Workflow automation: eight examples from Icelandic companies

Research shows 25 to 56% faster completion of individual tasks with AI (Harvard/BCG 2023). But the most interesting cases are not the tasks that get faster. They are the workflows that disappear entirely: nobody sits over them any more, they simply happen.

Four characteristics of a workflow that can be automated

Before we look at the examples, it helps to know what characterises a workflow that is suited to automation. Having built more than 50 AI solutions since 2025, we see the same pattern again and again. The best workflows have four characteristics:

• Repetition. The task is done daily or weekly, not once a year.
• Clear input and output. It is obvious what goes in (an email, a document, a spreadsheet) and what should come out (a reply, a summary, an entry).
• Evaluation rather than judgement. The task requires reading, sorting and rephrasing, not making major strategic decisions.
• A person can review the result in seconds. Approving a draft takes a fraction of the time it takes to write one from scratch.

If a workflow meets three or four of these, it is a candidate. If it meets one or none, leave it, at least for now.

Eight examples, department by department

Here are the eight workflows we see most often in Icelandic companies. None of them requires coding skills from the staff who use the solution.

Customer service. workflow: Email sorting and draft replies · what changes: Emails are pre-sorted and draft replies are waiting. An employee reviews and sends.

Finance. workflow: Draft monthly reports · what changes: The data is pulled and a first draft of the text and analysis is written automatically.

Management. workflow: Meeting minutes into action lists · what changes: A meeting recording becomes minutes with owners and deadlines.

Accounting. workflow: Matching invoices to purchase orders · what changes: Incoming invoices are read and matched. Only exceptions land on a person's desk.

HR. workflow: Job adverts and application screening · what changes: The advert is written from the job description, and applications are summarised into a comparable overview.

Sales and proposals. workflow: Draft tender and proposal documents · what changes: Previous proposals and the requirements brief become a first draft in an hour instead of days.

Sales. workflow: CRM clean-up and logging · what changes: Interactions are logged automatically against the right customer, and duplicates and outdated data are flagged.

Development. workflow: First-pass code review · what changes: AI reviews changes before a person does and catches common errors.

Notice the pattern: in every example, a person has the final say. The AI delivers a draft, a sorting or a match. The employee approves, corrects or rejects it. That way, responsibility stays where it belongs and errors do not slip through unseen.

Some workflows do not get faster, they disappear

There is a difference between two kinds of results. On the one hand, tasks that become 25 to 56% faster (Harvard/BCG 2023): the employee still does them, just faster. On the other hand, workflows that disappear: nobody "does" invoice matching any more, the system matches and a person only looks at the exceptions.

In the examples above, meeting minutes, invoice matching and CRM logging usually fall into the second category. These are the tasks nobody misses. They added no value, they were simply necessary. When they disappear, time is freed up for the work that genuinely requires judgement: customer relationships, analysis, decisions.

This is also why we never promise "10x for everyone". The honest picture is this: a measurable speed-up on individual tasks and a handful of workflows that disappear entirely. That is enough to change how a department runs, without exaggerating anything.

So why do most companies still fail at this?

95% of corporate AI projects deliver no measurable results (MIT NANDA). The reason is rarely the technology. It is the order in which things are done. Companies buy tools, send an email to staff and hope for the best.

What works is the reverse order: start with the workflows, not the tools. Map where the time goes, pick two or three workflows with the four characteristics, build the solution with the people who own the workflow, and measure before and after. When the team sees its own numbers, not statistics from foreign reports, attitudes change.

Staff training is the other half. Automation without training becomes a black box that nobody trusts or maintains. That is why we weave the two together in the AI Sprint: the teams choose the workflows, we build the solutions together, and in-house specialists keep the development going.