Analysis of Sponsored Data in the Case of Competing

Wireless Service Providers

Patrick Maillé, Bruno Tuffin

To cite this version:

Patrick Maillé, Bruno Tuffin. Analysis of Sponsored Data in the Case of Competing Wireless Ser-
vice Providers. NETGCOOP 2018 - 9th International Conference on Network Games, Control and
Optimization, Nov 2018, New York City, United States. pp.1-16. �hal-01670139v2�

HAL Id: hal-01670139

https://hal.inria.fr/hal-01670139v2
Submitted on 12 Sep 2018

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Analysis of Sponsored Data Practices in the Case
of Competing Wireless Service Providers

Patrick Maillé and Bruno Tuffin

Abstract With wireless sponsored data, a third party, content or service provider,
can pay for some of your data traffic so that it is not counted in your plan’s monthly
cap. This type of behavior is currently under scrutiny, with telecommunication regu-
lators wondering if it could be applied to prevent competitors from entering the mar-
ket, and what the impact on all telecommunication actors can be. To answer those
questions, we design and analyze in this paper a model where a Content Provider
(CP) can choose the proportion of data to sponsor and a level of advertisement to get
a return on investment, with several Internet Service Providers (ISPs) in competi-
tion. We distinguish three scenarios: no sponsoring, the same sponsoring to all users,
and a different sponsoring depending on the ISP you have subscribed to. This last
possibility may particularly be considered an infringement of the network neutrality
principle. We see that sponsoring can be beneficial to users and ISPs depending on
the chosen advertisement level. We also discuss the impact of zero-rating where an
ISP offers free data to a CP to attract more customers, and vertical integration where
a CP and an ISP are the same company.

1 Introduction

With the improving capacity of smartphones, wireless data consumption is exponen-

tially increasing. According to previsions1 , mobile data consumption will be seven-
fold larger by 2021 than by 2017. Wireless communication subscription offers are

Patrick Maillé
IMT Atlantique / IRISA, Rennes, France, e-mail: patrick.maille@imt.fr
Bruno Tuffin
Inria, Rennes, France e-mail: bruno.tuffin@inria.fr
1 see http://www.businessinsider.fr/us/mobile-data-will-skyrocket-700-by-2021-2017-2/ among
others

1
2 Patrick Maillé and Bruno Tuffin

often made of unlimited telephony but caps on data over which a volume-based fee
is applied. Users may therefore be forced to monitor and even limit their data usage.
Content providers (CPs) on the other hand are economically dependent on the
amount of data consumed by users, typically via displayed advertisements: the more
volume consumed, the more advertisement and then revenue. For this reason, there
might be an incentive for content providers to sponsor data, that is, to at least par-
tially pay the Internet Service Provider (ISP) for the user’s consumption of its own
services. This way, users may expand their usage, CPs may get more revenue due
to advertisement, and ISPs are paid for the traffic increase. Examples in the USA
are: Netflix or Binge-On with T-Mobile, DIRECTV and U-verse Data Free TV with
AT&T, etc. ISPs are even offering services such that CPs can, if they wish, apply
sponsored data: examples are Verizon Wireless, AT&T Mobility and T-Mobile US,
or Orange with DataMI in France, among others.
But sponsored data bring concerns from user associations and small content
providers. It is claimed to give an unfair advantage to some content/service providers
and to eventually prevent some actors from entering the market, due to lower visi-
bility and high entrance costs to the sponsored data system if they want to get the
same service as incumbent and big providers already present. The culmination of
the sponsored data principle is the so-called zero-rating where ISPs (freely) remove
some content providers from their data caps, hoping more customers will subscribe
due to potentially unlimited usage. Those providers could then hardly be challenged.
The French ISP SFR did it with YouTube in its offer RED a few years ago. Sim-
ilarly, Facebook, Google and Wikipedia have built special programs in developing
countries, with the claimed goal to increase connectivity and Internet access. For
those reasons, sponsored data and zero-rating are currently under investigation by
regulators to determine if rules should be imposed to prevent or limit their use. In
Chile for example, the national telecom regulator has stated that it was violating net
neutrality laws and that it should be forbidden. Net neutrality means that all pack-
ets/flows are treated the same, independently of their origin, destination, and type of
service [3, 4].
The purpose of this paper is to design and analyze a mathematical model of spon-
sored data to understand if this type of behavior should be banned or encouraged.
To our knowledge, very few similar works exist, that we now summarize and from
which we differentiate ourselves. The notable works are first [2], where the authors
consider a model with a discrete set of users, a single ISP, and several (complement)
CPs able to choose a different level of sponsoring per user. They show that spon-
soring can benefit more to users than to CPs. Our model presents similarities with
that one but strongly differs on many aspects: we use a different model for CPs’
advertisement level and determination of the best strategy; our model also repre-
sents the negative externality of advertisements on users, something often forgotten,
and a different model of volume consumption highlighting the correlation between
willingness-to-pay for connectivity and consumed volume of data; finally and more
importantly, we model competition between ISPs to better link the sponsored data
practice to the network neutrality debate. We can also mention [7], dealing with
several substitutable CPs in competition, or [8, 9] including network externalities,
Sponsored data practices by competing wireless service providers 3

