Baemin's 1.7-times accuracy claim implies merchant-typed ETAs were right under 30% of the time

Fifty divided by 1.7 is 29.4. Taking Baemin's two disclosed figures at their floors, the cooking and delivery times restaurant owners typed in by hand were right fewer than three times in ten ^[5][1]. The article does not define what counts as an accurate forecast, so the tolerance behind both numbers is unstated ^[6].

What replaces the typing reads 58 kinds of data, among them order volumes at individual restaurants, congestion in the surrounding area and how many riders are free ^[4]. Baemin says it built the system to take the estimating burden off owners and cut forecast error ^[3].

Over the year to last month Baemin added 1.34 million monthly active users and Coupang Eats added 2.70 million, almost exactly twice as many ^[2][3][4]. Baemin's lead narrowed from 11.32 million to 9.96 million ^[5]. Coupang Eats now has 59.2% of Baemin's monthly user count, against 50.9% a year earlier ^[6].

Sedaily reports the industry is spending on accuracy because it treats the delivery quality users actually experience as the factor that determines repeat use, and that free-delivery promotions can bring in users who leave again as late-delivery complaints pile up ^[10][11]. A subsidy is paid on every order. A forecasting model is built once and then runs, and the substitution only pays if better arrival times hold users who would otherwise have gone.

I'd expect this to settle as a cost both platforms carry with neither taking users off the other because of it. Coupang Eats grew 23.0% over the year with a coordinate fix that shipped only recently, and its official said store-location inquiries have fallen sharply since adoption and that orders taking more than an hour also declined ^[13][7][9]. The 50% figure could equally reflect a strict tolerance, in which case the model is better than the number sounds ^[5][6].

"The fiercer the competition among delivery apps becomes, the more this will emerge as an important service advantage," an official in the retail industry said ^[14].

The check arrives in the monthly user counts after launch. If Coupang Eats keeps adding around 2.7 million users a year once Baemin's forecasts improve, the accuracy work is maintenance on a base of 24.4 million ^[3][12]. If Baemin's growth moves up off 5.8% without a new fee promotion, arrival-time accuracy is the plausible reason ^[12].