Most retrospectives produce a tidy list of things nobody was afraid to say.
The genuinely useful material — the decision that should have been questioned, the dependency everyone could see coming, the fact that the deadline was never real — stays in people's heads, and the team carries it into the next piece of work fully confident it has been reviewed.
This is not a courage problem. It is a format problem, and you can fix a lot of it without hiring anyone. Here is the sequence, and you can run it yourself at your next retro.
Start with silence, not a round-table
Before any discussion opens, everyone writes. Individually, privately, on their own paper or in their own doc, no talking. Give the room one clear prompt — what actually got in the way — and let people fill the page.
The point is that the first thing said out loud in a retro sets the ceiling for everything after it. If the first contribution is mild, the room calibrates to mild. Writing first means every position exists before anyone has heard another one, so you get independent accounts rather than an echo of whoever spoke first.
Round-robin, and everyone reads their own
Then go around the room in order, and have each person read out what they wrote. Not a facilitator summarising a pile of sticky notes — each person, their own words.
Two things this fixes at once. Nobody can pass invisibly, which is how the quietest and often sharpest reads get lost. And it removes the moment of decision that stops most people speaking: you are not choosing to raise something contentious in front of the room, you are reading out what you already wrote down, which is a much smaller act.
Cluster before you argue
Once everything is out, group the items that are really the same item. You will usually find three or four themes doing the work of twenty complaints, and the clustering is worth doing before any debate starts.
Resist the urge to solve as you go. The moment you start problem-solving item one, the rest of the list gets less attention than it deserves, and the loudest item wins on running order rather than importance.
Leave with one change and a name against it
Pick one. Not a list of nine, one. Then attach a person and a date to it, out loud, in the room.
A retro that produces nine improvements produces none, because nobody owns any of them. A retro that produces one change with a name against it changes something, and the next retro has evidence that raising things works — which is the actual mechanism that makes people willing to raise things again.
Where this stops being enough
The technique above handles the ordinary case, and for a lot of teams it is the whole fix. It runs into a wall in two situations. One is where the person who owns the decision under review is also in the room and holds the pen on performance reviews — no amount of format will fully get past that, and it usually needs somebody from outside chairing it. The other is where the same theme has now come up in three retros running and still nothing has shifted, which means the problem is not that the team cannot see it.
Both of those are the point at which an outside facilitator earns their fee, and the wrap-up and review format exists for exactly that. If you would rather check first whether you actually need one, the fit assessment will tell you if you do not.