Orders typed into the ERP. The same customer emails, every day. Contracts read line by line. GoldenDane builds AI that does the first pass inside the systems you already use, and proves its accuracy on your own data before anyone relies on it.
80 anchors ordered, 64 in stock. Split the delivery or offer the 41-0095 alternative?
↳ Stock: warehouse 1 · Customer terms: partial delivery allowed“Hi, my order hasn't arrived yet. If it comes after the weekend, can I still return it?”
Hi Anna, your order left our warehouse on 6 October and is out for delivery today. And yes: you can return it within 30 days of delivery, free of charge.
↳ Order system #10482 · Returns policy §211 more delivery-status emails drafted and waiting for a quick review.
Fatigue design for steel components per EN 1993-1-9, detail category ≥ 71.
↳ Specification B, p. 87 §4.3.2As-built BIM model due within 30 days. Conflicts with Appendix 2 (60 days).
↳ Contract, p. 12 · Appendix 2, p. 3GoldenDane ApS · CVR 46351789. Verify on virk.dk ↗
Data processing agreement signed before we see any data. EU processing where possible.
Pilots are priced up front. You decide what happens after you've seen the results.
We report accuracy on your own data before go-live, and keep measuring after.
These are the three kinds of work we automate most. Each runs inside the systems you already use, flags anything it isn't sure about, and links every answer to where it came from.
For order desks, wholesalers, brands and finance teams.
For customer service, webshops, travel, utilities and property managers.
For law firms, brokers, accountants, engineering and construction.
We analyse one export from your ERP, ticketing or case system and show you where work really waits, loops and piles up. No system access, fixed price.
Setpoint optimisation, soft sensors and early warnings for district heating, water, and process manufacturing. Tested in simulation on your history first.
You don't need to be technical, and you don't need new software. These are the teams we build for.
Orders and invoices entered automatically. Your team checks the exceptions.
"Where's my order" and returns answered in draft from your own data.
Key terms and deviations, each linked to the exact clause.
Quotes and policy wordings compared side by side, with sources.
Tenant emails, leases and move-out paperwork sorted and drafted.
Receipts and invoices extracted and checked, reminders drafted.
Requirements lists and conflict checks from tenders in minutes.
A Process X-ray shows where cases wait. Then we fix it.
Don't see your team? Tell us which task eats your week and we'll say honestly whether we can help.
Mette works at a 60-person engineering consultancy. A new tender lands every week, and each one runs to hundreds of pages. Here's her Monday before and after.
Mette is an illustrative example of the workflows we build for, not a client. Savings depend on your documents and team, which is exactly why we measure them in a short pilot before you commit to anything.
A Process X-ray is a two-week analysis of the activity log your business systems already keep. Like an X-ray, it shows what's inside without opening anything up: no new software and no access to your systems, just one data export.
case ID step timestamp person or teamHere's what an X-ray found in 4,580 real support tickets from a public dataset:
Tickets took 40 days to close but only 6 to fix. Most of the difference was a fixed confirmation window, not work. Public data from an Italian software company's IT help desk; findings are associations, not proof of cause.
Tickets picked up within a day were fixed in 1.2 days. The queue, not the work, made the difference.
Tickets solved by the first person took 1 day. Every hand-off multiplied the time.
A bottleneck in busy weeks, and a risk whenever they're away.
You never pay for a big project on faith. Every engagement starts with one workflow and a number we can measure.
You show us the workflow that eats your team's week. We ask how often it happens, who does it and which systems hold the data.
We build a working prototype on your real documents or process data and measure its accuracy against your team's own results.
We connect it to your systems, add access control, monitoring and approval steps, and train your team.
From the first call to measured results: about a month.
ChatGPT and Copilot are great for one-off tasks, and we use the same kind of models. The difference is everything around the model: where it runs, how it's tested, and who is responsible for it.
Every system is evaluated on a labelled sample of your own data, and you see the accuracy before it goes live.
Every extracted value and answer links back to the page, paragraph or record it came from.
AI takes the first pass. Your team approves, corrects and stays accountable, with a full audit trail.
Data handling agreed in writing before we start. Never used to train models for anyone else.
GoldenDane is a small, focused team from DTU, the Technical University of Denmark. The people you meet on the first call are the people who build and support your system.
Soheil leads client work: he maps your workflow, scopes the pilot and makes sure results are measured in hours and kroner, not buzzwords. Background in Human-Centered Artificial Intelligence at DTU.
Your first call and your main contact.
International Physics Olympiad medalist with a background in algorithms, applied mathematics and machine learning. Vadim designs and builds the systems, from document pipelines to optimisation models.
Builds and tests your system.
Something else? Write to hello@goldenda.com or call +45 50 39 84 24.
Yes, that's exactly who it's for. You don't need data scientists or new software. If your team repeats the same orders, emails, documents or process work every week, we can usually tell in one 30-minute call whether it's worth automating.
You walk us through one workflow that takes up your team's time. We ask a few questions about volume, systems and data, and tell you honestly whether AI can help, what a pilot would look like and what it would involve. No slides and no obligation.
A pilot is a fixed price agreed up front, scoped to one workflow and two to three weeks of work. We give you the number after the free call, once we know the size of the problem, and you decide whether to go ahead.
We sign a data processing agreement before we touch anything, keep processing inside the EU where possible, delete working copies when the project ends, and never use your data to train models for other clients.
For one person's one-off tasks, often yes. But a chat window doesn't read your inbox, enter orders in your ERP or run on its own, and nobody measures whether its answers are right. We build the system around the model: connected to your tools, tested on your data, covered by a data processing agreement, and maintained by us. If a chat assistant is all you need, we'll tell you on the first call.
A two-week, fixed-price analysis of the activity log one of your business systems already keeps. We call it an X-ray because it shows what happens inside a process without changing anything. You get a map of how cases really flow, where time is lost, who is overloaded, and the fixes worth making first. All we need is one export with four columns: case ID, step, timestamp and person or team. We never need access to your systems.
No. For industrial processes, the AI sits on top of your existing PID and SCADA control and suggests or sets better setpoints within limits you define. It's tested in simulation on your historical data first, and your operators can always override it.
Fair question. That's why we start with a small, fixed-price pilot and show measured results before you commit to more. You work directly with the two engineers who build your system, everything is covered by a written agreement, and you can verify our company registration on virk.dk.
Tell us in two or three sentences. Here's what happens next:
Prefer the phone? Call Soheil on +45 50 39 84 24
GoldenDane ApS · CVR 46351789 · Copenhagen, Denmark