but again a single ISP, or [6] which nicely combines sponsored data and caching
strategies, but not considering any ISP in the equation.
Our contributions are the following: i) the use of a model different from the lit-
erature to represent user behavior. This model separates the willingness-to-pay for
connectivity and for CP content. Remark that using a new model is important to
better understand the assumptions that may change the results, or see the robustness
of conclusions. ii) We focus on CPs’ strategical decisions and their impact on all
other actors (ISPs and users). The decision variables are the amount (proportion) of
data to be sponsored and the amount of advertisement CPs will set; our study allows
us to see if sponsoring data also means more advertisement for a return on invest-
ment for CPs. iii) We investigate the best sponsoring and advertisement strategies in
the case of competition between ISPs; three situations are compared: (a) no spon-
sored data (b) the same level of sponsoring at all ISPs (c) the possibility to sponsor
differently according to the ISP; these situations also correspond to three different
levels of neutrality in the net neutrality debate. iv) We investigate and compare with
all other situations the case of zero-rating, when the CP content is not counted in
the user data cap; ISPs could indeed be in favor of such a strategy if it brings addi-
tional subscriptions. v) Finally we look at the case when the CP is managed by one
ISP, that is the so-called vertical integration; does it lead to an unfair advantage if
data are sponsored with respect to a non-sponsored situation? This corresponds to
practical situations within an increasingly vertically-integrated ecosystem.
The remainder of the paper is organized as follows. Section 2 presents the model,
the decision variables and their hierarchy. Section 3 discusses the last level, how
users distributes themselves among ISPs depending on prices and strategies of CPs.
Then Section 4 describes the three sponsoring strategies and compares their impact
on all actors in the case of independent CPs and ISPs. The particular situation of
zero-rating, with the CP content not included in one ISP data plan but included in
the other is investigated in Section 5. Section 6 presents the case of a vertically-
integrated CP-ISP. Finally, from all the described scenarios, we conclude in Sec-
tion 7 on the need to regulate sponsored data.

2 Model

We consider three types of actors: users, CPs and ISPs. We describe here the model
with full generality, and will later restrict the number of actors to focus on specific–
and to our knowledge un-investigated–aspects of the sponsored data debate.
Consider M CPs indexed by j and N ISPs indexed by i.
4 Patrick Maillé and Bruno Tuffin

2.1 Users

We index users by θ ∈ R+ , which characterizes their type. We consider a continuum

of users of total mass 1, and denote by F the cumulative distribution function of θ .
Each user θ has several variables of decision: its choice of ISP, that we will denote
by i(θ ), but also the volume of CP j data (without advertisements) he will consume
if using ISP i, which we will denote by vi, j (θ ). The dependence on i comes from
the fact that prices for data may differ between ISPs, and the consumed volume too
as a consequence.
To simplify, let θ be the willingness-to-pay for connectivity and pi the subscrip-
tion price at ISP i, which would be giving a “utility for connectivity” θ − pi . But we
additionally weigh the difference using an ISP reputation parameter ai ≥ 0, mean-
ing that users value more the gain obtained with incumbent providers than with
newcomers, resulting in a “weighted utility for connectivity”

ai (θ − pi ).

In addition, User θ gains satisfaction from using each CP j. The cost per unit of
volume he pays in his data plan is denoted by ci, j , which depends on j because
that usage can be sponsored by CP j as we will see in the next subsection. More-
over there is a valuation for each unit of volume of CP j. Let rθ0 , j be the marginal
valuation, that we will consider to be

rθ0 , j (x) = [θ − (α j s2j )x]+

for the x-th unit of useful volume, i.e., without advertising, where [y]+ := max(y, 0),
α j is a fixed parameter, and s j ≥ 1 corresponds to the relative increase of volume
due to advertisement displayed by CP j, expressed as the total downloaded volume
divided by the volume of “useful” data, excluding ads. The larger this value of s j , the
smaller the service valuation for users. We consider a square value of s j to assure the
reasonable assumption that at one point, for the CPs, the loss due to users unpleased
by advertisements exceeds the gain from those ads (see later (2) and (3) with s j
tending to infinity); any other choice can be used without difficulty though. Then
the willingness-to-pay rθ , j (x) for consuming a volume x of CP j data over a month
is  2
 θ x − α j s2j x2 if x ≤ θ 2
α js j
rθ , j (x) =
 θ2 2 otherwise.
2α s
j j

