This article was downloaded by: [140.116.206.

221] On: 28 August 2023, At: 22:20

Publisher: Institute for Operations Research and the Management Sciences (INFORMS)
INFORMS is located in Maryland, USA

Management Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org

Food Delivery Service and Restaurant: Friend or Foe?

Manlu Chen, Ming Hu, Jianfu Wang

To cite this article:

Manlu Chen, Ming Hu, Jianfu Wang (2022) Food Delivery Service and Restaurant: Friend or Foe?. Management Science
68(9):6539-6551. https://doi.org/10.1287/mnsc.2021.4245

Full terms and conditions of use: https://pubsonline.informs.org/Publications/Librarians-Portal/PubsOnLine-Terms-and-

Conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.

Copyright © 2022, INFORMS

Please scroll down for article—it is on subsequent pages

With 12,500 members from nearly 90 countries, INFORMS is the largest international association of operations research (O.R.)
and analytics professionals and students. INFORMS provides unique networking and learning opportunities for individual
professionals, and organizations of all types and sizes, to better understand and use O.R. and analytics tools and methods to
transform strategic visions and achieve better outcomes.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
 MANAGEMENT SCIENCE
Vol. 68, No. 9, September 2022, pp. 6539–6551
http://pubsonline.informs.org/journal/mnsc ISSN 0025-1909 (print), ISSN 1526-5501 (online)

Food Delivery Service and Restaurant: Friend or Foe?

Manlu Chen,a Ming Hu,b Jianfu Wangc
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

a
School of Business, Renmin University of China, Haidian District, Beijing 100872, China; b Rotman School of Management, University of
Toronto, Toronto, Ontario M5S 3E6, Canada; c College of Business, City University of Hong Kong, Kowloon 999077, Hong Kong SAR
Contact: chenmanlu@rmbs.ruc.edu.cn, https://orcid.org/0000-0002-7325-0372 (MC); ming.hu@rotman.utoronto.ca,
https://orcid.org/0000-0003-0900-7631 (MH); jf.wang@cityu.edu.hk, https://orcid.org/0000-0002-2393-8750 (JW)

Received: December 11, 2019 Abstract. With food delivery services, customers can hire delivery workers to pick up
Revised: November 17, 2020; July 2, 2021 food on their behalf. To investigate the long-term impact of food delivery services on the
Accepted: August 9, 2022 restaurant industry, we model a restaurant serving food to customers as a stylized single-
Published Online in Articles in Advance: server queue with two streams of customers. One stream consists of tech-savvy customers
February 8, 2022 who have access to a food delivery service platform. The other stream consists of traditional
https://doi.org/10.1287/mnsc.2021.4245 customers who are not able to use a food delivery service and only walk in by them-
selves. We study a Stackelberg game, in which the restaurant first sets the food price; the
Copyright: © 2022 INFORMS food delivery platform then sets the delivery fee; and, last, rational customers decide
whether to walk in, balk, or use a food delivery service if they have access to one. If the
restaurant has a sufficiently large established base of traditional customers, we show
that the food delivery platform does not necessarily increase demand but may just
change the composition of customers, as the segment of tech-savvy customers grows.
Hence, paying the platform for bringing in customers may hurt the restaurant’s profit-
ability. We demonstrate that either a one-way revenue-sharing contract with a price
ceiling or a two-way revenue-sharing contract can coordinate the system and create a
win-win situation. Furthermore, under conditions of no coordination between the restau-
rant and the platform, we show, somewhat surprisingly, that more customers having ac-
cess to a food delivery service may hurt the platform itself and the society, when the food
delivery service is sufficiently convenient, and the delivery-worker pool is large enough.
This is because the restaurant can become a delivery-only kitchen and raise its food price
by focusing on food-delivery customers only, leaving little surplus for the platform. This
implies that limiting the number of delivery workers can provide a simple yet effective
means for the platform to improve its own profitability while benefiting social welfare.

History: Accepted by Charles Corbett, operations management.

Funding: M. Hu was supported by the Natural Sciences and Engineering Research Council of Canada
[Grants RGPIN-2015-06757 and RGPIN-2021-04295]. M. Chen and J. Wang were supported by a
grant from City University of Hong Kong [Grant 7200664].
Supplemental Material: The data files and online appendices are available at https://doi.org/10.1287/
mnsc.2021.4245.

Keywords: food delivery • on-demand economy • service operations • omnichannel operations • queueing economics •
channel coordination • revenue-sharing contract • labor welfare

1. Introduction Adopting food delivery services, people with high

The first pizza delivery order was placed by Queen opportunity costs can avoid the simple yet unproduc-
Margherita of Italy during her visit to Naples in tive task of traveling to restaurants and waiting for
1889. In recent years, food delivery has become an eas- food there, allowing them to devote this time to more
ily accessible service for commoners rather than a royal fruitful tasks instead. Although the number of food
privilege. By simply swiping a smartphone, customers delivery orders grows exponentially over the years,
can order food via food delivery platforms such as there are still people lagging behind (Steingoltz and
Uber Eats, Grubhub, Postmates, and DoorDash. Food Picciola 2019). Some people are not tech-savvy enough
delivery is a booming industry. Online restaurant de- to operate digital devices such as smartphones, some
livery sales are estimated to grow to $62 billion in are simply reluctant to try out the food delivery op-
2022 from about $25 billion in 2019. Venture-capital tion, and some are sensitive to the delivery fee. These
firms invested $5 billion in U.S. food and grocery people still go to the restaurant as walk-in customers.
delivery services in 2018, more than four times the On the restaurant side, the benefit conferred by
amount invested in 2017 (Haddon and Jargon 2019). food delivery platforms is unclear. The current

6539
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
6540 Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS

practice in the industry is for platforms to take may not increase demand for the restaurant but just
10%–25% of food revenue as a service fee from the res- change the composition of customers, as the segment of
taurant and take a flat delivery fee of $3 per order tech-savvy customers grows. In those cases, paying the
from customers (Haddon and Jargon 2019). Platforms platform to bring in customers hurts the restaurant’s
consider themselves as demand generators for restau- profitability.
rants. The head of Uber Eats in North America pointed Second, we prove that this service system could be
out, “We exist for demand generation.” However, res- improved, either through a one-way revenue-sharing
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

