AI engagements, as they are

Logistics

A logistics company in Gothenburg receives requests from its clients by email and works across some eight different systems. The company has no IT department of its own.

• A new freight booking platform that connects the transport management system, a booking portal, the shared inbox and EDI flows.
• Language models read client emails, turn them into structured freight requests and suggest a carrier based on previous shipments.
• AI drafts every outgoing email in the recipient's language.

A person approves every message before it is sent.

Insurance

A Nordic insurance company wanted to try a faster way to reach a new market.

• A proof of concept for the Norwegian market: the policyholder reports glass damage on their phone and photographs it in the app.
• The app runs on iOS and Android and was built with AI tools end to end.

Manufacturing

A manufacturing company collects data points from its equipment and needs alarms from those readings.

• AI measures the data points.
• Alarms are raised from those measurements.

Private models

Can a small model, fine-tuned on one task, beat a large general-purpose model that only gets instructions? The use case: answering customer support tickets.

• A small model, fine-tuned on growing amounts of support-ticket data.
• A synthetic test set, held out from training, that scores both models the same way.

• 3.5×3.5×

  better reply quality

• ~16×~16×

  better key fields

• ~50×~50×

  faster replies

• ~125x~125x

  estimated lower running cost per ticket (internal estimate)

Results on a support-ticket task. Your results depend on your data.

Software

The work of building software should run as autonomous workflows.

• A software factory that runs those workflows.
• The factory carries out the build, tests it, reviews it, deploys it and verifies it.

A person only guides the product.