This form is commonly adopted in [1, 5] and presents the advantage of positively
correlating willingness-to-pay for data and willingness-to-pay for connectivity. We
have added in the model the negative correlation due to advertisement (some kind
of “pollution” which can advert users). Note that a positive effect (only) from ad-
vertisement was considered in [2] due to relevant ads which can be clicked, but ads
seems to us rather negatively perceived.
Our assumptions give a utility for user θ at ISP i
Sponsored data practices by competing wireless service providers 5

M

Ui (θ ) := ai (θ − pi ) + ∑ rθ , j (vi, j (θ )) − ci, j vi, j (θ )s j , (1)
j=1

using the fact that the total downloaded volume is actually vi, j (θ )s j from the defini-
tion of s j . Users indeed also download advertisements, which are not differentiated
from “real” content by ISPs. The additive expression indicates that CPs are assumed
to be independent in terms of content.
Our modeling, unlike those of the literature ([2] for example), chooses to repre-
sent user preferences for ISPs, in terms of connectivity and a correlation between
willingness-to-pay for connectivity and consumed volume of data; it separates what
comes from ISPs from what comes from CPs.
If subscribing to ISP i, the volume vi, j (θ ) User θ chooses is the one maximizing
rθ , j (vi, j (θ )) − ci, j s j vi, j (θ ), and can easily be computed [1] to be
" #+
θ − ci, j s j
vi, j (θ ) = , (2)
α j s2j

(θ −ci, j s j )2
leading to rθ , j (vi, j (θ )) − ci, j s j vi, j (θ ) = 2(α j s2j )
1l{θ >ci, j s j } .
Summarizing, User θ chooses ISP i(θ ) maximizing his overall utility:
M
(θ − ci, j s j )2
i(θ ) = argmaxiUi (θ ) = argmaxi ai (θ − pi ) + ∑ 1l{θ >ci, j s j }
j=1 2(α j s2j )

if the max is non-negative, otherwise i(θ ) = 0, meaning no subscription at all.

2.2 CPs

CP j is assumed to have a utility (a revenue) linearly increasing with the volume

of displayed advertisement, with CP-dependent linear parameter β j . The advertise-
ment volume is the total volume s j vi(θ ), j (θ ) minus the “real” data volume vi(θ ), j (θ ),
hence (s j − 1)vi(θ ), j (θ ). As a consequence, the gain for CP j is
Z
β j (s j − 1)vi(θ ), j (θ )dF(θ ).
θ

Each CP j can decide to sponsor a fraction γi, j of the data usage cost of ISP i’s
users. This could be an incentive to consume more CP j content and therefore gen-
erate more revenue from advertisement. Of course, there is a trade-off, which is one
purpose of this paper, between sponsoring cost and generated volume of consumed
data. We will investigate three different sponsoring policies:
1. No sponsored data: γi, j = 0 ∀i, j;
2. The same data sponsoring level for all ISPs: γi, j = γ j ∀i, j;
6 Patrick Maillé and Bruno Tuffin

3. A possible differentiation between ISPs, with γi, j 6= γi0 , j for i 6= i0 .

In terms of the network neutrality debate, those three scenarios correspond to three
different levels of neutrality. Option 1 corresponds to full neutrality. Option 2 seems
neutral with respect to ISPs, even if non-neutral between applications because dif-
ferent applications can have different levels of sponsoring (note on the other hand
that advocates of sponsored data also claim that packets are treated the same within
the network, but it is not the issue we want to address here). Option 3 is the fully
non-neutral one.
Let qi be the unit price ISP i is (officially) charging users for data. After data
sponsoring, the unit cost seen by users when consuming data of CP j is then

ci, j = qi (1 − γi, j ).

If sponsoring data, CP j has to pay to each

R
ISP for the proportion of volume
it has chosen to sponsor, hence a total cost: θ γi(θ ), j qi(θ ) s j vi(θ ), j (θ )dF(θ ), where
the proportion of volume paid, is the proportion γ j of the total volume s j vi(θ ), j (θ ),
(which includes ads).
Compiling all the elements, the revenue of CP j is
Z
Gj = (β j (s j − 1) − γ j qi(θ ) s j )vi(θ ), j (θ )dF(θ ). (3)
θ

CP j has several decision variables: the sponsoring levels γi, j ∀i and the adver-
tisement volume increase level s j ≥ 1. Note that we could also consider advertising
levels that would depend on the user ISP (i.e., si, j instead of s j ): we rather impose
an equal advertising level, since our focus is on the sponsoring strategies, and CPs
may not know the ISP of the user upon a request for content.