taurants feel differently. As the owner of a small res- contract with a price ceiling—in which the platform
taurant chain in California put it, “We saw a direct shares a fraction of its revenue with the restaurant un-
correlation between the delivery services and the re- der the condition that the restaurant caps its food
duction of our income” (Isaac and Yaffe-Bellany prices—or a two-way revenue-sharing contract, in
2019). which the restaurant and platform each shares a pre-
Our paper examines the relationship between a committed fraction of its revenue with the other party.
food delivery platform and a restaurant in a changing These coordinating contracts would effectively force
environment, with an increasing number of online the platform to share profits with the restaurant (com-
customers brought in by the platform. In particular, pared with the equilibrium behavior without such a
we develop a stylized service chain model using contract) to induce food price reduction. This seems
queueing methodology. The restaurant serves food to the opposite of the current practice, where the restau-
customers, modeled as a single-server queue. The rant marks up its regular menu price and shares a frac-
third-party platform with a pool of delivery workers tion of its revenue with the platform as commissions.
offers a food delivery service to customers on the side. We also verify the robustness of these two contracts in
There are two streams of customers: traditional cus- the case where there are no traditional customers.
tomers, who have no access to the platform and can Third, we discover that a delivery-worker pool of a
only walk into the restaurant, and tech-savvy custom- limited size can curb the restaurant’s self-interested de-
ers, who have access to the platform so that they can sire to serve only the growing segment of food-delivery
use the food delivery service in addition to the tradi- customers, avoiding damage to the platform and social
tional walk-in option. The three parties of the restau- welfare. This is because, when the cost of hiring deliv-
rant, platform, and customers participate in a sequential ery workers is high as a result of a tight labor market, in
game, where the restaurant first sets the food price to
anticipation of the successive markup by the platform,
maximize its profit; then the platform determines the
the restaurant finds it unprofitable to charge a higher
delivery fee (and delivery worker’s wage) to maximize
price, serve only the tech-savvy customers, and become
its profit; and last, according to the food price and deliv-
a delivery-only kitchen. There are several implications
ery fee, customers decide whether to walk in or balk,
of this result. Without any coordination with the restau-
or, if they are tech-savvy, to use the food delivery
rant, the platform has a unilateral way to avoid the po-
service.
tential damage to its profits induced by the restaurant’s
If the restaurant has a large enough established
desire to turn into a delivery-only kitchen and reap the
base of traditional customers, we obtain the following
greatest surplus from its food delivery business. Some-
main results.
First, in our base model with infinitely many deliv- what counterintuitively, this requires the platform not
ery workers, we show that, in the sequential game, if to overgrow its labor pool and instead to cap the num-
the food delivery service is sufficiently convenient ber of delivery workers registered with the platform. In
and the number of tech-savvy customers is sufficiently contrast, the social planner who aims at maximizing
large, the restaurant finds it profitable to become a total social welfare including delivery workers’ total
delivery-only kitchen, and then both the platform’s utility also has an implementable and effective way to
profit and social welfare may drop sharply. In this regulate the service system by capping the size of the
case, to squeeze out tech-savvy customers’ residual delivery-worker pool.
surpluses generated by the food delivery service, the
restaurant raises its food price significantly; the price 2. Literature Review
would otherwise be kept lower to accommodate Our paper is related to the extensive literature on
walk-in customers. This hurts the food delivery plat- queueing economics, which dates back to Naor (1969).
form’s profit and social welfare. However, this phe- Hassin and Haviv (2003) and Hassin (2016) provide
nomenon will not occur when the food delivery service comprehensive reviews. We consider a stylized unob-
is not sufficiently convenient, or the tech-savvy customer servable queueing model, which was first studied in
segment is not large enough. Although the platform’s Edelson and Hilderbrand (1975).
profit and social welfare are both (weakly) increasing In the operations management literature, the paper
in the arrival rate of tech-savvy customers, the platform by Feldman et al. (2021) is one of the first to study the
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS 6541

operations of food delivery platforms and the closest delivery-worker pool, which have not been considered
to ours. The authors consider one stream of customers in their paper. Benjaafar et al. (2021) study the labor
who choose between food delivery or dining in. Cus- welfare in on-demand service platforms that crowd-
tomers incur a waiting cost proportional to the total source freelancers to serve customers. They show that
volume of food delivery and dine-in customers. They the labor pool size has a nonmonotonic effect on labor
find that a one-way revenue sharing (RS) contract and welfare. In a different setting, we show that capping
several common modifications, such as commission the labor pool size can effectively suppress the restau-
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

caps and price floors, are ineffective at coordinating rant’s self-interested desire to increase food prices and
the system. Moreover, they show that a generalized can improve the platform’s profit and social welfare.
RS contract, in which the platform pays the restaurant There is an extensive body of literature on supply
a fixed fee and a percentage of revenue, coordinates chain contracting. For reviews of this literature, see
the restaurant and the platform to achieve the optimal Cachon (2003) and Lariviere (2016). Cachon and Lari-
centralized revenue. Our paper examines the platform viere (2005) demonstrate that a RS contract can coordi-
using an unobservable queue model, with two streams nate a supply chain with a single retailer or multiple
of customers: one having access to the food delivery retailers competing in quantities. In a service opera-
service and one not. The unobservable queue formula- tions setting, we demonstrate the effectiveness of a
tion is appropriate when restaurants do not broadcast two-way RS contract and a one-way RS contract with
their real-time queue length to the public, although a price ceiling.
their long-term average waiting time is inferable from Last, our work is related to an emerging literature
their pages on Yelp or Google My Business. We focus on the operations of omnichannel retailing. Gallino
on how the decentralized and centralized systems and Moreno (2014) empirically show that offering a
would change if there were more tech-savvy custom- “buy-online, pick-up-in-store” (BOPS) option reduces
ers, which can shed light on how the industry evolves. online sales but increases offline sales. Gao and Su
We show that a one-way RS contract with a price ceil- (2017a) examine the impact of the BOPS initiative on
ing can coordinate the service chain to maximize the omnichannel store operations and show that BOPS
total profit and that limiting the number of delivery may benefit or hurt the retailer, depending on the
workers available to the platform is an easy-to-imple- product characteristics. Other papers in this line of re-
ment tool to regulate the system. search include Gao and Su (2017b), Gao and Su (2018),
Food delivery services allow customers with high op-
Hu et al. (2021), Yuan and Roet-Green (2020), and Bar-
portunity costs to avoid waiting in a queue by paying
on et al. (2021). Although these papers assume that
delivery workers to pick up and deliver food from res-
the same firm owns both channels, in contrast, we fo-
taurants. A related theme in service systems is priority
cus on the strategic interaction between a restaurant
purchasing. Kleinrock (1967) introduces bribery for a po-
serving offline/online customers and a platform that
sition in a queue, where customers who pay a higher
makes deliveries to online customers.
bribe will be placed ahead of those who pay a lower
bribe. Lui (1985), Glazer and Hassin (1986), and Hassin
(1995) (respectively, Balachandran 1972) analyze the het- 3. Base Model
erogeneous (respectively, homogeneous) customers’ We model a restaurant serving food to customers as a
bribing behavior in an unobservable (respectively, ob- stylized single-server queue under the first-in-first-out
servable) queueing system. Other papers that explore discipline. The customers arrive according to a Pois-
priority pricing of services include Mendelson and son process. The food price is p. The food preparation
Whang (1990), Van Mieghem (2000), Afèche and Men- time for a customer follows an independent and iden-
delson (2004), and Afèche and Pavlin (2016). More re- tically distributed exponential distribution with mean
cently, Lariviere (2020) shows that a priority scheme is 1=µ. Upon getting the food, a customer receives a ser-
superior to the first-in-first-out scheme for both the ser- vice reward R. We do not differentiate between cus-
vice provider’s revenue and social welfare, but priorities tomers who take out the food and those who dine in
often hurt the consumer surplus. the restaurant. Thus, we assume that dine-in, walk-in-
In a context closely related to that of our paper, Cui take-out, and food-delivery orders all generate the
et al. (2020) study the line-sitting service, in which cus- same service reward.
tomers can hire surrogates to wait in line on their be- A third-party platform with a pool of N delivery
half. The authors compare line-sitting with priority workers offers a food delivery service on the side.
purchasing in an unobservable queue model. In con- Any customer can pay the platform a flat delivery fee
trast, we focus on the interaction between the restau- θ to have a delivery worker pick up and deliver food
rant, the platform, and customers in a sequential game. to her doorstep. For simplicity, we assume that one
We investigate the system-wide impacts of a growing delivery worker fulfills at most one order per unit
segment of tech-savvy customers and the size of the of time. This assumption aligns with the one-on-one
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
6542 Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS

