0% found this document useful (0 votes)
11 views10 pages

Analytics and Data-Driven Methods and Practices in Platform Ecosystems 2023

Uploaded by

chianti-wu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
11 views10 pages

Analytics and Data-Driven Methods and Practices in Platform Ecosystems 2023

Uploaded by

chianti-wu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 10

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/377069086

Analytics and Data-Driven Methods and Practices in Platform Ecosystems: a


systematic literature review

Conference Paper · September 2023


DOI: 10.1109/SEAA60479.2023.00018

CITATIONS READS

0 90

4 authors:

Shady Hegazy Christoph Elsner


Siemens Siemens
3 PUBLICATIONS 0 CITATIONS 46 PUBLICATIONS 358 CITATIONS

SEE PROFILE SEE PROFILE

Jan Bosch Helena Holmstrom Olsson


Chalmers University of Technology Malmö University
616 PUBLICATIONS 21,961 CITATIONS 221 PUBLICATIONS 5,777 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Shady Hegazy on 19 March 2024.

The user has requested enhancement of the downloaded file.


Analytics and Data-Driven Methods and Practices in
Platform Ecosystems: a systematic literature review
Shady Hegazy ∗, Christoph Elsner ∗, Jan Bosch †, and Helena Holmström Olsson ‡
∗ Siemens
Technology, Munich, Germany
{firstname.lastname}@siemens.com.
† Department of Computer Science and Engineering, Chalmers University of Technology, Göteborg, Sweden

[email protected]
‡ Department of Computer Science and Media Technology, Malmö University, Malmö, Sweden

[email protected]