2.3 ISPs

Each ISP i tries to maximize its revenue. Revenue comes from subscriptions and
consumed data:
Z
!
Ri = pi + ∑ qi s j vi, j (θ ) 1l{i(θ )=i} dF(θ ).
θ j

2.4 What we analyze

Our purpose in this paper is to analyze the impact of the decisions of CPs on all ac-
tors, in the context of competing ISPs. In particular, we would like to see the impact
of a “neutrally” sponsoring CP (that is, γi, j = γi0 , j for any ISPs i, i0 ) with respect to
sponsoring differently depending on ISPs and with respect to no sponsoring.
Sponsored data practices by competing wireless service providers 7

Three scenarios are analyzed in the next sections:

1. In the case of independent ISPs and CPs, what are the best advertising and spon-
soring levels of CPs and the impact on all actors? We can compare the outputs
with optimal solutions for the three sponsoring possibilities.
2. Does zero-rating, where an ISP leaves a CP out of its data plan to try to attract
customers, harm users and the other ISPs? Should it be forbidden?
3. Does vertical integration of a CP and an ISP have a positive or negative impact
on the other ISPs and on users? This, with a dominant or a non-dominant ISP.
Since CPs are complements (due to the additive form of users’ utility), we will
consider without much loss of generality in the next sections the case of a single
CP. This in order to focus on our purpose, that is the impact of advertisement and
sponsoring strategies on competition between ISPs, and on users. Hence we can
simplify the writing by removing the CP-relative indices in the notations defined
before. We will also limit ourselves to 2 ISPs, labeled 1 and 2.

3 User subscription decisions

User decisions depend on prices set by ISPs, but also on decisions of the CP. Even
if the CP “plays” first, its optimal strategy should anticipate users’ choices. For this
reason, we discuss here the repartition of users depending on fixed CP decisions. The
optimal CP decisions are analyzed in the next sections, anticipating this subsequent
choice (this implicitly assumes that the CP knows the distribution F).
In other works using the same type of models for user preferences [1, 5], we
had depending on the value of θ an interval over which users do not subscribe,
then an interval (possibly empty) over which one ISP is chosen, and then another
one for the other ISP (that is, users with small valuations do no subscribe, those
with “intermediate” valuation go with one ISP and those with the largest go with
the other). Here things can be more complicated, with interestingly more intricate
intervals. To illustrate this, consider the values a1 = 1.3, a2 = 1, s = 1, α = 1,
c1 = 2, c2 = 1, p1 = 1, and p2 = 0.97, i.e., ISP 1 has a better reputation but higher
subscription and usage prices. Figure 1 displays the difference max(U1 (θ ), 0) −
max(U2 (θ ), 0) in terms of θ . We consider the max with 0 to show only the situations
where a user θ is willing to subscribe to an ISP.
We can see (when the curve gives zero) that users with small valuations θ prefer
not to subscribe to any ISP; then there is an interval where ISP 2 is preferred, then
an interval where it is ISP 1, and then again ISP 2. An interpretation is as follows:
users with small valuation (but large enough to subscribe) are first interested in
connectivity with a smaller connectivity price at ISP 2. At one point, due to its larger
reputation, connectivity attractiveness of ISP 1 becomes larger, but it is counter-
balanced by the data cost of ISP 2 (the parabola).
In general, in terms of the combination of prices, we can express how many such
successive intervals above the “no subscription” possibility can be found, between
8 Patrick Maillé and Bruno Tuffin

[U1 (θ )]+ − [U2 (θ )]+

Users prefer ISP 2

Users prefer ISP 1

Users prefer ISP 2

Users do not subscribe

0 0.5 1 1.5
θ

Fig. 1 [U1 (θ )]+ − [U2 (θ )]+ in terms of θ . In the left zone (small θ values), users do not subscribe
to any provider.

one and three. We do not give such details here since it is a list of cases which is
not very instructive. The situation becomes even more intricate if we increase the
number of ISPs (and CPs). Solving it numerically in the next section does not pose
any problem.

