Why ARK is built this way: the evidence
Four findings decided the shape of ARK: skill transfers only when it is used on real work, spacing beats cramming, retrieval beats re-reading, and most change programs fail without rigor at every stage. None of these are ARK's results. They are the reasons ARK looks the way it does.
TRANSFER. Baldwin and Ford's review of training research found that what happens in the classroom predicts very little on its own. Behavior changes when the skill is applied to real tasks and reinforced afterwards by the work environment. The training event is not the variable. The work is.
SPACING. Cepeda and colleagues pooled hundreds of experiments and found the same result each time: material revisited across days is retained far better than the same total time spent in one block. Cramming feels efficient in the moment and decays quickly.
RETRIEVAL. Roediger and Karpicke showed that recalling something from memory produces durable learning, while re-reading it produces confidence without retention. The version that feels harder is the version that lasts.
CHANGE. McKinsey's long-running work on transformations reports that roughly a third succeed. Where rigor is applied at every stage of the program, the reported success rate rises to about three quarters.
Read together, those four findings rule out most of what the market sells as AI upskilling. A course is a training event with no work attached, which fails the transfer test. A workshop is one block, which fails the spacing test. A slide deck to re-read fails the retrieval test. A launch with no cadence fails the change test.
What survives is unglamorous: short daily practice on the person's own tasks, the same ideas returning across weeks, answers produced from memory rather than recognized on a slide, and a measured record of what actually changed at work.
Practice on the real task, not a case study. Transfer is the only outcome that counts.
Spread the same concept across weeks instead of stacking it into one session.
Make people produce the answer before showing it to them.
Treat capability building as a program with a cadence and a measurement, not an announcement.
Pick one dreaded recurring task and rebuild it with AI this week. That single rebuild teaches more than a full course.
Schedule the return, not just the first session. A concept seen again on day 2, 4 and 8 is a concept you keep.
Before opening your notes, write down what you remember. The gap you find is the thing worth studying.
Record the before and after in minutes. Without a measurement, nobody can tell practice from theatre.
These findings shaped the design. They are not evidence about ARK's own results, and we do not present them as such.
Spacing and retrieval make learning feel harder in the moment. Teams reading comfort as progress will drift back to passive formats.
The McKinsey change figures come from self-reported transformation programs across industries, not from AI capability work specifically. Treat them as direction, not as a benchmark for your company.
Baldwin, T. T. and Ford, J. K., 'Transfer of Training: A Review and Directions for Future Research,' Personnel Psychology, 1988; Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T. and Rohrer, D., 'Distributed Practice in Verbal Recall Tasks: A Review and Quantitative Synthesis,' Psychological Bulletin, 2006; Roediger, H. L. and Karpicke, J. D., 'Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention,' Psychological Science, 2006; McKinsey and Company, research on transformation success rates.