Data analysis and machine learning in Gothenburg
Fiive builds forecasts, risk models and decision support that go into production — from data preparation to a model your systems actually use. We start with a decision worth improving.
Gothenburg
Local team at Järntorget
Pilot first
Measurements before a big investment
In production
Models in operation, not in a report
Handover-ready
Models, code and documentation
You do not need a data warehouse in place. What you do need is a decision that is made often, a measurable outcome and someone who can act on the answer.
The same judgement is made over and over in Excel or in the head of an experienced colleague. You want the answer to be there when the decision is made, not the week after.
Forecasts and decision support in your flows
The right customer, the right case or the right maintenance job first. You need a model that ranks what is worth acting on while there is still time.
Risk scoring and prioritised action lists
You want to know whether the data holds up and whether the value is there — without first committing to a platform, a data warehouse or a long programme.
A defined pilot with measurements
Less suitable when a simple rule or a report already solves the need.
Case study · Machinery manufacturer
We built a RAG solution that lets service technicians ask the service manuals directly, with a reference to the right section. The setup was a defined proof of concept — prove the method worked on their actual manuals, at a known cost. After the pilot the client took over and developed the solution further internally.
One contact and one responsibility from raw data to a model in production. We do not hand over a presentation — we hand over something that gets used in your processes.
01
We start with the data you have. It is cleaned, harmonised and connected until the basis is stable enough to make decisions on — and we tell you if it is not enough.
02
Forecasting, prioritisation and classification where the same kind of decision is made often. We choose the simplest method that solves the problem — not the most advanced.
03
A model nobody maintains loses value quietly. We put the solution into production, connect it to your workflows and track accuracy over time.
How an engagement works
Describe a recurring decision or flow. Together we assess what data exists, whether a pilot is the right next step and what it needs to show — before you commit to anything bigger.
Map a decisionWe start from a recurring decision, review the data behind it and assess whether machine learning is the right tool.
You get: an assessment of the data, a recommended scope and a price
We build a first model and measure how it performs on your actual data — before you invest in putting it into production.
You get: a working model and measurements on the business value
We connect the solution to your systems so the answers reach the right person at the right moment, and retrain when reality shifts.
You get: a live solution, documentation and a plan forward
Technology you can own
We build with established tools — Python, SQL and common machine learning frameworks — and run in the cloud environment you already have, such as Azure, AWS or Google Cloud. We choose the simplest method that solves the problem and document why, so someone else can retrain the model after us.
The value of data analysis is decided by the sector knowledge behind the models. In these three sectors, we have it.
We work with companies across industries. Read our previous projects and what outcomes the solutions delivered in practice.
The key things to sort out before you choose a partner, a delivery model and a first scope.
It depends on how much data preparation is required, how many systems need connecting and whether the solution goes into production or is only evaluated. We start with a defined discovery phase that sets a fixed price for a pilot — you know the cost before the model building begins. Our pricing guide for software development gives a first impression of the levels.
We aim for a defined pilot on your real data rather than a long investigation project. The pilot should produce measurements on a concrete decision, so you can judge the value before paying to expand the solution and put it into production.
No. We need history for what you want to predict and something that links the records together — not a finished data warehouse. The first step is to look at the data you actually have and say outright whether it is enough. If it is not, data preparation is part of the engagement, not a prerequisite you have to solve yourself first.
Yes, that works fine. Often it is a model that worked on delivery but has lost accuracy, or that nobody knows how to retrain any more. We map what it actually does, measure how it performs today and take over operations and retraining.
You own the code, the models, the environments and your data. We document key decisions so your own team or another partner can retrain and keep building. Maintenance is a contract you can terminate — no lock-in.
Fixed price when the delivery can be clearly defined — which discovery and pilot phases usually can. For ongoing maintenance and retraining we work with a transparent, prioritised backlog. After the first mapping we propose the model that gives you the most control.
Practical articles on document search, NLP and applied machine learning.
How RAG and graph databases are revolutionizing document search in large enterprises, and what it means for building smarter knowledge retrieval systems.
Read more about Advanced Document Search: Harnessing AI and Graph Databases
Delve into data augmentation techniques, including back translation, to enhance NER systems. Learn about the challenges and solutions associated with NER.
Read more about Data augmentation for NER using Back Translation
How BERT models improve Named Entity Recognition in healthcare — extracting diagnoses, medications, and clinical entities from patient records with greater accuracy.
Read more about Named Entity Recognition in Healthcare⚡ We answer within 24h!