4 Analysis with independent CP and ISPs

In all our numerical investigations, θ follows an exponential distribution with rate

1 truncated at 2, i.e., F(θ ) = (1 − e−θ )/(1 − e−2 ) for θ ∈ [0, 2]. Unless specified
otherwise (i.e., when a parameter explicitly varies), the parameter values we con-
sider throughout all the paper are summarized in Table 1. ISP 1 has the position

a1 a2 p1 p2 q1 q2 β α
1.5 1 1 0.6 0.6 0.4 1 1
Table 1 Parameter values for the numerical investigations

of an incumbent: with a better reputation than its competitor, it can charge higher
subscription and per-usage prices.
We discuss in this section the case where the ISP and CP are independent. We
display in Figure 2 the CP revenue as s varies for fixed values of the other param-
eters, including fixed values of the sponsoring coefficient γi on ISP i = 1, 2, but
using the subsequent user subscription decision. It can be checked that there is an
optimal level of advertising s∗ around 2.7 above which increasing the displayed ads
diminishes revenue.
In Figure 3 we display the CP revenue-maximizing sponsoring levels γi as s
varies, and the corresponding CP Revenue. We can see that the optimal γi is in-
Sponsored data practices by competing wireless service providers 9

CP revenue
1 2 3 4 5 6
total volume
Advertisement overhead s = volume without advertisement

Fig. 2 CP revenue, with γ1 = 0.7, γ2 = 0.4.

·10−2
1 4

CP revenue G
0.5 2 γ1 = γ2 = 0
γ

Opt. γ1 = γ2
γ1 = γ2 Opt. (γ1 , γ2 )
γ1
0 γ2 0
2 4 2 4
s s

Fig. 3 CP revenue-maximizing sponsoring coefficient γ (left), and corresponding CP revenue

(right) in each scenario.

creasing with the level of advertisement s: the more ads you put, the more you can
sponsor data because you earn more, and the more you have to do to compensate the
negative externality for users. Typically also, applying the same level of sponsoring
(γ1 = γ2 ) leads to a sponsoring level in between the differentiated levels of spon-
soring (γ1 6= γ2 ). Similarly to Figure 2 where sponsoring was fixed, in the present
case of optimal sponsoring for each s, there is an optimal advertising level for each
sponsoring scenario in Figure 3 (right). As expected, the CP revenue with differen-
tiated sponsoring is larger than the one with identical sponsoring, itself larger than
the one with no sponsoring, because the optimization is on a larger set of sponsoring
parameters. When the advertisement level is low, there is no significative difference
between the three options; it is significant when s is large. The optimal advertis-
ing levels s∗ are 1.34 with no sponsoring, 1.34 too with identical sponsoring, and
1.44 with differentiated sponsoring, a larger value. “No sponsoring” or “identical
sponsoring” yield a similar optimal revenue, while it is significantly better for dif-
ferentiated sponsoring which then seems the appropriate option for the CP.
10 Patrick Maillé and Bruno Tuffin

γ1 = γ2 = 0
0.4 0.45
ISP 1 γ1 = γ2

Consumer Surplus
Opt. (γ1 , γ2 )
Revenue ISPs 0.3 γ1 = γ2 = 0
γ1 = γ2
0.4
0.2 ISP 2 opt. (γ1 , γ2 )

0.1
0.35

2 4 2 4
Advertisement level s Advertisement level s

Fig. 4 ISP revenues (left) and Consumer Surplus (right) with CP-revenue-maximizing γ values.

Figure 4 displays the impact of sponsoring on the revenues of ISPs and on con-
sumer surplus. In terms of ISP revenue first, we can see that ISP 1, the one with the
highest reputation, makes more money than ISP 2. For very low values of s the three
policies give the same revenues. ISP 2 always prefers sponsoring to no sponsoring,
while for ISP 1 an identical sponsoring is always the best option, something not so
obvious at first sight. As a second choice for ISP 1, depending on the advertisement
level, preference varies between no sponsoring and a differentiated one. Sponsor-
ing is also always preferred by users, since giving a higher consumer surplus. On
the other hand, the consumer surplus decreases as the advertisement level increases:
the increased sponsoring level does not compensate enough the negative external-
ity of ads. Interestingly, there is no significant difference for users between the two
sponsoring options, and no dominance either.
From the figures, sponsoring seems a relevant option for all actors, with some
differences in preferences between the two sponsoring options depending on the
actors.
Table 2 compares the outputs at the optimal advertising levels selected by the CP.
Note that in the case of identical sponsoring policy, the optimal advertising level is

s∗ γi G R1 R2 CS
No sponsoring 1.34 (0,0) 0.0378 0.4124 0.0931 0.3813
Id. sponsoring 1.34 (0,0) 0.0378 0.4124 0.0931 0.3813
Dif. sponsoring 1.45 (0,0.41) 0.0411 0.3317 0.1559 0.3770
Table 2 Comparison of the three sponsoring scenarios with CP-revenue-maximizing advertise-
ment and sponsoring levels, for the parameter values of Table 1.