delivery practice of some platforms such as Instacart Λ0  0. The case of 0 < Λ0 < µ can be analyzed with
(see, e.g., Deighton and Kornfeld 2017); other plat- much greater complexity but generates insights simi-
forms may combine multiple delivery orders in one lar to the two extreme cases.
trip, so that delivery workers can pick up orders from We assume that the kitchen’s status is unobservable
several restaurants in the same area or deliver to vari- to all customers at the moment when their demand
ous customers in the same neighborhood (see Chen arises. As we study the long-term relationship be-
and Hu 2021 for examples of batch processing in food tween the restaurant and the platform, this assump-
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

delivery). Our model can be easily modified to allow tion captures the first-order interaction by focusing on
for a general maximum number of orders delivered by the customers’ expected service experiences of the sys-
each worker per time unit. However, such a modifica- tem over repeated interactions. As a standard as-
tion may not easily capture same-side or cross-side ex- sumption in the relevant literature and consistent
ternalities among customers and workers; we leave with the focus on repeated interactions, all parameters
this for future research. Each delivery worker’s oppor- are assumed to be common knowledge. There are
tunity cost σ per unit of time follows a probability dis- other information structures for related settings (see,
tribution. In our base model, we assume an ample e.g., Debo and Veeraraghavan 2014, Kremer and Debo
supply of delivery workers, that is, N → ∞, so that 2015, Cui and Veeraraghavan 2016, Wang and Hu
there are sufficient delivery workers to fill all the de- 2020). We leave those alternative information struc-
sired delivery orders. We relax this assumption and tures for future research.
consider a finite delivery-worker pool in Section 6. In The food price p is assumed to be the same for both
our base model, we normalize the delivery workers’ walk-in and food-delivery customers, which is a prac-
opportunity cost to zero (see also Cui et al. 2020). tice of Uber Eats and a common assumption in the
Customers incur a linear waiting cost with the mar- omnichannel literature. We first derive customers’
ginal rate c when they wait in the restaurant for food. equilibrium behavior under the food price p and the
By using the food delivery service, customers do not delivery fee θ. Following Edelson and Hilderbrand
need to physically go to the restaurant, but instead, (1975), we assume that both tech-savvy and traditional
can wait for food at home, and put their time to better customers use a symmetric mixed strategy to choose
use. We assume that the food delivery service does between the options available to them when arriving
not affect the quality of the food as perceived by cus- to the system. The traditional customers’ strategy can
tomers. Let φ ∈ [0, c) denote the customers’ waiting be described by their joining rate λ0W ∈ [0, Λ0 ], where-
cost rate while using the food delivery service. To as tech-savvy customers’ strategy can be expressed by
some extent, the value of φ measures the convenience a tuple of (λD , λ1W ), where λD and λ1W are the joining
of the food delivery service. If the platform provides a rates of food-delivery and walk-in tech-savvy custom-
satisfactory and seamless service, the value of φ will ers, respectively. Let λW  λ0W + λ1W denote the total
be small; otherwise, if the food delivery service is not joining rate of walk-in customers. The effective arrival
that convenient, the value of φ will be large and can rate to the system is λ  λW + λD . As all customers
be close to the offline waiting cost c. need to be served by the kitchen, food preparation is
Not all customers have access to a food delivery ser- the bottleneck, and hence the expected waiting time of
vice at the moment when their demand arises. There all customers for food preparation is 1=(µ − λ). Clear-
are two streams of customers. One stream consists of ly, if Λ1 ≥ µ, because of the lower waiting cost of using
tech-savvy customers who can access a food delivery the food delivery service, the restaurant can obtain a
service (but may not use it); the other stream consists higher profit by serving only food-delivery customers,
of traditional customers who cannot. Let Λ1 and Λ0 de- in which case the restaurant becomes a delivery-only
note the arrival rates of tech-savvy and traditional restaurant, also referred to as a “delivery-only
customers, respectively. When the need for food kitchen” or “ghost kitchen.”
arises, a traditional customer has two options: walk in Under the food price p and the delivery fee θ, the ex-
or balk. The tech-savvy customers are identical to the pected utility of (tech-savvy or traditional) walk-in cus-
traditional customers, except that, when in need of tomers is the service reward less the food price less the
food, they have one more option: using a food deliv- expected waiting cost; that is, UW (λ)  R − p − c=(µ − λ).
ery service if they find this a better choice. These We focus on the food preparation process as the bottle-
tech-savvy customers are the potential customers of neck and assume away the delivery time after the food
the platform. For simplicity, we focus on the case of is made if customers choose the food delivery service.
abundant potential demand from traditional custom- This assumption is innocuous because the expected
ers, that is, Λ0 ≥ µ, which applies to restaurants that waiting cost due to the delivery delay can be factored
have already established a large customer base before into the delivery fee θ and does not qualitatively change
the introduction of food delivery services. In Section our insights. The same assumption is also made in
A of the online appendix, we analyze the case of Feldman et al. (2021). Then the expected utility of
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS 6543

food-delivery customers is the service reward less the regime; an increasing number of tech-savvy customers ben-
food price less the delivery fee less the expected cost of efits both the platform’s profit and social welfare; however,
waiting for food preparation; that is, UD (λ)  R − p − θ it does not increase demand for the restaurant, and instead
−φ=(µ − λ). If customers choose to balk, they receive only changes the composition of customers. Otherwise, if
zero utility. As a tie-breaking rule, we assume that tech- the food delivery service is sufficiently convenient and the
savvy customers use food delivery services if the two number of tech-savvy customers increases to a critical
options of using the service and walking in are equally mass, the restaurant will switch to a delivery-only
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

attractive, that is, UD (λ)  UW (λ) ≥ 0. This rule is innoc- regime—it will raise its food price significantly and serve
uous because the platform can always lower its delivery only food-delivery customers; the platform’s profit, social
fee by an infinitesimal amount to make the food delivery welfare, and the restaurant’s demand will suffer from this
service more attractive with almost no impact on its food-price surge; and if the food delivery service is suffi-
profit. ciently convenient and the number of tech-savvy customers
Our model without a food delivery service (or is sufficiently high, the demand will increase beyond the level
equivalently, without any tech-savvy customers, that it had reached before the introduction of the food delivery ser-
is, Λ1  0) is essentially a classical unobservable queue vice. Formally, there exist φ1 and λ1 such that if (i) φ > φ1
model with sufficient demand, as Λ0 ≥ µ. In this case, or (ii) φ ≤ φ1 and Λ1 ≤ λ1 , we have p∗ and λ∗D + λ∗W stay
the traditional customers’ equilibrium join-up-to level as constants, as Λ1 increases, whereas π∗ , S∗ , λ∗D , and −λ∗W
is µ − c=(R − p) (see, e.g., Edelson and Hilderbrand are weakly increasing in Λ1 . Otherwise, if φ ≤ φ1 , we have
1975), which is decreasing in the food price p. Lemma p∗ | Λ1 ≤λ1 < p∗ | Λ1 >λ1 , π∗ | Λ1 λ1 > π∗ | Λ1 λ1 , and S∗ | Λ1 λ1 >








