95% of corporate AI projects deliver no measurable results (MIT NANDA). Not because the technology does not work, but because the implementation fails. The same five mistakes come up again and again, and they are all predictable.
Mistake 1: subscriptions bought, training skipped
The most common start: the company buys ChatGPT or Copilot subscriptions for everyone and hopes for the best. Three months later few people use the tools to any real extent, and those who do use them the same way they did on day one: type a question, get a mediocre answer, give up.
Tools without training are not half an implementation, they are no implementation at all. Research shows that proper use delivers 25 to 56% faster completion of tasks (Harvard/BCG 2023), but untrained use can actually make experienced people slower and more confident in the wrong answer. The difference between the two is not the tool. It is the training.
In Iceland the picture is clear: 80% of professionals use AI at work but only 34% have received training from their employer (Viska 2025). The gap between those two numbers is exactly where the value is lost.
Mistake 2: no policy on safe use
Only 14% of Icelandic companies have an AI policy in place (Statistics Iceland 2025). At the other 86%, AI is still being used every day. Staff paste business data, contracts and personal data into free tools without knowing whether they are allowed to.
The lack of a policy does double damage. On the one hand, a security risk: data goes into tools where it can end up in model training. On the other, hesitation: conscientious staff do not dare to use the technology because nobody has told them what is acceptable. Both problems are solved by the same document: clear rules about which data may go where, which tools are approved and how to verify results.
Mistake 3: only the enthusiasts get trained
When training is finally offered, the people who turn up are the ones who would have managed on their own. The tech enthusiasts sit in the front row, the rest carry on as before. The result: two departments in the same company, one many times more productive than the other.
This is priorities in reverse. Research shows that the biggest gains from AI are not among the most experienced and tech-savvy people, but among the broad majority: those with the least experience improved by 43% (Harvard/BCG 2023). Training that only reaches the enthusiasts leaves the biggest gains on the table.
Mistake 4: nobody measures anything
Ask a manager who bought an AI course last year: what did it deliver? The most common answer is silence. No baseline was measured before the training, nothing was measured after it, and so it is impossible to know whether the investment paid off.
Without measurement, AI adoption is a matter of faith. With measurement, it is a business decision. The minimum is simple: pick three to five workflows, measure how long they take today, and measure again 30 days after training. Same people, same tasks, your own numbers.
Mistake 5: one workshop and then nothing
The fifth trap is the lack of follow-up. The company holds a great workshop, everyone goes home excited, and four weeks later everything has slipped back into the old routine. Skills that are not used and supported disappear within a few weeks.
What works is a system, not an event. The difference is clearest in comparison:
A lecture about the possibilities. systematic adoption: Hands-on practice on your own tasks
Generic examples. systematic adoption: Use cases in every job
Everyone goes home with ideas. systematic adoption: Everyone goes home with working workflows
Nobody responsible afterwards. systematic adoption: In-house specialists keep the development going
Results unmeasured. systematic adoption: A baseline and a repeat measurement on day 30
Companies that succeed at adoption have this in common: training for everyone, clear rules, practice on real tasks, designated people who keep the ball rolling, and measurement that shows in black and white what changed.
What you can do right now
- Do a health check on your own company. Go through the list above and tick off: have we trained everyone, not just some? Do we have a written policy? Do we measure anything? Three or more noes means your subscriptions are probably delivering little.
- Write a one-page set of interim rules. Three points are enough to start: no personal or confidential data in free tools, use company accounts where they exist, verify every fact before anything leaves the building.
- Pick three workflows and measure them. Document how long they take today. Then you have a baseline, and whatever training you choose later can be evaluated with real numbers instead of gut feeling.
[ Get in touch ]
Book a free assessment
90 minutes that pay off immediately: we map your AI usage, risks and 3 to 5 automatable workflows, and deliver a report within a week. No commitment.