such that there is no sponsoring at all. Hence no sponsoring and identical sponsoring
lead to the same output with our set of parameters. Differentiated sponsoring leads
to slightly more ads. With this optimal choice, only customers of ISP 2 get spon-
sored data, 41% of data being sponsored. It leads to a larger CP revenue than the
two other options. The revenue of ISP 1 decreases because attracting less customers,
Sponsored data practices by competing wireless service providers 11

to the benefit of ISP 2. Consumer surplus is then slightly decreased. In conclusion,

with this set of parameters and optimal advertising, identical sponsoring is of no
help since leading to no sponsoring at all, and differentiated sponsoring just slightly
changes consumer surplus but really modifies ISP revenues to the advantage of the
one with the lowest reputation; this could be beneficial for competition with incum-
bent providers having usually a better reputation than newcomers.

5 Zero-rating

Zero-rating consists in leaving a CP out of the data plan, making it free of access for
users. The potential interest for an ISP is to attract more customers, and therefore
compensate the volume-based revenue loss by a subscription revenue increase. We
study here whether this kind of strategy makes sense and what is the impact on all
actors. Implementing zero-rating translates into ci, j = 0. In the case of a single CP,
it is equivalent to considering qi = 0.
Figures 5 to 8 alternatively display the outputs in the cases where ISP 1 or ISP 2
implement zero-rating. Looking at the two scenarios helps to see if zero-rating is
more relevant and has more impact depending on the market position of the ISP.
Indeed, by keeping all values of parameters the same as in the previous section,
ISP 1 is more established than ISP 2.
We see in Figure 5 (left) that in the case of zero-rating for ISP 1, the optimal
sponsoring strategy for ISP 2 does not impact much the CP revenue, whatever the
level of advertising. With respect to no zero-rating in Figure 3, the CP gain is sub-
stantial. Similar results can be observed when it is ISP 2 which applies zero-rating:

γ1 = 0
γ1 > 0
0.1 0.1
CP revenue

CP revenue

0.05 0.05

γ2 = 0
0 γ2 > 0 0
1 2 3 4 1 2 3 4
Advertisement level s Advertisement level s

Fig. 5 CP revenue with zero-rating for ISP 1 (left) or for ISP 2 (right).

substantial gain for the CP, but at a smaller advertising level than when it is applied
by ISP 1, while at high advertising levels the difference between no sponsoring or
sponsoring for ISP 1 is significant. At the optimal advertising level, there is no dif-
12 Patrick Maillé and Bruno Tuffin

ference though. In Figure 6, we can see that, again, the sponsoring level increases

0.5
γ

Opt. γ2 (zero-rating for ISP 1)

0 Opt. γ1 (zero-rating for ISP 2)

1 1.5 2 2.5 3 3.5 4

Advertisement level s

Fig. 6 Optimal γ for an ISP, with zero-rating for the other ISP.

with the level of advertisement. But it can be checked that for smaller advertising
levels, there is no sponsoring (which is understandable due to the low(er) revenue
of the CP based on visits) and that sponsoring (for ISP 2 users) starts at a smaller
advertising level for zero-rating with ISP 1 than in the opposite situation.

0.4

0.4
ISP Revenues

ISP revenues

ISP 1, γ2 = 0
ISP 2, γ2 = 0 0.2
0.2 ISP 1, γ2 > 0
ISP 2, γ2 > 0

0 0
1 2 3 4 1 2 3 4
Advertisement level s Advertisement level s

Fig. 7 ISP revenues with zero-rating for ISP 1 (left) and for ISP 2 (right).

In terms of ISP revenue in Figure 7, a zero-rating at ISP i induces for small adver-
tising levels no revenue at the other ISP, which can be seen as against competition
and could lead to action from a regulator, especially given that the optimal advertis-
ing levels are actually quite low from Figures 5 to 8. Sponsoring tends to reduce the
gap in revenue with respect to no sponsoring when zero-rating is applied at ISP 1
only, while it is the opposite when applied at ISP 2.
Interestingly in Figure 8, consumer surplus does not change much with the policy
if zero-rating is applied at ISP 1, the largest ISP. It changes a bit more when it is
Sponsored data practices by competing wireless service providers 13

0.7 γ2 = 0, zero-rating for ISP 1

γ2 > 0, zero-rating for ISP 1

Consumer Surplus
0.6 γ1 = 0, zero-rating for ISP 2
γ1 > 0, zero-rating for ISP 2

0.5

0.4

1 1.5 2 2.5 3 3.5 4

Advertisement level s

Fig. 8 Consumer Surplus with zero-rating for one ISP.

applied at ISP 2, with large levels of advertising only. In both cases anyway, note
that sponsoring is beneficial to users, but at the advertising level chosen optimally
by the CP, there is no real difference (because the selected value is γ1 = 0 in case of
zero-rating at ISP 2).

