Workshops, Retrospectives, and Making It Stick

A workshop is not a long meeting, and the difference in how it should be planned is specific.

The common failure. Someone books a day, invites people, prepares topics, and runs discussion. Everyone finds it interesting and engaging. Nothing exists at the end except a photograph of some sticky notes and a shared sense that it was useful. Two weeks later nobody can say what was decided.

The design principle that prevents it. Start from the artefact you need to exist at the end, then work backwards to the activities that produce it.

What counts as an artefact. A prioritised list. A decision with its reasoning recorded. A plan with owners and dates. A map of a process with the problems marked. A written statement of what the group means by something contested. Each is a concrete object that survives the room.

Why this changes the design rather than just the reporting. Each activity now has a required output that feeds the next, so you can tell during the day whether you are on track. A workshop designed as topics cannot fail visibly until it is over.

The practical method. Write the artefact's structure before designing anything. If the output is a prioritised list, you need candidates, criteria, and a prioritisation step, in that order, and each has a time cost. That immediately tells you whether the day is realistic, which is the question most workshop plans never confront.

And the honest test to apply while planning. If we do this and produce nothing written, what was it for? If the answer is that people would understand each other better, that is a legitimate goal and should be stated, because it changes the design entirely.

The structural pattern underneath almost every workshop technique, and knowing it lets you design sessions without a book of exercises.

Any productive session alternates between two modes.

Divergence expands the space: generating options, surfacing problems, collecting perspectives. It needs breadth, quantity over quality, and an explicit prohibition on evaluation, because evaluation stops generation immediately.

Convergence narrows it: grouping, prioritising, choosing, deciding. It needs criteria, critical judgement, and a mechanism for concluding.

The failure the pattern names, and it is the most common thing that goes wrong in workshops. Groups attempt both simultaneously. Someone offers an idea and someone else immediately explains why it will not work. Within three exchanges, contribution stops, because people have learned that offering something costs them.

The uncomfortable middle, sometimes called the groan zone in Sam Kaner's work on participatory decision-making, is worth knowing about. Between divergence and convergence there is a period where many options are on the table, none is obviously right, and the group feels stuck and frustrated. Inexperienced facilitators treat that as failure and rescue the group by narrowing prematurely. It is a normal and necessary stage, and saying so out loud when it arrives is genuinely helpful.

The practical design. Each block is explicitly one mode, announced. We are collecting for twenty minutes and we will not evaluate anything. Then a break. Then we are choosing, and here are the criteria.

The break matters more than it seems. It marks the mode change physically, and it lets people stop defending what they generated before they have to evaluate it.

A day-long session has a shape that a one-hour meeting does not, and ignoring it wastes the most valuable hours.

The opening block establishes why everyone is here, what will exist at the end, and how decisions will be made. Fifteen minutes, and it prevents the drift that otherwise consumes the morning.

The first working block is where the hardest thinking should go, because attention is highest early. Groups routinely spend it on context-setting and warm-up, then attempt the difficult decision after lunch when the room is at its worst.

Then divergence, in a bounded block with evaluation explicitly prohibited.

Then a break, which marks the mode change.

Then convergence, with stated criteria and a mechanism for concluding, in the block after the break rather than in the last hour.

Then commitment: owners, dates, what happens next. This must not be squeezed, and it is what gets squeezed.

And throughout, breaks every sixty to ninety minutes, which is not indulgence. Sustained group attention degrades, and a session that runs three hours without a break has people present and not thinking.

The scheduling principle worth extracting. Put the decision that matters most in the morning, and never immediately after lunch. Facilitators plan around the agenda's logic and forget that the room's capacity varies by hours in a way that dominates the ordering.

Narrowing a long list is where workshops most often produce a result nobody believes in, and the mechanism chosen determines the quality.

Why simple voting is weak. Show of hands is anchored by who raises first and by watching others, which is the conformity problem from the previous lesson. It also gives no information about intensity: an option that everyone finds acceptable beats one that half the room considers essential, which is frequently the wrong outcome.

Mechanisms that work better, and each suits a different situation.

Dot voting, where each person distributes a fixed number of marks across options and may stack several on one. Simple, fast, and it captures intensity because someone can spend everything on one item. Have people place marks simultaneously, or the first placements anchor the rest.

Ranking rather than selecting, which forces trade-offs and prevents everything being important.

Scoring against explicit criteria agreed before the options are scored. Slower, and it makes the reasoning visible and produces a record of why, which matters when the decision is questioned later.

And consent rather than consensus, which is drawn from sociocratic practice and is the most useful for a group that keeps failing to conclude. The question is not does everyone prefer this, but does anyone have a serious objection. Those are different, and the second is answerable. Groups seeking preference-agreement can deadlock for hours over options nobody actually objects to.

The facilitator's discipline in all of these. Agree the criteria before seeing the options where possible, because criteria chosen afterwards are chosen to select a preferred answer, which is the self-serving-choice problem from the data storytelling cursus arriving in a room.

The retrospective is the most common recurring workshop and the one that most reliably degrades into ritual, for reasons worth understanding.