C.2 in the online appendix presents customers’ equi- S∗ | Λ1 λ1 , and furthermore, if φ < c= (2 Rµ=c− 1) ≤ φ1 ,
librium behavior in our model in the presence of the we have λ∗D + λ∗W | Λ1 ≤λ1 < λ∗D + λ∗W | Λ1 µ .
food delivery service.
We note that, ceteris paribus, food-delivery custom-
4. Decentralized System ers generate more surplus than walk-in ones, because
In this section, we study a decentralized system where of their lower marginal waiting cost. The restaurant
the restaurant and the platform are operated indepen- can either set a high food price to extract more surplus
dently to maximize their own profits. The interaction from food-delivery customers while abandoning
of the restaurant, the platform, and two streams of walk-in customers (i.e., in the delivery-only regime),
customers forms a sequential game. Consistent with or it can use a low food price to serve more (potentially
the supply chain contracting literature, the sequence walk-in) customers but leave the food-delivery
of events is as follows. First, the restaurant sets the customers’ extra surplus for the platform to reap
food price p. Then, the platform sets the delivery fee (i.e., in the delivery-irrelevant regime). The restau-
θ. At last, customers decide whether to join the queue rant’s optimal choice depends on the number of tech-
savvy customers and the convenience of the food
or balk, and if they join and are tech-savvy, whether
delivery service (or equivalently the waiting cost while
to use the food delivery service or to walk in, based
using food delivery service φ), which determine
on the food price p, delivery fee θ and their prior be-
the food-delivery customers’ residual surplus. When
lief about the wait; in equilibrium, their prior belief is
the food delivery service is convenient enough and the
consistent with actual experiences over repeated inter-
number of tech-savvy customers is sufficiently large,
actions. Given the food price p and the delivery fee θ,
the restaurant will be more profitable focusing only on
the equilibrium joining behavior by customers at the
food-delivery customers and staying in the delivery-
third stage is characterized in Lemma C.2 in the online
only regime; otherwise, it will prefer operating in the
appendix. In the base model, the delivery-worker
delivery-irrelevant regime.
pool size N is large enough that in equilibrium any
We then illustrate Proposition 1 numerically. Specif-
tech-savvy customer who is willing to pay for the
ically, we display the following equilibrium measures:
food delivery service can obtain it.
(i) the restaurant’s profit, the platform’s profit, and the
We use backward induction to derive the equilibrium
resulting social welfare; (ii) the food price and the de-
food price p∗ and delivery fee θ∗ . Then we derive the
livery fee; and (iii) the joining rates of food-delivery
restaurant’s profit Π∗ , the platform’s profit π∗ , and so-
and walk-in customers, and the resulting throughput,
cial welfare S∗ (i.e., the total surpluses from all stake-
in equilibrium, as a function of the arrival rate of tech-
holders) in equilibrium.
savvy customers Λ1 , for φ  0:4, 0.3, and 0.1 in Figures
Proposition 1 (Equilibrium in Decentralized System). 1–3, respectively. (In our setting with two segments of
If the food delivery service is sufficiently inconvenient or homogeneous traditional and tech-savvy customers,
the number of tech-savvy customers is relatively low, the in equilibrium, all customers’ surpluses will be ex-
restaurant will not react to the introduction of the food de- tracted by the food price and/or delivery fee.) Under
livery service and will operate in a delivery-irrelevant this setting where R  10 and Λ0  µ  c  1, we have
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
6544 Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS

the threshold values in Proposition 1 as φ1 ≈ 0:3554 ∈








general and independent of whether there is a con-
(0:3, 0:4) and c=(2 Rµ=c − 1) ≈ 0:1878. tract between the restaurant and the platform. Thus,
When the food delivery service is not so convenient the food delivery service does not necessarily increase
(see, e.g., φ > φ1 in Figure 1) or the arrival rate of tech- demand for the restaurant, especially when the restau-
savvy customers is not high (see, e.g., φ ≤ φ1 and Λ1 ≤ rant has sufficient demand from traditional customers
0:55 in Figure 2), the restaurant does not react to the and the food price stays the same. The introduction
introduction of the food delivery service, so it applies of the food delivery service may simply change the
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

the same strategy as in an unobservable queue with composition of customers—it brings in more food-
only traditional customers for various Λ1 . However, delivery orders while driving away traditional cus-
when the food delivery service is relatively conve- tomers who used to walk in. This phenomenon occurs
nient (see, e.g., φ ≤ φ1 in Figure 2), if the arrival rate of even when the number of traditional customers is un-
tech-savvy customers Λ1 increases to a critical level changed, and will be more likely to occur as more tra-
λ1, it becomes more beneficial for the restaurant to ca- ditional customers become tech-savvy over time.
ter completely to the food-delivery customers. Then, Proposition 1 identifies two conditions under which
the restaurant will abruptly increase its food price p∗ the food delivery service does not benefit the restau-
(see, e.g., the dash-dotted curve in Figure 2(b)) to rant: either the food delivery service is not sufficiently
drive away traditional customers all at once (see, e.g., convenient, or it is but the number of tech-savvy cus-
the densely dotted curve in Figure 2(c)) and become a tomers is not large enough. Under either of the two
delivery-only kitchen. In response, the platform has to conditions, the restaurant does not benefit from the
lower the delivery fee θ (see, e.g., the dashed curve in food delivery service by obtaining a higher through-
Figure 2(b)). As a result, the platform’s profit and so- put. As a result, if the restaurant pays the platform for
cial welfare drastically drop in the neighborhood, bringing in customers as seems to be the current prac-
whereas the restaurant’s profit weakly increases, as tice by platforms such as Postmates, having more
shown in Figure 2(a). food delivery orders actually hurts the restaurant’s
When the arrival rate of tech-savvy customers Λ1 profitability. Isaac and Yaffe-Bellany (2019) docu-
increases, if the food price stays unchanged and the mented such an instance. After offering delivery
restaurant accommodates both the food-delivery and through a platform in 2016, two pizzerias with the
walk-in customers, the throughput at the restaurant same owner, whom we quoted in the introduction,
stays constant (see, e.g., Λ1 ≤ 0:68 in Figure 1(c) took a sharp turn from generating annual profits of
and Λ1 ≤ 0:55 in Figure 2(c)). When the arrival rate of $50,000 to $100,000 to losing $40,000 a year. This was
tech-savvy customers Λ1 increases while that of tradi- precisely because customers who used to order directly
tional customers Λ0 stays unchanged, the platform from the pizzerias switched to the platform, which
may bring in more food-delivery customers, but the forced the owner to pay commissions. The business
traditional customers will join less often so that the owner said, “It was like death by a thousand cuts,”
throughput rate stays at µ − c=(R − p), which is the and he eventually closed these two locations.
throughput in a classical unobservable queue with The food delivery service may increase demand at
only traditional customers under price p, and their ex- the restaurant compared with the scenario of no food
pected utility is zero. We note that this insight is delivery service, under a stricter condition than that

Figure 1. (Color online) Decentralized Equilibrium System Behavior as a Function of Λ1 for R  10, Λ0  µ  c  1, and φ  0:4

(a) (b) (c)

Social Welfare and Profits Prices Joining Rates
10 12 1

8 10 0.8
8
6 0.6
6
4 0.4
4
2 2 0.2

0 0 0
0 0.5 1 0 0.5 1 0 0.5 1
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS 6545

Figure 2. (Color online) Decentralized Equilibrium System Behavior as a Function of Λ1 for R  10, Λ0  µ  c  1, and φ  0:3