6 Vertical Integration

We know look at the case of vertical integration: when the CP and one of the ISPs
are the same firm. This situation becomes increasingly frequent in practice, with
ISPs offering video-on-demand services, among other. In that case the optimization
of parameters (sponsoring and advertising) is based on the combined revenues of
the CP and the ISP, by just adding them.
Figure 9 (left) displays the revenue of the integrated CP-ISP in terms of s, in the
cases where it is either ISP 1 or ISP 2 which is integrated, and with fixed sponsoring
levels. ISP 1 being more established, it is not surprising that integrating it yields a
larger revenue. But the optimal advertising level is larger in the case of ISP 2 with
those parameters. Figure 9 (right) shows the revenue of the integrated entity (CP-
ISP 1, or CP-ISP 2) with the corresponding optimal γ. Recall that in the case of
non-integration, the optimal advertisement level was 1.35 for no or identical spon-
soring, and 1.45 for differentiated sponsoring. We still have the dominance such that
differentiated sponsoring is better than identical one, itself better than no sponsor-
ing. It can be seen that for an integrated CP-ISP 1, identical sponsoring leads to the
most ads (a level around 1.6, and 1.0–i.e., no ads–for no sponsoring and 1.4 for a
differentiated one), while it is differentiated sponsoring in the case of an integrated
CP-ISP 2 (a level around 1.65, and 1.05 both for no sponsoring and for an identical
one). So the optimal advertising level maybe less or more for the same sponsoring
policy if the CP is integrated or not. But if choosing the optimal policy, advertising
is larger if vertical integration is applied.
14 Patrick Maillé and Bruno Tuffin

γ1 = γ2 = 0
γ1 = γ2
0.8 0.6 CP-ISP 1

Revenue CP-ISP
Opt. (γ1 , γ2 )

Revenue Integrated CP-ISP 1

0.6 Integrated CP-ISP 2
0.4
CP-ISP 2

0.4
2 4 6 2 4
Advertisement level s Advertisement level s

Fig. 9 CP+ISP revenue in the two integration scenarios, when γ1 = 0.7, γ2 = 0.4 (left), and with
optimized values of (γ1 , γ2 ) (right).

γ1 = γ2 = 0 γ1 = γ2 = 0
γ1 = γ2 γ1 = γ2
0.5 0.45
Consumer Surplus

Opt. γi With opt. γi

No integ., opt. γi User Welfare No integ., opt. γi
0.45
0.4
0.4

0.35 0.35

2 4 2 4
sA sA

Fig. 10 Consumer surplus with integrated CP-ISP 1 (left) or integrated CP-ISP 2 (right), for βA =
1.00, βB = 1.50, r1 = 1.50, r2 = 1.00, p1 = 1.00, p2 = 0.60, q1 = 0.60, q2 = 0.40, αA = 1.00, αB =
1.00, sB = 5.00.

Figure 10 gives the consumer surplus in terms of s for the two integration cases,
and compares it with the optimal non-integrated strategy obtained in Section 4. It
allows to see if integration has a positive or negative impact on users, something
of interest for the regulator. We observe that vertical integration is good for users,
especially if sponsoring data is allowed. An identical sponsoring is the best option
with this regard. The reason is that it induces much more data sponsoring resulting
in lower user cost.
Finally, Figure 11 displays the revenue of the non-integrated ISP for all strategies,
in order to illustrate whether integration brings down competition. We also again
plot the revenues in cases of no integration for comparison sake. An integrated ISP 1
reduces drastically ISP 2 revenue, which was expected, but at a lesser extent for the
identical sponsoring strategy because of less freedom. Integrating the incumbent ISP
is typically against competition. Surprisingly, when it is ISP 2 who is integrated and
Sponsored data practices by competing wireless service providers 15

0.25 0.5
γ1 = γ2 = 0
γ1 = γ2
0.2 Opt. γi 0.4
ISP 2 revenue

ISP 1 revenue
No integ., γi = 0
No integ., γ1 = γ2
0.15 0.3
No integ., opt. γi

0.1 0.2

0.1
2 4 2 4
Advertisement level s Advertisement level s

Fig. 11 Revenue of the non-integrated ISP if ISP 1 (left) or ISP 2 (right) is integrated.

