In 1977 Trevor Stokes and Donald Baer did something that someone in every discipline should eventually do and almost nobody enjoys doing: they went through the first ten years of their own journal, the Journal of Applied Behavior Analysis. Along the way they sorted every technique their colleagues had used to carry a behavior learned in a therapy session out into the world, into the classroom, the home or the company of strangers. Nine categories came out of it, and the two of them gave the first one a name so dry that you read it twice before you laugh: “train and hope”. It described an approach in which a behavior is carefully built in one place, nothing is planned for the way out, and someone checks afterwards whether it happened to come along.
You could take this for an anecdote from the early days of a young science, if so many of the companies I have seen from the inside over the last few years did not work in exactly this way, only with bigger budgets and better slides. A learning programme, a change initiative or an AI rollout is built over months, with an agency, a learning platform and a kick-off, and for the way into everyday work there is a reminder email, a dashboard of usage figures and, well, hope. Each of these programmes is, soberly put, a bet that a few hundred people will do something differently on an ordinary Monday than they do today.
How that bet tends to turn out is, unfortunately, quite well documented. Training professionals from 150 organizations estimated for Alan Saks and Monica Belcourt that one year after a training 34 percent of participants still apply what they learned. MIT’s NANDA initiative concluded in 2025 that 95 percent of the companies it studied get no measurable return from their investments in generative AI. At Engaginglab we call the point where a new behavior breaks off in everyday work the transfer break, and because that point is often visible in the plan already, I turned it into a review step I call the Monday Test. It puts five questions to a programme eight to 16 weeks before rollout.
Why before and not after
After rollout almost every company checks very thoroughly, it just checks the wrong thing, because the figures of the first weeks, registrations, completion rates and active users, nearly always look good. An AI assistant that most of the workforce has opened once appears in the report as a success, even if it has taken over a task at hardly any workstation. By the time anyone notices the difference, the licences are paid, the agency has invoiced and the programme lead has put her name under a success report.
A finding after rollout is therefore always an accusation as well, and organizations are remarkably good at not hearing accusations. The same finding before rollout is a correction that costs no more than a change to the plan: an exercise on real cases instead of the agency’s examples, a metric that is reported differently, or a contest that is never announced in the first place.
Who does what differently on Monday
The first question of the Monday Test sounds so banal that you are almost embarrassed to ask it: is it named who is supposed to do what differently on Monday? In a surprising number of plans this spot holds a word like “competence”, “culture” or “AI readiness”, an attitude nobody can see in anyone at work. What is needed instead is a situation with a person, a trigger and an action, such as the case handler who lets the assistant draft a reply to every customer complaint before she types anything herself. Edwin Locke and Gary Latham showed over decades that specific goals work more reliably than general appeals, and a Monday behavior is nothing but a goal specific enough to be observed.
The second question checks whether the person can handle this situation and whether their everyday work allows it, and it draws on the COM-B model by Susan Michie, Maartje van Stralen and Robert West, according to which behavior needs capability, opportunity and motivation. Opportunity is what plans overlook most often, because it never comes up in the training room. The tool sits two clicks away from the ticket system instead of inside it, nobody has written down which customer data may go into the assistant, and the manager who championed the rollout still asks on Thursday why the pile is not done yet.
A programme is a bet that people will do something differently on an ordinary Monday, and the bet can only be changed cheaply before anyone has paid the stake.
The reward that replaces the behavior
In 1973 Mark Lepper, David Greene and Richard Nisbett promised preschool children who already liked drawing with felt-tip markers a certificate for good drawing, and one to two weeks later exactly these children drew less in free play than children who had received no certificate or had received it as a surprise. The third question of the Monday Test is therefore called the starter aid, and it asks whether a reward builds the behavior or replaces it, that is, whether there is any reason left to do the thing once the bonus, the points or the contest are over.