(a) (b) (c)

Social Welfare and Profits Prices Joining Rates
10 12 1

8 10 0.8
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

8
6 0.6
6
4 0.4
4
2 2 0.2

0 0 0
0 0.5 1 0 0.5 1 0 0.5 1

for a delivery-only regime: the arrival rate of tech- platform’s profitability. Thus, the platform or the so-
savvy customers becomes significantly high and the cial planner may want to coordinate actions by the
food delivery service is highly convenient (see, e.g., restaurant and platform as the number of food deliv-
Λ1 > 0:6838 in Figure 3(c) where φ  0:1). The restau- ery orders grows, which motivates us to explore ap-
rant’s optimal choice of turning away traditional cus- proaches to coordinating the service system.
tomers and operating in a delivery-only regime makes As a remark, in our base model, we assume abun-
it more difficult for the platform to raise demand for dant demand from traditional customers, that is,
the restaurant—the potential arrival rate of tech-savvy Λ0 ≥ µ, so an increase in food-delivery customers will
customers needs to rise significantly beyond the criti- inevitably turn away some traditional customers. In
cal mass at which the restaurant would switch to the practice, there are also cases with no traditional cus-
delivery-only regime. It may not be a simple task for tomers, that is, Λ0  0. For example, think about new
the platform to prove its helpfulness in demand crea- restaurants without an established customer base or
tion to the restaurant. restaurants during the pandemic that were barred
Our result here suggests that the no-contract rela- from having dine-in customers. For this case, see
tionship, that is, the decentralized system studied in Proposition A.1 in the online appendix. There we
this section, may not be ideal, especially when the show that the restaurant prefers the delivery-only re-
food delivery service is convenient and the arrival gime to the delivery-irrelevant one, especially when
rate of tech-savvy customers is high. In those circum- the arrival rate of tech-savvy customers Λ1 is low—
stances, the increasing demand rate of tech-savvy the restaurant will not be able to attract more custom-
customers may jeopardize social welfare and the ers by setting a low food price, and it can only focus

Figure 3. (Color online) Decentralized Equilibrium System Behavior as a Function of Λ1 for R  10, Λ0  µ  c  1, and φ  0:1

(a) (b) (c)

Social Welfare and Profits Prices Joining Rates
10 12 1

8 10 0.8
8
6 0.6
6
4 0.4
4
2 2 0.2

0 0 0
0 0.5 1 0 0.5 1 0 0.5 1
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
6546 Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS

on extracting more residual surplus from food- Moreover, in observable queues, the social planner’s
delivery customers by charging a high food price. incentive does not align perfectly with the profit
Moreover, the platform that connects a restaurant maximizer’s—the social welfare maximizing price
with no traditional customers to a growing pool of is greater than the profit maximizing price (see chap-
tech-savvy customers will certainly increase demand ter 2 of Hassin and Haviv 2003). We expect the
for the restaurant. In this case, the platform will be able same misalignment in the observable version of our
to fulfill its stated mission of demand creation. Howev- model.
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

er, there are other much-debated issues such as high To illustrate the analytical results of Lemma 1, we
commissions charged by food delivery platforms (see, plot the optimal performance measures as a function
e.g., Tkacik 2020) that we leave for future research. of the arrival rate of tech-savvy customers Λ1 in
Figure 4 (similar to what we have in Figure 2). Lemma
5. Centralized System 1 shows that the optimal throughput increases in the
In this section, we consider the profit maximization arrival rate of tech-savvy customers Λ1 . As we dis-
problem of controlling the food price and delivery fee cussed in Section 3, if the food price p stays the same
when Λ1 increases, the traditional customers will join
from the perspective of a centralized owner of the
less often so that the throughput stays at µ − c=(R − p).
food catering system.
Thus, when Λ1 increases, the centralized owner needs
Lemma 1 (Optimal Monopoly Prices). If the restaurant to lower the food price p to attract more traditional
sets the optimal monopoly food price po , the platform will customers (see, e.g., the dash-dotted curve in Figure
choose the optimal monopoly delivery fee θo as its best re- 4(b)). At the same time, when the food price is re-
sponse. (The expressions of po and θo are given by (C.3) duced, the food-delivery customers have more residu-
and (C.4) in the online appendix.) Moreover, the optimal al surplus, so the centralized owner needs to raise the
monopoly food price po, the total price po + θo , and the cor- delivery fee θ to reap it from food-delivery customers
responding restaurant’s profit Πo are weakly decreasing in (see, e.g., the dashed curve in Figure 4(b)). Here, the
Λ1 . The optimal monopoly delivery fee θo, the correspond- social planner may use delivery fees other than θo—as
ing throughput λo D + λo W , the platform’s profit πo, and so- long as all tech-savvy customers use the food-
cial welfare So are weakly increasing in Λ1 . delivery service, the delivery fee only changes the
distribution of surpluses between the platform and
From Lemma 1, we see that it is not necessary for
customers, whereas social welfare stays the same. In
the centralized owner to dictate both the food price general, the centralized owner reduces the total price
and delivery fee. Instead, the centralized owner can p + θ paid by the food-delivery customers when Λ1
achieve the monopolistically optimal solution by reg- increases, because of the congestion caused by a
ulating only the food price. Under the optimal mo- higher throughput (see, e.g., the solid curves in Fig-
nopoly food price, the platform will voluntarily set ure 4, (b) and (c)).
the delivery fee at the centrally optimal level. The rea- Lemma 1 also sheds light on the coordination of the
son is that, given the optimal monopoly food price, restaurant’s and the platform’s operations in the decen-
the centralized owner’s goal is to set the delivery fee tralized system. There are two important features in
to extract the maximum surplus from food-delivery any coordinating scheme that can maximize the aggre-
customers, which is also the platform’s goal in a de- gated profit of the restaurant and the platform. The first
centralized system. is, as mentioned in the centralized solution, when the
The optimal monopoly food price po and delivery arrival rate of tech-savvy customers Λ1 increases, the
fee θo maximize not only the aggregated profit but restaurant needs to set the food price at po, which de-
also social welfare. Because of our unobservable creases in Λ1 , to attract more customers to purchase
queue assumption and customers’ homogeneity in from the restaurant. Then, the platform’s best response
their service reward and marginal waiting cost, the is to set the delivery fee at θo, which increases in Λ1 ,
centralized owner can extract all customer surplus as to extract the residual surplus generated by having a
profit by setting the food price and delivery fee (see lower food price (see, e.g., Figure 4(b)). As a result, the
chapter 3 of Hassin and Haviv 2003 for a single- restaurant’s direct profit from food sales decreases while
segment problem). Thus, the centralized owner’s goal the platform’s profit from the delivery service increases
of maximizing the aggregated profit aligns perfectly as the segment of tech-savvy customers Λ1 grows (see,
with the social planner’s goal of maximizing social e.g., Figure 4(a)). The second feature is that it is essential
welfare; however, when setting the food price at po, for the platform to share its delivery profit with the res-
the social planner can also achieve various surplus taurant to incentivize the restaurant’s participation in
distributions between the platform and customers by this coordinating effort. Our result suggests that the cur-
varying the delivery fee, which is different from that un- rent practice—that the restaurant marks up its regular
der the profit maximization by the centralized owner. menu price and shares a portion of its revenue with the
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS 6547

Figure 4. (Color online) Centralized Optimal System Behavior as a Function of Λ1 for R  10, Λ0  µ  c  1, and φ  0:3

