On this page
Every bar that has ever had a consistency problem has already tried training. Usually more than once. The recipes get written down, the team gets walked through them, everyone agrees, and for about three weeks the drinks are excellent. Then service gets busy, the new starter learns from whoever is next to them rather than from the sheet, and the standard drifts back to wherever it was.
This is not a failure of the training. It is a misunderstanding of what training is for. Training establishes a standard. Holding a standard is a separate job, and most bars have never assigned it to anything.
Why the drift is structural
Consider what a busy service actually rewards. Speed is visible and immediate: the queue shortens, the guest is happier, the shift feels better. Accuracy is invisible and delayed: a heavy pour costs money that shows up in a stock count weeks later, attributed to no particular night and no particular person.
Put a feedback loop that fast against one that slow and the fast one wins every time. That is not a character flaw in the team. It is what any reasonable person does when one signal arrives in seconds and the other arrives in a month, if it arrives at all.
Four things that actually hold a recipe
1. Make the standard physically easier than the alternative
Jiggers, speed pourers with a known count, pre batched high volume cocktails. None of this is new and none of it is exciting, but it is the only intervention that works by removing the choice rather than by asking people to make the right one under pressure.
The objection is that it slows service. Sometimes it does, at first. It is worth measuring rather than assuming: a jigger costs a second, and a remake costs a minute.
2. Teach the reason, not just the number
A bartender told to pour 50ml will pour 50ml when someone is watching. A bartender who understands that the drink is balanced at 50ml and tastes thin at 40 and hot at 60 will pour 50ml because they want the drink to be right. The second one holds under pressure. The first one does not.
This is the difference between compliance and skill, and it is the reason training on the floor beats training in a binder. The moment to explain why a pour matters is the moment it drifted, not six weeks later in a review.
3. Close the loop within the week
If variance is only visible at month end, nobody can connect it to anything they did. By the time the number exists, the shifts that produced it are a blur. Bringing that loop inside the week is the single change that makes the feedback usable, because people can still remember the nights in question.
This is the argument for looking at the transaction record at the level of the shift rather than the period. Not because shift level data is inherently better, but because it arrives while the memory of the shift is still intact.
4. Make it about the drink, not the person
The fastest way to make a consistency programme fail is to run it as surveillance. If the output of your system is a list of who poured heavy, you will get defensiveness, and you will also get people who learn to game the measurement rather than pour better.
The framing that works is the drink. This one drifted, here is what it costs, here is how we get it back. Fair and visible beats personal every time, and it is the only version that survives the bartender in question hearing about it.
What this looks like in practice
A workable weekly rhythm for a single site is short. Look at the products where variance concentrated. Look at which shifts they concentrated in. Pick one thing. Say it to the team as a drink problem with a number attached. Check the same thing the following week.
That is a fifteen minute job if the data is in front of you and a two hour job if you have to assemble it first, which is the actual reason most bars do not do it rather than any disagreement about whether it would help.
The part worth being honest about
None of this makes every drink identical. Bartenders are not machines and a bar that treats them as machines will lose the ones worth keeping. The realistic goal is a narrower band, not a single point: a guest ordering the same drink on two different nights gets a drink they recognise.
If you want to see where your own band currently sits, that is what a CoreTAP demo shows against your own trading. If you would rather read the underlying study first, it is on the research page.