Re:McDonald's will stop testing AI to take drive-thru orders...
by moose_man on Monday June 17, 2024 @12:18AM
(#40702260)
I'm not sure I believe that math though. The person taking the order is also the person responsible for the drinks and the payment. Most of the inside is already done through app and kiosks. Drive through also has the ability to pick up app orders. The drive through order taking isn't the big cost here. You would still need that person to hand food, prep drinks. The real time is in prepping the order not taking the order. Reason for the line at the drive through is the meal prep not the ordering process. If AI slows down the ordering process in any way it is actually a net negative.
Re:McDonald's will stop testing AI to take drive-thru orders...
by devoutsalsa on Monday June 17, 2024 @12:58AM
(#40702441)
Many McDonald's have 2 windows in the drive-true, one for payment and one for food & drinks.
Re:McDonald's will stop testing AI to take drive-thru orders...
by spaceship__sun on Monday June 17, 2024 @12:17AM
(#40702259)
Exactly. A textbook example of using first principles for reasoning rather than just saying 'The cost of implementing this system vs paying a teenager minimum wage makes it not worth' - the equivalent of cavemen reasoning.
Re:McDonald's will stop testing AI to take drive-thru orders...
by lmm on Monday June 17, 2024 @01:04AM
(#40702474)
> Minimum wage is $16/hr in California + employee overheard so lets say $25/hr. With 3 people covering shifts at the drive thru, that's $219k per year.
The part that you can get an AI to do, you can just as easily outsource to Indonesia or somewhere.
That's a math conclusion so lets show our work
Minimum wage is $16/hr in California + employee overheard so lets say $25/hr. With 3 people covering shifts at the drive thru, that's $219k per year.
Now for this system, lets look at off the shelf retail API prices to get a conservative guess. $0.006/min or $8.64 per 24 hours to listen & transcribe the customer, then add TTS + GPT-4o for $50 per million tokens in and out (thats way more than they need for a day). So that's $21k per year.
So back of the napkin math says it's 10x cheaper to implement this system. Unless you're referring to the corporate R&D budget. But that's a one time cost.