Four stations, one hour, one constraint
A hundred and eighty drinks in the peak hour — forty espresso, a hundred and ten milk, thirty poured — needs 150 baskets. At nine seconds a basket that is 1,350 seconds of grinder against 7,200 available across two grinders: nineteen percent. At forty seconds of group head occupancy per basket it is 6,000 seconds against 7,200: eighty-three percent. The wands take 3,850 of 7,200, the hands 4,410 of 7,200.
The group heads bind. Scale the mix up until they fill the hour and it is 216 drinks. That is the ceiling, and the other three stations have no say in it.
The second grinder that does nothing
Add a third grinder and grinder capacity goes to 10,800 seconds against 1,350 of demand. The ceiling stays at 216 drinks, exactly, because it was never the grinder. This is the most expensive mistake on a small bar and it is made constantly, usually because the grinder is the station where the queue visibly forms — the barista is standing there, so that must be the problem.
Add a third group head instead and the head capacity goes to 10,800 against 6,000, which allows 324 drinks. The hands now bind at 4,410 seconds, which allows 294. The ceiling moves from 216 to 294, a gain of 78 drinks an hour. Same money, wildly different answer.
The input that decides everything
Group head seconds. Almost everybody enters the shot time — twenty-eight seconds, say — when the honest figure is how long the head is unavailable: lock in, extract, unlock, knock out, rinse. Forty is not a long estimate for that. Enter twenty-eight instead and the demand drops to 4,200 seconds, the heads fall to fifty-eight percent, the hands take over as the constraint and the ceiling reads 294. The bar has not changed at all. The number was wrong.
The same trap sits in the steam seconds, where purging and wiping are usually left out, and in the grinder seconds, where people time the dose rather than the portafilter round trip.
What a capacity number cannot tell you
How long anyone waits. This is a production ceiling under an assumption that work arrives evenly, and it does not. Arrivals bunch, and a bar running well inside its ceiling can still have a ten minute line when a bus stops outside. Capacity and waiting are different calculations and confusing them produces confident, wrong staffing decisions in both directions.
The model also pools stations. Two group heads count as two heads worth of seconds, which holds while somebody is free to start the next shot and stops holding when one person is working both. On a two person bar at full tilt, the real capacity sits somewhere below what the table says.
Questions people ask
What limits how many drinks an espresso bar can make?
Whichever station runs out of seconds first, and it is usually not the one where the queue is visible. With the defaults here the group heads bind at 83 percent while the grinders sit at 19, so the ceiling is 216 drinks an hour and no amount of grinding capacity changes it. Which station binds moves with the drink mix, so a milk-heavy hour and an americano-heavy hour do not have the same answer.
Is a second grinder worth it?
Only if the grinder is what binds, and on most bars it is not. The page works the ceiling with and without one so you can see the gain in drinks per hour rather than argue about it. With the defaults the gain is zero, because the group heads were the constraint. Change the mix or the grind time and that flips.
How many seconds should I put for a group head?
How long the head is occupied and unavailable, not how long the shot runs. Lock in, extract, unlock, knock out, rinse. Timing the shot alone is the single most common error on this form and it moves the answer a long way — with the defaults, entering 28 rather than 40 changes both the ceiling and which station binds.
Does this tell me how long customers wait?
No, and it is important that it does not pretend to. This is a capacity ceiling under an assumption of even arrivals. Real arrivals bunch, so a bar comfortably inside its ceiling can still build a long line for ten minutes. Waiting is a queueing calculation and there is a separate page for it.
Why is the till not in this?
Because taking orders and payments is driven by arrival variability far more than by average service time, and folding it into a capacity model would make both answers worse. If the register is your constraint it shows as a line at the counter rather than as undelivered drinks, and that is the queueing calculator rather than this one.