(a) (b) (c)

Social Welfare and Profits Prices Joining Rates
10 15 1.5

8
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

10 1
6

4
5 0.5
2

0 0 0
0 0.5 1 0 0.5 1 0 0.5 1

platform by paying a commission—does not seem able Without this price ceiling, the food price chosen by
to coordinate the service system. the self-interested restaurant will surpass po, which
Next, we propose two RS contracts with the previ- leads to a suboptimal total revenue for the whole ser-
ous two features that can perfectly coordinate the ser- vice chain. Different from Feldman et al. (2021) who
vice system. Of course, to negotiate and enforce such show that a one-way RS contract cannot achieve the
a RS contract calls for the engagement of both parties. profit of a centralized system and often performs
The restaurant’s and the platform’s profit levels in a worse for the restaurant than having no delivery ser-
decentralized system—shown in Proposition C.4 and vice, we show that a one-way RS contract with a price
Corollary C.3 in the online appendix—represent the ceiling—a mild modification—can coordinate the sys-
lower bounds of both parties’ profit targets in a RS tem to achieve the maximum aggregated profit. On
contract. Then their bargaining power can determine the other hand, in the two-way RS contract, neither
how they will divide the extra surplus generated food price nor delivery fee is specified, and both par-
through this coordination. ties just agree to share part of their revenue with each
other so that each party earns a given fraction of the
Proposition 2 (RS Contracts). The following RS con-
total revenue. Under this contract, the restaurant’s
tracts can coordinate the system and achieve the maximum
profit becomes an affine transformation of the aggre-
aggregated profit.
gated profit of the service system, so the restaurant
(i) One-way RS contract with a price ceiling—The plat-
will set the optimal monopoly food price po. Under
form allocates a fraction γ1 of its revenue to the restaurant
both contracts, the platform promises to share a frac-
while the restaurant cannot set a food price higher than po.
tion of its profit with the restaurant, which enforces
(ii) Two-way RS contract—Both the restaurant and the
the latter’s engagement in such coordinating
platform agree that a fraction γ2 of their aggregated revenue
contracts.
be allocated to the restaurant.
Proposition 2 specifies closed-form expressions of
There always exists a range of sharing fractions that make
the ranges of sharing fractions that can achieve a win-
both parties weakly better off than they would be without any
win situation for both the restaurant and platform. In
contract. In particular, the sharing fractions in (C.5) (respec-
general, such a range can be obtained by taking ratios
tively, (C.6)) of the online appendix achieve a win-win for
of the profit levels of the restaurant and platform in
both parties under the one-way RS contract with a price ceil-
the decentralized system over the centralized total
ing (respectively, two-way RS contract).
profit.
Proposition 2 provides two contracts to coordinate We close this section with two remarks. First, we
the system and arbitrarily share the maximized total also verify the robustness of the results of Lemma 1
profit. In the proposed one-way RS contract, the food and Proposition 2 under the condition of no traditional
price is capped at po . Because the restaurant’s
  profit customers, that is, Λ0  0 (see Lemma A.1 and Proposi-
increases with the food price p on 0, po in the de- tion A.2 in the online appendix). The same insights
centralized system (see Section C.3 in the online ap- hold. Second, in our base model, we assume that the
pendix), the restaurant will self-interestedly set the restaurant is unable to price-differentiate between
food price at po. walk-in and food-delivery customers. Thus, the food
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
6548 Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS

price stays the same for all purchases. However, if the Λ̄ T such that if N ≤ N̄, in equilibrium the restaurant’s food








restaurant was allowed to set different food prices for price p∗ and profit Π∗ stay at p∗ (Λ1 )  R − Rc=µ and
food-delivery and walk-in customers as po + θo and po , 



√

Π∗ (Λ1 )  ( Rµ − c)2 for all Λ1 . If N > N̄ and Λ1 > Λ̄ T ,
where po and θo are given in Lemma 1, the restaurant in equilibrium the restaurant sets a food price
could extract all social welfare as profit—the platform 







p∗ (Λ1 ) ≥ R − Rc=µ.
and customers would have zero surpluses. Note that,
in this case, the platform would still provide a food de- In the decentralized system with a finite delivery-
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

livery service to tech-savvy customers, because the worker pool, Proposition 3 shows that, as in the base
cost of hiring delivery workers is assumed to be zero model, the restaurant has two operating regimes: (i)
in the base model. the delivery-irrelevant regime, where the restaurant
behaves the same as in an unobservable queue with
6. Finite Delivery-Worker Pool only traditional customers, and (ii) the delivery-only
In this section, we extend our base model with an infi- regime, where the restaurant becomes a delivery-only
nite pool of delivery workers by assuming that the kitchen and serves only the food-delivery customers.
platform has a delivery-worker pool of size N < ∞. For the restaurant, the delivery-worker pool size N af-
We assume that the delivery workers’ opportunity fects the maximum profit in the delivery-only regime
 
cost per unit of time σ follows a distribution on 0, β but has no impact in the delivery-irrelevant regime.
with the cumulative distribution function denoted by For the platform, when the delivery-worker pool size
F(·). When the  platform
 sets the guaranteed delivery N increases, it becomes less costly to hire delivery
wage at w ∈ 0, β per unit of time, the expected supply workers to fulfill the same unconstrained demand for
of delivery workers is ν(w)  N · F(w). the food delivery service from the tech-savvy custom-
As with the base model, we study a decentralized ers, which benefits the platform for a fixed food price.
system where the restaurant and the platform partici- However, when N reaches a certain level N̄ and there
pate in a Stackelberg game. The setting is exactly the is a sufficiently number of tech-savvy customers, the
same as in the base model, except that in the second restaurant finds it more beneficial to shift to the
stage, the platform decides on and posts a wage w for delivery-only regime. That is, the restaurant will raise
delivery workers together with the delivery fee θ for the food price significantly and squeeze more residual
customers. The game can be solved using backward surplus from the platform, which can cause the plat-
induction as well. Recall our assumption that one de- form’s profit to drop sharply, in spite of the benefit of
livery worker fulfills at most one order per unit of accessing a cheaper labor pool. This phenomenon is
time. Lemma C.2 characterizes the tech-savvy cus- similar to what has been shown in the case of the infi-
tomers’ demand λD , unconstrained by the supply of nite delivery-worker pool in our base model (see, e.g., Fig-
delivery workers, for the food delivery service under ure 2(a)). However, before N increases to this threshold
delivery fee θ. The joining rate of food-delivery cus- of N̄, the restaurant does not find it beneficial to make
tomers is the minimum of demand and supply of such a regime shift.
delivery workers, that is, min (λD , ν(w)). The plat- Proposition 3 implies that, under no contract be-
form’s revenue is the product of the delivery fee and tween the restaurant and platform, the delivery-
the joining rate of food-delivery customers, that is, worker pool size may be a useful lever for the
θ · min (λD , ν(w)). The platform pays w · ν(w) per unit platform and social planner. By limiting the size of
of time to those ν(w) numbers of delivery workers. this pool, the platform can limit the restaurant’s ten-
Thus, under delivery fee θ and delivery wage w, the dency to become a delivery-only kitchen, a situation
platform’s profit is π p, θ, w) θ · min (λD , ν(w))− that may hurt the platform’s profit and social welfare
w · ν(w), and the delivery
 w workers’ total utility is because of the potential price increased by the restau-
uD (θ, w)  w · ν(w) − N x dF(x), which is part of so- rant. (Section B.1 of the online appendix provides a
0
cial welfare in this extension. We characterize the equi- more detailed discussion of the impact of the
delivery-worker pool size N on social welfare when
librium behavior as follows.
there is sufficient demand from tech-savvy customers.
Proposition 3 (Impact of Delivery-Worker Pool Size Social welfare could be hurt by a larger delivery-
on Restaurant). Consider the decentralized system. When worker pool despite more workers getting hired, be-
the delivery-worker pool size is capped at a specific level, the cause of the possible price increase by the restaurant.)
restaurant operates in a delivery-irrelevant regime, with no More importantly, by limiting the delivery-worker
response to the introduction of the food delivery service. pool size, the platform may reap all surpluses from
Otherwise, the restaurant operates in a delivery-only re- the introduction of the food delivery service.
gime when the tech-savvy segment is sufficiently large. For- Proposition 3 also provides the social planner with
  2when σ follows the uniform distribution over
mally, a simple yet effective approach to improving social
0, β , µ β ≤ cN and φ ≤ φ1 , there exist thresholds N̄ and welfare. Recall that, in Section 5, we demonstrate the
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS 6549

