Most of the questions we get from readers aren’t about the research. They’re about the order. Where do I actually start? What do I do in week one? At what point do I bring the partners in? So this note is one reader’s sequence, in her words, with permission — including the bits that didn’t go to plan.

The reader is the head of people at a 90-person creative agency in Lyon. We’ll call her Camille. The agency had two formal harassment complaints in the eighteen months before she started the programme; six months after she finished, it had a quarterly pulse score on “I feel respected by my colleagues” that had moved from 6.4 to 8.1 out of 10. What follows is what she did between those two numbers.

Step one — assess before you announce

Camille’s first mistake, in her own telling, was an instinct to announce the programme on day one. She didn’t. Instead, she ran an anonymous baseline survey for two weeks — sixteen questions, mostly drawn from the climate-survey instruments that the CDC uses in its school-based work and adapted to the office. Response rate was 71%. Three teams scored noticeably worse than the agency average on two questions: “I have seen behaviour at work in the last month that I would describe as disrespectful” and “If I reported a problem, I trust it would be handled fairly.”

Without the baseline she would have rolled the workshop out evenly across the agency. With it, she knew which three teams needed the workshop first, and which two questions to track in the follow-up pulse. The Olweus literature is unambiguous on this: the programmes that move the needle are the ones with a real before-number. Without one, you cannot tell whether anything you did mattered.

Step two — name the behaviour, in French and in English

The Lyon office is bilingual, and Camille noticed early that the English vocabulary the team used — “banter”, “rough patch”, “bit much” — softened conduct that the French harcèlement moral framing would have called by its name. So her second move was a 30-minute all-hands in which she walked through the four definitions in the handbook — harcèlement moral, harcèlement sexuel, agissement sexiste, discrimination — with the Code du travail article numbers on screen, in both languages.

She did not soften the legal language. The point of the session, she said afterwards, was to make the words usable. A team lead who has heard the phrase “agissement sexiste” said out loud, in a calm meeting, by the head of people, is more likely to use it later when something happens.

Step three — train the bystanders, not the offenders

This is where Camille leaned hardest on the Olweus model. The two-hour facilitator-led workshop in the ArtPartner programme is built around the bystander, not the actor — ten scenario cards, all drawn from real incidents (anonymised), all asking the same question: you’re the third person at the table. What do you do in the next ninety seconds?

She ran the workshop seven times in eight weeks — small groups, mixed seniority, no opt-outs for partners. The partners going through the same workshop as the junior designers was, she said, “the most expensive decision I made and the one that changed the room.” The Olweus research, the StopBullying.gov practitioner guides, and twenty years of school-based replications all converge on this: training only the people who already care is theatre. The signal comes from training the people whose silence has been read as endorsement.

Step four — document the procedure, then publish it

By month four Camille had a written reporting procedure that named two referents (one HR, one elected from the CSE), three intake channels (email, in-person, an anonymous form), and a 48-hour acknowledgement standard. The procedure went on the intranet, in both languages, with the article numbers and the names of the referents on the front page. The DUERP — the French employer’s legal risk-assessment document — was updated to include psychosocial risks, with the survey data attached as evidence.

None of this is glamorous. All of it is what the law assumes you have done. The handbook chapter on documentation exists because, in the cases where things later went badly, the absence of these documents was usually the thing the labour inspector and the conseil de prud’hommes pointed at first.

Step five — review on a calendar, not a crisis

Camille set the pulse survey to repeat every quarter, with a standing 30-minute slot on the executive committee agenda to read the results out loud. Not a dashboard. Not a slide. Read out loud, with the worst three open-text comments quoted verbatim. That, she said, was the discipline that kept the programme alive past the launch high.

A programme that survives is a programme with a calendar. A programme that dies is a programme with a launch.

Eight months in, the agency has a baseline, a vocabulary, a trained team, a written procedure, and a quarterly review — the five pieces the programme is designed around. It is not finished, and Camille would be the first to say so. But it is no longer one person’s side project. It is the way the office works.