The failure pattern. The same surface issues are raised each time. Nobody mentions the actual problem, which is usually about a person, a decision made above the team, or a constraint nobody feels able to name. Actions are generated and not done. Attendance becomes obligation, and the session confirms that raising things achieves nothing.

The root cause is almost always safety rather than technique. Amy Edmondson's research on psychological safety, developed over three decades, found that team performance depends heavily on whether members believe they can raise problems and admit mistakes without penalty. In its absence, no format produces honesty, and a retrospective that people do not feel safe in produces a list of tooling complaints.

What helps, in rough order of importance.

Whoever holds power over the participants should consider not attending, or should speak last and least. People do not discuss what went wrong in front of the person who decides their promotion, however good that person's intentions are.

Written first, again. Collect observations silently, and anonymously if the topic is sensitive, before any discussion. This is the same anchoring fix as the previous lesson and it matters more here.

Separate what happened from what to do. Groups jump to solutions to escape discomfort, and the diagnosis is where the value is.

Pick one thing. A retrospective producing eight actions produces zero. One change, with an owner, checked at the next session, is the entire mechanism by which retrospectives improve anything.

And start by reviewing whether the last action happened. If it did not, that is the most important item on the agenda, because it is why nobody believes in the process.

The most common complaint about workshops is that nothing changes afterwards, and the causes are identifiable and mostly fixable.

The actions had no owner. We should improve handovers is nobody's job. This is the single largest cause and it is entirely preventable in the last ten minutes of the session.

The actions were too large. Redesign the onboarding process is a project, not an action, and it will not start. What is needed is the first concrete step with a date: draft the new checklist by the fourteenth.

Nobody had capacity. The actions were added to people already fully committed, and workshop outputs lose every time against existing obligations, because those have deadlines and stakeholders. Unless something was removed, nothing was actually decided.

The decision-maker was not there. The group agreed something requiring approval, and the approval never came, which teaches everyone that the workshop was advisory theatre.

Nothing was written while memory was fresh. Notes produced a week later lose the reasoning, and reasoning is what lets people apply a decision to the cases nobody anticipated.

And nobody checked. Without a scheduled point where progress is reviewed, actions decay silently, and the next session's participants know it.

The fixes are unglamorous and they work. Every action gets a named person and a date, stated aloud before people leave. Actions are small enough to start this week. Someone asks explicitly what will be dropped to make room. Notes go out the same day. And there is a date in the calendar, set during the session, for checking what happened.

That last one does most of the work. A workshop with a scheduled follow-up produces different behaviour from one without, because participants know the commitment will be revisited.

Most facilitation guidance assumes a neutral facilitator. In practice the person running the session usually has a stake in the outcome, and pretending otherwise is worse than acknowledging it.

The difficulty. A facilitator controls the agenda, who speaks, when discussion ends, and how conclusions are summarised. Each of those is an opportunity to steer, and a facilitator with a preferred outcome will steer, mostly without intending to: by giving supporters more room, by summarising in favourable terms, by closing the discussion when it has reached the right place.

Participants detect this reliably, and once detected the session is discounted entirely, including the parts that were fair.

The options, in order of preference.

Get someone else to facilitate. A colleague from another team, or someone junior with the explicit authority to run the process. Cheap and it solves the problem outright.

Separate the roles explicitly if you must do both. State it: I have a view on this, I will give it at this point, and while I am facilitating I am not arguing for it. Then hold that line, and give your view once, marked as your view.

Use mechanisms that constrain you. Written collection, structured rounds, and scored criteria all reduce the facilitator's discretion, which is exactly what you want when your discretion is compromised.

And invite the challenge. Tell me if you think I am steering this. Saying it makes it possible for someone to raise, which it otherwise would not be.

The worst option, which is common, is claiming neutrality you do not have. Participants who suspect it stop engaging honestly and say so afterwards to each other rather than to you, and you lose both the input and the ability to know you lost it.

Three lessons, as things to do.

Before the meeting exists, which is where most of the outcome is determined. Ask whether it could be a document, because transmitting information and collecting input are both better written. Name the type, and never mix generating with deciding. Write outcomes rather than topics. Give every attendee a reason for being there, and send the notes to everyone else. And for recurring meetings, set an end date at creation, because ending one otherwise requires a decision nobody owns.

In the room. Say how the decision will be made, at the start, because the frustration comes from the mismatch rather than the outcome. Collect views in writing before discussion, since the first speaker anchors everyone and the loudest voice is not the most informed. Allocate airtime deliberately, or status will allocate it. Treat easy agreement as a warning and run a pre-mortem. Diagnose disagreements by type, because factual, predictive and values disagreements need completely different handling. And defend the last five minutes for decisions, owners and dates.

For longer sessions. Design backwards from the artefact that must exist at the end. Alternate divergence and convergence explicitly, never both at once, and expect the uncomfortable middle. Put the hardest thinking early and never after lunch. Converge with a mechanism that captures intensity, and agree criteria before seeing options. And in retrospectives, safety determines honesty more than format does, so consider who is in the room.

And afterwards, which is where it usually fails. Named owners, small enough actions, something explicitly dropped to make room, notes the same day, and a scheduled date to check.

The unifying observation. Almost nothing here is about running a room well. It is about decisions made before and after, which is where meetings are actually won or lost.