The fourth question concerns the metric and asks whether the easiest way to the target leads through the work itself. In 1902 the French colonial administration in Hanoi paid one cent for every rat tail handed in to curb a rat plague. Soon tailless rats were running through the city, released again so they could breed more tails, and health inspectors found rat farms on the outskirts, as the historian Michael Vann has reconstructed. A dashboard that counts active users or completed learning modules pays for the tail on the same principle. Lisa Ordóñez and her colleagues showed in “Goals Gone Wild” how reliably people find the detour as soon as money or recognition hangs on the number. Their first example is the Sears auto centers, whose mechanics were given a sales quota of 147 dollars per hour in the early 1990s and went on to sell repairs nobody needed.
1 Monday behavior: Is it named who is supposed to do what differently on Monday?
2 Capability and opportunity: Can the person handle their Monday situation, and does their everyday work allow it?
3 Starter aid: Does the reward build the behavior, or does it replace it?
4 Direction of the target: Does the easiest way to the target lead through the work itself?
5 Handover to everyday work: What carries the behavior once training and campaign are over?
What still reminds people in week twelve
The fifth question is the most uncomfortable, because it aims at a period for which hardly any plan has pages left, the weeks after the campaign, when the agency has invoiced and the reminder emails have gone quiet. Phillippa Lally and her colleagues at University College London observed in 2010 that a new everyday behavior took a median of 66 days of daily repetition before it ran automatically, with a range from 18 to 254 days, and hardly any rollout plans more than a survey for that period.
Katherine Milkman showed at the University of Pennsylvania how quickly even a well-built effect fades when nobody takes it over: students who were only allowed to listen to gripping audiobooks at the gym went to the gym 51 percent more often than the control group. Over the Thanksgiving break visits then dropped in all groups, and Milkman commented with a sentence that could hang in every training department: “habits formed during school terms tend to be eliminated during breaks”. A programme therefore needs a recipient that takes the behavior over when the campaign leaves, an activity that gives something back by itself above a certain level of skill, a fixed trigger in the workflow, or at least a management report that measures the Monday situation instead of the usage figure.
An AI assistant, checked once
How the five questions work together is shown by a case assembled from patterns that are typical of AI rollouts. An insurer wants to introduce an AI assistant at 1,400 workstations, the rollout starts in eleven weeks, and the agency’s plan is carefully made, with training videos, a prompt library and a contest in which the team with the most prompts in the first month wins a dinner together. In the six conversations with employees everyone names the same situation, the reply to a customer complaint that takes twenty minutes today, so the first question holds. The training, however, never practises exactly this situation, and the release of customer data is still with the works council, both of which can be fixed before the start.
The plan breaks at the contest, because it rewards the number of prompts instead of their use and simply ends with the dinner, and the dashboard needs work, because it reports active users as success. The verdict is therefore release with conditions: the training practises the complaint reply on real cases, the contest gives way to recognition for concrete solved cases, and the report to management counts the complaints answered with the assistant instead of the times it was opened. None of these conditions costs much while the programme is not yet running, and each of them would have been an admission after rollout.
Three conditions, all of which can be met before launch: training on real complaints, contest dropped, report counts answered complaints.
The Bauprobe
German theatres have an appointment the audience never gets to see and at which a great deal about the evening is decided all the same, the Bauprobe, a kind of set mock-up rehearsal. Before the workshops build the stage set, it is marked out on the real stage with simple means at a scale of one to one, so that everyone can see whether the sightlines work and whether the design that looked so convincing in the model also works in the room. The Monday Test is a Bauprobe for programmes and deliberately takes place where the behavior will later have to survive. That is why six conversations with employees sit next to the documents, because anyone can read a plan, but only the people who live through Monday every week know it.
Whoever is responsible for a programme is, in this picture, the last person who can still change something about the set before it is built, and that fits the verdict the Monday Test ends with: release, release with conditions, or no release. Even a review note saying “Stopped before launch” pins down a risk at the only point where it does not yet belong to anyone. The most honest of the five questions can, incidentally, be put to any planned programme today, without any review at all: which situation next Monday is supposed to run differently afterwards, and who in the organization could describe it in one sentence?