practicality of a RS contract to maximize social wel- the restaurant may find it more profitable to cater
fare. However, the RS contract needs participation only to the food-delivery customers and will raise its
from both the restaurant and platform. Negotiating food price significantly to extract more surpluses from
and enforcing such a contract may be costly. Alterna- them. This action hurts the platform’s profit. On the
tively, the social planner can regulate the number of other hand, this will not happen when the number of
delivery workers registered with the platform to im- delivery workers is limited. In such a situation, the
prove social welfare, especially when the demand rate platform’s capacity to provide the food delivery ser-
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

of tech-savvy customers is high enough and the food vice is capped, so the restaurant cannot benefit from
delivery service is sufficiently convenient. Although concentrating only on tech-savvy customers. Similar
such regulation does not lead to optimal social wel- results hold for social welfare, for the same reason.
fare, it generates a reasonably good outcome and can These results have several implications. When there
be easy to implement—it aligns with the platform’s in- is no contract between the restaurant and itself, the
terest, so the platform is less likely to push back. platform may not gain from a larger potential market
In general, the closed-form solution of the sequen- for the food delivery service. If the arrival rate of tech-
tial game under a general opportunity cost distribu- savvy customers is high, it may be more beneficial for
tion may not exist. We adopt a numerical approach to the platform to limit the number of delivery workers
deriving all desired performance measures in Section so that the restaurant does not have the incentive to
B of the online appendix (for opportunity costs follow- serve only food-delivery customers, which will hurt
ing uniform and beta distributions) and demonstrate the platform’s profit. More importantly, such a cap by
that our insights continue to hold for various opportu- the platform can leave all the extra surpluses generated
nity cost distributions. by the food delivery service to the platform itself.
Under the same condition of no contract between
7. Conclusion the restaurant and the platform, the social planner
The food delivery service is an innovative and eco- may not prefer a high arrival rate of tech-savvy
nomically sensible business model that allows people customers, who can use the food delivery service
with high opportunity costs to outsource the task of to reduce their waiting costs. The social planner
waiting in line to people with low opportunity costs can improve social welfare by regulating the
to reduce their own waiting cost. The recent explosion delivery-worker pool size to curb the restaurant’s
of information technology has enabled the food deliv- interest in keeping the food price high. Finally, our
ery business to expand so that more people have study generates insights into the contracts that co-
access to it. In this paper, we study a stylized single- ordinate a food catering system and maximize the
server restaurant with a third-party platform provid- total profit obtained by the restaurant and the plat-
ing a food delivery service to customers on the side. form. When the number of tech-savvy customers
There are two streams of customers. The traditional increases, a coordinating contract will incentivize
customers cannot access this food delivery service, the restaurant to reduce its food price and thereby
while the tech-savvy customers can. The interplay of attract more orders. At the same time, the platform
the four parties—the restaurant, the food delivery will raise its delivery fee to extract the residual sur-
platform, the traditional and tech-savvy customers— pluses from food-delivery customers. Hence, it is
forms a sequential game. We solve the game analyti- essential for the platform to share its food delivery
cally for decentralized and centralized systems, under profit with the restaurant, not the other way
the condition of abundant traditional customers that around, so that the restaurant is motivated to take
applies to restaurants with a strong existing customer part in the coordination.
base. In practice, restaurants can set up in-house food de-
We discover that the platform does not necessarily livery services or use third-party food delivery plat-
increase demand for the restaurant, especially when forms. Before delivery apps were introduced, some
the food price remains unchanged and there are suffi- restaurants, especially pizza restaurants, had their
cient traditional customers. Then, if the restaurant has own in-house food delivery services. Now even those
to pay the platform for bringing in customers, the pizza businesses are split on whether to work with
more orders made through the platform, the greater apps like Uber Eats (Melton 2021). By going with a
loss the restaurant may incur. Moreover, we show food delivery platform, a restaurant can lower its de-
that when the pool of delivery workers is large, the livery cost and enjoy the otherwise expensive-to-build
platform’s profit may decline sharply when the num- digital infrastructure, thanks to the provision of re-
ber of tech-savvy customers, which represents the po- sources such as drivers and capital by the platform.
tential market for the food delivery service, increases But there are also downsides in working with delivery
to a critical level. This happens because, when the ar- platforms. First, restaurants with their own drivers
rival rate of tech-savvy customers is sufficiently high, can provide faster average delivery times thanks to
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
6550 Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS

dedicated services, particularly in the suburbs, where Technological University, and Shanghai University of
delivery platforms still lack driver networks. Second, Finance and Economics.
because of the current practice of the platform charg-
ing a commission for each order, which gives restau- References
Afèche P, Mendelson H (2004) Pricing and priority auctions in
rants an incentive to mark up their menu prices on the
queueing systems with a generalized delay cost structure. Man-
app, eaters can find it more expensive to order from agement Sci. 50(7):869–882.
the app than directly from the restaurant. After all,
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

