Icelandic workplace software company 50skills has raised $6 million to expand an HR automation platform that combines fixed workflows with AI agents, as organisations start putting artificial intelligence into processes involving contracts, employee records, background checks, approvals, and workplace administration.

The round was co-led by Frumtak Ventures, Swiss Post Ventures, Avant, and DRD Investments. 50skills will use the funding across sales, marketing, community growth, and continued development of a platform already used by employers across Iceland, the Nordics, and the UK.

The company is concentrating on administrative processes around employees rather than another standalone HR chatbot. Teams can describe a workflow in plain language, after which deterministic automation handles predictable steps, AI agents interpret information where necessary, and people retain control over decisions requiring human judgement.

That distinction becomes important when workplace AI moves from drafting documents into systems that affect somebody’s employment, pay, access, or personal information.

HR automation moves beyond the assistant

50skills says millions of actions have run through its platform across onboarding, contracts and signatures, employee requests, background checks, equipment, system access, and related coordination.

Icelandair has digitised more than 20 HR workflows using the product and reports eliminating the equivalent of four to five months of administrative work each year. The airline also uses the platform to review more than 3,000 employee health-grant receipts annually against company policy.

Air Atlanta reports saving 50 hours a week through visa-processing automation, while Íslandshótel says it cut the time spent processing some employee-benefit applications by 80%. Those figures come from customers and the vendor rather than an independent productivity study, but they show the type of operational measures against which HR automation can be assessed.

Many of these processes are repetitive without being completely predictable. Missing documentation, unusual requests, inconsistent forms, and exceptions can force rigid workflow software into manual queues, giving language models a potential role in interpreting the variation before returning the task to a controlled process.

Judgement still needs an owner

HR is also one of the least forgiving places to automate carelessly. Employment records can include identity information, compensation, health-related documents, performance material, contractual data, and other information subject to privacy and employment obligations.

50skills is therefore separating work that should always follow fixed rules from steps where AI interpretation may help, while preserving human oversight at points where the organisation does not want a probabilistic system making the final decision.

The model is likely to become common across enterprise automation. Models supply interpretation, workflow engines provide structure, integrations connect applications, and people intervene where financial, legal, employment, or security consequences make full autonomy inappropriate.

That architecture also gives organisations a more measurable route into AI than redesigning a department around a general-purpose agent. A single onboarding, visa, or document-validation process can be compared with the time and error rate of the work it replaces before the approach is extended elsewhere.

The governance challenge grows as those workflows accumulate. Once AI is operating across dozens of employee processes, companies need consistent policies for permissions, data access, model behaviour, record keeping, and escalation rather than treating every automation as a separate experiment.

HR software has spent years becoming the formal system of record for employee information while much of the activity around those records continued through email, spreadsheets, and manual follow-up. Agentic automation is now being aimed at that connective work.

For 50skills, the commercial question is whether the combination of fixed rules, AI interpretation, and human oversight can remain manageable as customers push the system into more consequential processes. The productivity case is strongest where administrative work is repetitive; the governance case becomes more demanding each time software moves closer to decisions affecting the person behind the record.