The pattern is so consistent it is almost a law. Week one, review volume doubles. Week two, holds. Week three, drops by half. Week five, back to baseline, and by week eight nobody can remember whose job it was.
Then someone re-briefs the team, and the same curve happens again.
Worth being precise about why, because the standard diagnosis — a motivation problem — leads to the standard fix, which is another briefing.
Cause one: it is the last thing in a sequence
The ask happens at payment, which is the end of the table's journey and also the moment the server is furthest behind. Everything else they are doing at that moment has a consequence if skipped: an uncleared table blocks a booking, a missed check-back generates a complaint. Skipping the review ask has no consequence at all today.
So it is not competing on importance. It is competing on immediacy, and it loses every time the floor is busy — which is precisely when you have the most guests to ask.
This is why the decay is not gradual. It is a step change on the first genuinely hard shift, and after one shift of not doing it, the habit no longer exists.
Cause two: nobody sees the result
A server who asks eight tables on a Friday will never learn whether any of them wrote anything. The reviews arrive days later, attributed to nobody, and appear on a screen the server does not look at.
Behaviour without feedback does not persist. This is the most fixable of the three causes and the most consistently ignored: the loop is broken not because the information is unavailable but because nobody carries it back to the floor.
Cause three: the person who owned it left, or moved shifts
Review asking, in most venues, is genuinely held by one or two people rather than by a process — as who asks matters more than what they say argues, that concentration is actually correct. The consequence is fragility. One of them going on holiday takes the programme with them, and because volume falls gradually nobody connects the two events.
The three things that make it survive
Read the new reviews out at pre-shift. Out loud. By name.
Ninety seconds, three or four reviews, including who was mentioned. This single practice does more for sustained review volume than any script, because it closes the feedback loop from cause two and it does it in front of everybody. A server who hears a guest describe something they did will ask again that night.
It also has to be the new reviews, not a monthly summary. The connection between asking and appearing has to be close enough in time to feel causal.
Attach it to something that already has a consequence.
The ask survives if it lives inside a step that cannot be skipped. In practice that means the card machine or the receipt — a link on the receipt happens whether or not anyone remembers, and the verbal ask becomes an enhancement rather than the mechanism. Whether your link actually works when someone taps it is a separate and frequently fatal question.
Name a person and check weekly, in one sentence.
Not a dashboard, not a target. A GM asking two designated people "how many did you ask this week?" every Friday. The number does not need to be accurate. The asking of the question is the entire mechanism, and it takes eleven seconds.
What does not work, based on watching it not work
Targets. "Twenty reviews a month" produces gaming at best and gating at worst — staff asking only guests who look pleased, which is against platform policy and generates a review profile that reads as suspiciously uniform.
Incentives for staff. Same problem, more so, plus it makes the ask sound like what it now is. Guests can hear it.
Re-briefing the whole team. Produces the two-week curve again, reliably, and each repetition works slightly less well because the team has learnt that this initiative expires.
Posters in the staff area. Read twice, then invisible. Nothing on a wall survives more than a fortnight of being walked past.
The reasonable expectation
A sustained programme in a mid-sized venue moves review volume by something like a third to a half over baseline, not by a factor of three. The week-one number is not the number; week one is novelty, and treating it as the new normal is what makes weeks three onward look like failure.
Set the expectation at the start, or the programme will be judged against its own best fortnight and abandoned for underperforming.
The bit worth automating
The pre-shift read-out is the highest-value practice here and it depends on someone assembling the new reviews before every shift, which is exactly the kind of small recurring task that stops happening in week three for the same reasons everything else does.
That is the part OMMU covers — the new reviews arrive already gathered, with the names and dishes pulled out, so the ninety-second read at pre-shift needs no preparation from anybody.