with identical sponsoring, ISP 1 can benefit from the integration of the competitor.
Note it is not the case for the other sponsoring strategies.
Table 3 compares the outputs at the optimal advertising levels selected by the
CP in the two integrated scenarios. They can be compared with the non-integrated
case of Table 2. Vertical integration, except partially for the last line of Table 3,

CP integrated with ISP 1 CP integrated with ISP 2

s∗ (γ1 , γ2 ) G + R1 R2 CS s∗ (γ1 , γ2 ) G + R2 R1 CS
No spons. 1.0 (0,0) 0.6445 0.1653 0.4600 1.05 (0,0) 0.4370 0.4062 0.4420
Id. spons. 1.6 (0.998,0.998) 0.6614 0.1171 0.4833 1.05 (0,0) 0.4370 0.4062 0.4420
Dif. spons. 1.4 (0.783,0.995) 0.6697 0.1925 0.4882 1.65 (0.243,0.997) 0.4644 0.2021 0.3978
Table 3 Output at optimal s when the CP is integrated with an ISP.

always significantly benefits to users and competitors, something not so intuitive.

We see in Table 3 that vertical integration leads to more sponsoring than in Table 2,
so that consumer surplus is increased. With an integrated CP-ISP 2, the increased
satisfaction in the first two lines is due to the smaller optimal advertising level; for
the last line, the advertising level is larger but only partially counter-balanced by the
increased sponsoring.

7 Conclusions

The purpose of this paper was to study the impact of sponsored data, differentiated
or not, on all telecommunication actors, in the presence of competing ISPs. Our
main results are the following:
• Sponsoring can be beneficial to users and ISPs and the impact depends on the
chosen advertisement level.
16 Patrick Maillé and Bruno Tuffin

• Zero-rating at one ISP can lead to no revenue at the other ISP, which can be seen
as harming competition.
• Vertical integration can be beneficial for all actors in this context, if the CP is
integrated to the incumbent ISP (the one with the largest reputation).
Conclusions of course depend on the arbitrarily chosen set of parameters, but the
paper provides insight on the potential effects of sponsored data, and a model to
perform such an analysis.
We plan to extend the promising first results of this work in one main direction:
We aim at inserting another round of decisions, on ISP prices (for subscription and
for data usage). Our purpose here was to see the impact of having several ISPs on
the CP decisions leading to important conclusions. Inserting a pricing game will
complete the analysis but is computationally demanding. Other interesting exten-
sions include considering a larger set of ISPs to show that the obtained results are
without loss of generality in this respect, as well as considering CPs which can be
substitutes.

References

1. M. Cho and M. Choi. Pricing for mobile data services considering service evolution and change
of user heterogeneity. IEICE Transactions on Communications, E96-B(2):543–552, 2013.
2. C. Joe-Wong, S. Ha, and M. Chiang. Sponsoring mobile data: An economic analysis of the
impact on users and content providers. In Proc. of INFOCOM 2015, Hong-Kong, China, 2015.
3. P. Maillé, P. Reichl, and B. Tuffin. Internet governance and economics of network neutrality.
In A. Hadjiantonis and B. Stiller, editors, Telecommunications Economics - Selected Results of
the COST Action IS0605 EconTel, pages 108–116. Lecture Notes in Computer Science 7216,
Springer Verlag, 2012.
4. P. Maillé, G. Simon, and B. Tuffin. Toward a net neutrality debate that conforms to the 2010s.
IEEE Communications Magazine, 54(3):94–99, 2016.
5. P. Maillé and B. Tuffin. Users facing volume-based and flat-rate-based charging schemes at the
same time. In IEEE Explore, editor, 8th Latin American Network Operations and Management
Symposium (LANOMS 2015), pages 23–26, Joao Pessoa, Brazil, October 2015.
6. H. Pang, L. Gao, Q. Ding, and L. Sun. When data sponsoring meets edge caching: A game-
theoretic analysis. CoRR, abs/1709.00273, 2017.
7. R. Somogyi. The economics of zero-rating and net neutrality. CORE Discussion Papers
2016047, Université catholique de Louvain, Center for Operations Research and Economet-
rics (CORE), 2016.
8. Z. Xiong, S. Feng, D. Niyato, P. Wang, and Y. Zhang. Competition and Cooperation Analysis
for Data Sponsored Market: A Network Effects Model. ArXiv e-prints, November 2017.
9. L. Zhang, W. Wu, and D. Wang. Sponsored data plan: A two-class service model in wireless
data networks. SIGMETRICS Perform. Eval. Rev., 43(1):85–96, June 2015.