is designed to be natively GDPR-compliant. It runs in the EU or inside your own network, and prompts are checked for personal data before they can go to an external model. We provide a data processing agreement and a list of subprocessors.
Where you choose to run eokit: in your own data centre, in our EU cloud in Germany or Luxembourg, or in your own cloud account. It is not copied to other regions.
Yes. You describe once how your company writes, and drafts follow that tone. You can also add your logo and colours.
Exactly the people who can open the document today. eokit uses the permissions from your directory and file systems, so people get answers only from documents they are allowed to open.
Your documents stay where they are, because eokit does not move them. We remove our components and can hand over your activity logs.
A digital team member with one job. It knows which systems to read, which steps to follow and what to hand back, for example a report or a draft.
Yes. Someone who knows the task describes it in plain words, eokit sets up the agent, and you check it before it runs.
The one you choose: a self-hosted model, a model from an EU provider or a large commercial model. Each department can use a different one. Access rights stay the same.
Reading and drafting happen automatically. Sending, booking or changing a record needs a person's approval, unless you decide otherwise. Every step is logged.
Workflows take on recurring tasks with a clear result, for example writing weekly reports, preparing reminders, summarizing meetings or comparing quotes. Your team decides how the agent works and what a good result looks like.
A workflow starts on a schedule, when something changes in a connected system, or when someone asks in the chat. You choose the trigger that fits the task.
Yes. You can run workflows outside working hours, for example for large monthly reports. Actions that need approval still wait for the person in charge.
The person in charge is told about the error and its cause. The log shows which steps ran and where the run stopped.
Every run is logged. You see which information the agent used and what came out. Your team checks whether the result meets its guidelines. You decide which actions also need approval.
eokit has connectors for SharePoint, Google Drive, Outlook and Gmail. Other tools connect through MCP servers. Which triggers and actions are available depends on the connector and on existing access rights.
Nothing matches that yet. Ask us directly and we add it here.
eokit is designed to be natively GDPR-compliant. It runs in the EU or inside your own network, and prompts are checked for personal data before they can go to an external model. We provide a data processing agreement and a list of subprocessors.
Owner Data protectionApplies to EU cloud · on-premise
Where you choose to run eokit: in your own data centre, in our EU cloud in Germany or Luxembourg, or in your own cloud account. It is not copied to other regions.
Yes. You describe once how your company writes, and drafts follow that tone. You can also add your logo and colours.
Owner MarketingApplies to All plans
3 of 5
Compliance
Who can see which document?
Exactly the people who can open the document today. eokit uses the permissions from your directory and file systems, so people get answers only from documents they are allowed to open.
Owner IT securityApplies to SharePoint · Google Drive
4 of 5
Compliance
What happens when we cancel?
Your documents stay where they are, because eokit does not move them. We remove our components and can hand over your activity logs.
Owner LegalApplies to All contracts
5 of 5
Agents
What is an agent in eokit?
A digital team member with one job. It knows which systems to read, which steps to follow and what to hand back, for example a report or a draft.
Owner ProductApplies to All plans
1 of 4
Agents
Can we build our own agents?
Yes. Someone who knows the task describes it in plain words, eokit sets up the agent, and you check it before it runs.
Owner ProductApplies to Business plan and up
2 of 4
Agents
Which models do agents use?
The one you choose: a self-hosted model, a model from an EU provider or a large commercial model. Each department can use a different one. Access rights stay the same.
Owner ITApplies to Per workspace
3 of 4
Agents
Can an agent act without approval?
Reading and drafting happen automatically. Sending, booking or changing a record needs a person's approval, unless you decide otherwise. Every step is logged.
Owner GovernanceApplies to All agents
4 of 4
Workflows
What tasks can workflows take on?
Workflows take on recurring tasks with a clear result, for example writing weekly reports, preparing reminders, summarizing meetings or comparing quotes. Your team decides how the agent works and what a good result looks like.
Owner ProductApplies to All plans
1 of 6
Workflows
How does a workflow start?
A workflow starts on a schedule, when something changes in a connected system, or when someone asks in the chat. You choose the trigger that fits the task.
Owner ProductApplies to All plans
2 of 6
Workflows
Can workflows run at night?
Yes. You can run workflows outside working hours, for example for large monthly reports. Actions that need approval still wait for the person in charge.
Owner IT operationsApplies to All deployments
3 of 6
Workflows
What happens when a workflow fails?
The person in charge is told about the error and its cause. The log shows which steps ran and where the run stopped.
Owner IT operationsApplies to All workflows
4 of 6
Workflows
How do we check a workflow's results?
Every run is logged. You see which information the agent used and what came out. Your team checks whether the result meets its guidelines. You decide which actions also need approval.
Owner GovernanceApplies to All workflows
5 of 6
Workflows
Which systems can workflows connect to?
eokit has connectors for SharePoint, Google Drive, Outlook and Gmail. Other tools connect through MCP servers. Which triggers and actions are available depends on the connector and on existing access rights.
Owner IntegrationsApplies to All plans
6 of 6
EU hosting or on-premise No training on your data
Adopt AI and boost your productivity while keeping data secure