Afèche P, Pavlin JM (2016) Optimal price/lead-time menus for

restaurants need to balance the pros and cons. For queues with customer choice: Segmentation, pooling, and stra-
those that already maintain an in-house fleet, the plat- tegic delay. Management Sci. 62(8):2412–2436.
Balachandran KR (1972) Purchasing priorities in queues. Manage-
forms can be used contingently, for example, when
ment Sci. 18(5):319–326.
in-house drivers are busy. Regardless of whether a Baron O, Chen X, Li Y (2021) Omnichannel services: The false pre-
restaurant has dedicated drivers, our results shed mise and operational remedies. Preprint, submitted September
light on the forms of coordinating contracts when the 22, https://dx.doi.org/10.2139/ssrn.3444772.
restaurant is dealing with food delivery platforms. Benjaafar S, Ding J, Kong G, Taylor T (2021) Labor welfare in
on-demand service platforms. Manufacturing Service Oper. Man-
Our model has limitations. First, we assume that the
agement, ePub ahead of print April 6, https://doi.org/10.1287/
restaurant uses the same unchanged capacity to cope msom.2020.0964.
with the growing market of tech-savvy customers. Cachon GP (2003) Supply chain coordination with contracts.
Winkler and Jones (2019) report Uber’s cofounder’s Graves SC, de Kok AG, eds. Handbooks in Operations Re-
new startup CloudKitchens—a bet on the food- search and Management Science, vol. 11 (Elsevier, Amster-
dam), 227–339.
delivery boomlet. The firm buys cheap real estate, Cachon GP, Lariviere MA (2005) Supply chain coordination with
often near city centers, and builds delivery-only kitch- revenue-sharing contracts: Strengths and limitations. Manage-
ens to rent to restaurants that can prepare food ex- ment Sci. 51(1):30–44.
clusively for delivery. The attempt to add such an Chen M, Hu M (2021) Courier dispatch in on-demand delivery. Pre-
industrial-production component to food delivery print, submitted November 6, https://dx.doi.org/10.2139/ssrn.
3675063.
aims to solve key logistics hurdles by creating capacity Cui S, Veeraraghavan S (2016) Blind queues: The impact of con-
that is not tethered to restaurants that also serve dine-in sumer beliefs on revenues and congestion. Management Sci. 62
customers. However, implementing this idea still faces (12):3656–3672.
many practical challenges. Consideration of the restau- Cui S, Wang Z, Yang L (2020) The economics of line-sitting. Manage-
rant’s existing dine-in capacity and potentially new ment Sci. 66(1):227–242.
Debo L, Veeraraghavan S (2014) Equilibrium in queues under un-
delivery-only capacity can be a fruitful direction to ex- known service times and service value. Oper. Res. 62(1):38–57.
plore. Second, we assume that customers have a ho- Deighton J, Kornfeld L (2017) Instacart and the new wave of grocery
mogeneous waiting cost while waiting for food offline. startups. Harvard Business School Case 515-089, Cambridge,
Third, our model does not consider the cross-side or MA.
Edelson N, Hilderbrand D (1975) Congestion tolls for poisson queu-
same-side externalities among customers and drivers,
ing processes. Econometrica 43(1):81–92.
for example, the externalities generated through spa- Feldman P, Frazelle AE, Swinney R (2021) Managing relationships
tial pooling that the delivery fee may hinge on. Last, between restaurants and food delivery platforms: Conflict, con-
we assume that if customers choose food delivery, tracts, and coordination. Preprint, submitted July 30, http://dx.
there is no delivery delay after the food is made, and doi.org/10.2139/ssrn.3258739.
Gallino S, Moreno A (2014) Integration of online and offline chan-
no quality difference in food between offline and on-
nels in retail: The impact of sharing reliable inventory availabil-
line orders. Our results may change (respectively, ity information. Management Sci. 60(6):1434–1451.
should still qualitatively hold) when the loss of food Gao F, Su X (2017a) Omnichannel retail operations with buy-online-
quality with online orders is significant (respectively, and-pick-up-in-store. Management Sci. 63(8):2478–2492.
minor). We leave relaxing those assumptions and en- Gao F, Su X (2017b) Online and offline information for omnichannel
retailing. Manufacturing. Service Oper. Management 19(1):84–98.
riching the current model for future research. Gao F, Su X (2018) Omnichannel service operations with online and
offline self-order technologies. Management Sci. 64(8):3595–3608.
Acknowledgments Glazer A, Hassin R (1986) Stable priority purchasing in queues.
The authors thank Charles Corbett (department editor), Oper. Res. Lett. 4(6):285–288.
the associate editor, and two anonymous reviewers for Haddon H, Jargon J (2019) Consumers love food delivery. Restau-
valuable comments and suggestions on the manuscript rants and grocers hate it. Wall Street Journal (March 9),
throughout the whole review process. The authors also https://www.wsj.com/articles/consumers-love-food-delivery-
restaurants-and-grocers-hate-it-11552107610.
thank Chien-Ming Chen, Youhua Chen, Geoffrey Chua,
Hassin R (1995) Decentralized regulation of a queue. Management
Pengfei Guo, Albert Ha, Qing Li, S. Viswanathan, Wen- Sci. 41(1):163–173.
bin Wang, Zhenzhen Yan, Man Yu, Yimin Yu, Hongtao Hassin R (2016) Rational Queueing (CRC Press, Boca Raton, FL).
Zhang, Rachel Zhang, Shaohui Zheng, and the seminar Hassin R, Haviv M (2003) To Queue or not to Queue: Equilibrium Be-
attendants at City University of Hong Kong, Hong havior in Queueing Systems. International Series in Operations
Kong University of Science and Technology, Nanyang Research and Management Science (Springer, New York).
 Chen, Hu, and Wang: Food Delivery Service and Restaurant: Friend or Foe?
Management Science, 2022, vol. 68, no. 9, pp. 6539–6551, © 2022 INFORMS 6551

Hu M, Xu X, Xue W, Yang Y (2021) Demand pooling in omnichan- Naor P (1969) The regulation of queue size by levying tolls. Econo-
nel operations. Management Sci., ePub ahead of print March 23, metrica 37(1):15–24.
https://doi.org/10.1287/mnsc.2021.3964. Steingoltz M, Picciola M (2019) Meals on wheels: The digital order-
Isaac M, Yaffe-Bellany D (2019) The rise of the virtual restaurant. ing and delivery restaurant revolution. Accessed December 22,
New York Times (August 14), https://www.nytimes.com/2019/ 2021, https://www.lek.com/insights/ei/digital-restaurant-delivery.
08/14/technology/uber-eats-ghost-kitchens.html. Tkacik M (2020) Restaurants are barely surviving. Delivery apps
Kleinrock L (1967) Optimum bribing for queue position. Oper. Res. will kill them. Washington Post (May 29), https://www.
15(2):304–318. washingtonpost.com/outlook/2020/05/29/delivery-apps-
Downloaded from informs.org by [140.116.206.221] on 28 August 2023, at 22:20 . For personal use only, all rights reserved.

Kremer M, Debo L (2015) Inferring quality from wait time. Manage- restaurants-coronavirus.
ment Sci. 62(10):3023–3038. Van Mieghem J (2000) Price and service discrimination in queuing
Lariviere MA (2016) Supply chain contracting: Doughnuts to bub- systems: Incentive compatibility of Gcµ scheduling. Manage-
bles. Manufacturing Service Oper. Management 18(3):309–313. ment Sci. 46(9):1249–1267.
Lariviere MA (2020) Priority queues and consumer surplus. Preprint, Wang J, Hu M (2020) Efficient inaccuracy: User-generated informa-
submitted July 30, https://dx.doi.org/10.2139/ssrn.3241559. tion sharing in a queue. Management Sci. 66(10):4648–4666.
Lui F (1985) An equilibrium queuing model of bribery. J. Political Winkler R, Jones R (2019) Meet Travis Kalanick’s Secret Startup,
Econom. 93(4):760–781. CloudKitchens. Wall Street Journal (November 7), https://www.wsj.
Melton A (2021) The pizza business is divided on delivery. Wall com/articles/meet-travis-kalanicks-secret-startup-cloudkitchens-
Street Journal (May 30), https://www.wsj.com/articles/the- 11573122602.
pizza-business-is-divided-on-delivery-11622367002. Yuan Y, Roet-Green R (2020) Information visibility in omnichannel
Mendelson H, Whang S (1990) Optimal incentive-compatible priori- queues. Preprint, submitted August 7, https://dx.doi.org/10.
ty pricing for the M/M/1 queue. Oper. Res. 38(5):870–883. 2139/ssrn.3485810.