Abstract—The emergence of platform ecosystems has trans- Google’s Android, it struggled to keep up with the pace of
formed the business landscape in many industries, giving rise evolution and innovation that was brought into the market
to novel modes of interorganizational cooperation and value co- by these platforms[5]. By 2012, Apple had nearly quarter
creation, as well as unconventional challenges. The vast traces of
data generated by platform ecosystems makes them ripe for the million external innovators, while Blackberry had only 8000,
use of analytics and data-driven methods aimed at improving representing a clear competitive advantage for one side on
their health, performance, business outcomes, and evolution. the ecosystem level [5]. Around two million apps in Apple’s
However, the research on the application of analytics within app store [6], and around three million apps in Google’s play
platform ecosystems is limited and spread across multiple disci- store [7] are an example of platform ecosystems catalyzing
plines. To address this gap, we conducted a systematic literature
review on the application of analytics and data-driven methods innovation and evolution in scales that conventional firms may
and practices within platform ecosystems. A total of 56 studies not be able to match. However, such scale of innovation and
were reviewed, and underwent data extraction, analysis, and evolution comes with many challenges. Orchestration of plat-
synthesis processes. In addition to presenting themes and patterns form ecosystems requires a complex decision making process
in the recent and relevant literature on platform ecosystems to deal with the variability and dynamism that distinguishes
analytics, our review offers the following outcomes: an actionable
overview of the analytics toolbox currently used within platform such environment [2]. The vast amount of data produced by the
ecosystems—spanning domains such as machine learning, deep networked actors within platform ecosystems provide a great
learning, data science, modelling, simulation, among others—; a potential for data-driven decision making, design optimization,
roadmap for practitioners to achieve analytics maturity; and a and transformation [8]. The application of analytics and data-
summary of underexplored research areas. driven methods and approaches is therefore needed to harness
Index Terms—Platform ecosystems, software ecosystems, ana- value out of such data [9]. However, such methods have
lytics, data-driven, big data, data science, machine learning, deep
learning, modelling, simulation, maturity model. received limited focus in primary and secondary research
on platform ecosystems [10]. We carry out this systematic
literature review to help alleviate this shortcoming.
I. I NTRODUCTION
The contribution of this systematic literature review is four-
In contrast to conventional firms that create value within the fold. First, we distill and discuss the recurring themes and
confines of a company or a supply chain, platforms leverage patterns in the research on platform ecosystems analytics.
ecosystems of autonomous actors, such as software developers Second, we present an overview of the analytics toolbox
and service providers, to co-create value within an ecosystem that can be used within platform ecosystems to provide valu-
that revolves around the platform as the core technological able outcomes. Third, we provide an actionable roadmap for
foundation [1, 2]. The interconnectedness, mutual dependence, practitioners to unlock the potential of analytics in platform
and omnidirectional collaboration within platform ecosystems ecosystems. Fourth, we highlight underexplored research areas
accelerate innovation and lead to evolution leaps within the that require the attention of researchers in the field for future
technological fields they operate in [3]. Platforms change research. In the following sections, we discuss the details
the competitive landscape in the businesses and industries of the research methodology and results. Section II presents
that they emerge within, as the competition shifts from the the background for this study. Section III outlines the design
product level to the ecosystem level [4]. It ”no longer revolves and architecture of the review protocol. Section IV discusses
around how to control the value chain but around attracting the implementation of the search strategy. In section V, we
generative activities associated with a platform” [1, p. 1]. The report the results of the descriptive and thematic synthesis
smartphone industry is a stark example of such evolution of the extracted data. Section VI discusses the answer to
leaps and competition transformation. Blackberry once had the research question, the implications of the study findings
a dominant share of the smartphone market, but with the for practitioners and researchers, and discusses the threats to
emergence of platform ecosystems such as Apple’s iOS and
the validity of the study findings. Section VII presents the there have been approaches to embed analytics at the core
conclusions of this study. of platform ecosystems to utilize big data generated within
the platform to allow for data-driven decision making, design
II. BACKGROUND optimization, and transformation. Tian et al. [13], together
with a research team from IBM, designed and implemented a
Actors within a platform ecosystem have different and central analytics unit within a platform ecosystem to harness
interwound goals and interests. However, platform operator, the generated data on three different tiers: data-driven business
complementors, third-party developers, users, and partners model redesign; service experiments and analytics; and data-
collectively co-create value and co-benefit from value within driven platform self-transformation. The analytics unit used
the platform ecosystem [11]. Platform operators continuously the collected data to accomplish tasks such as: price optimiza-
orchestrate and adapt the platform’s governance strategies tion, service performance benchmarking, churn prediction,
to achieve synergy between such different and sometimes marketing effectiveness estimation, value distribution analysis,
conflicting interests [5, 11]. In addition, many factors and network effects analysis, among others. Another remarkable
tradeoffs have to be accounted for during the decision making initiative was carried out within the software ecosystem of
process. For example, platform openness enhances generativ- the financial giant Fannie Mae. An analytics platform was
ity and attracts complementors, but excessive openness may deployed within the ecosystem, and data integration routines
reduce the quality of value generated within the platform, were set-up to make the data available for the analytics unit.
thus repelling business partners from joining the ecosystem Analytics were applied to the generated data to provide in-
[3]. Another challenging governance decision is the balance sights regarding transformation effects on productivity, quality,
between platform orchestrator’s control over value co-creation and overall ecosystem performance [14].
processes and the impediment caused by such control on the Despite the growing interest in understanding the structure,
generativity among platform participants, which is known as governance, and evolution of platform ecosystems, there has
the generativity-control tension [3, 11]. In addition, platform been a lack of primary and secondary research on the appli-
ecosystems landscape is characterized by continuous evolution cation of analytics within platform ecosystems [1, 8, 10, 15].
and boundless possibilities of transformation along many axes In 2013, Barbosa et al. [10] carried out a systematic mapping
[1, 4]. Not only the complexity, but also the pace of evolution study of software ecosystems from a general perspective. The
is boosted by platform ecosystems, as ”dynamics visible only results of the mapping study showed that the application
over 30–40 years in most industries can be observed in of analytics within software ecosystems is an underexplored
5–7 years in platform businesses”—a phenomenon described research area compared to other commonly researched areas
as ”compressed evolution” [5, p. 155]. All these aspects such as ecosystem governance, evolution, and modelling [10].
contribute to the complexity of decision-making processes, In 2021, Schüler and Petrik [8] conducted a co-citation and
and gives rise to multi-level and multi-aspect competition systematic literature review on objectives of research on
between different ecosystems [4]. With many factors to be platform ecosystems from a general perspective. The review
taken into account, ”platform owners repeatedly face paradox- results showed that research on platform ecosystems rarely
ical design decisions” [11, p. 2], necessitating organizations uses analytics, data science, or data mining techniques. The
to continuously adapt and refine their strategies to maintain review results confirms findings from other reviews that many
competitiveness and ensure long-term sustainability [4]. areas of platform ecosystem research ”could benefit from the
The extensive network of actors and participants within application of data science methods” [8, p. 21]. In 2022,
a platform ecosystem ”generates an enormous volume of Malcher et al. [15] carried out a tertiary study to review
digital traces as by-products, which can be used for further and thematically analyze secondary research on software
innovation and is commonly referred to as big data” [8, p. ecosystems from a general perspective. The study results
14]. Information on API usage, service consumption, artifact showed that the research on analytics methods and techniques
creation, interactions between actors, and end-user actions within software ecosystems is quite immature compared to
all represent valuable data points. The platform’s inherent research on more conceptual and theoretical aspects of soft-
connectivity allows for the processing of these data points ware ecosystems which shows ”maturity signs” [15, p. 35].
as continuous streams as the actors and participants are net- The study also highlights the lack of analysis methods and
worked together within the same ecosystem [5]. Davenport and tools designed specifically to address the unique challenges of
Harris [9] suggested that firms need to develop comprehensive software ecosystems.
strategies for using analytics in order to extract every possible
value out of the deluge of data made available through their III. M ETHODOLOGY
platforms [9]. Acting upon the amassed data is more relevant
A. Research Question
in the case of platform ecosystems due to the complexity of
decision making and the multifaceted nature of competition In order to translate the study goal—surveying analytics and
between ecosystems [12]. Analytics enable organizations to data-driven methods applied within platform ecosystems—
harness their data to generate insights enabling data-driven into a research question, we employed the Population, Inter-
actions and informed decision-making [9]. As early as 2008, vention, Comparison, Outcome, Context (PICOC) criteria for
research question formulation [16], as suggested in [17, 18]. was excluded because of the time and location dependence
Using an adaptation of that criteria for software engineering of its results, in addition to an insufficient character limit
studies by Kitchenham and Charters [17], the study goal was [19]. DBLP was excluded because of inconsistency issues
unpacked into its constituent concepts and entities as follows: in boolean operators execution [19]. Wiley Online Library
Population Platform ecosystems. was excluded as it restricts the number of accessible hits.
Intervention Analytics and data-driven methods and tech- CrossRef was excluded as it has significant coverage overlap
niques. with Scopus, while offering less complete metadata [20].
Comparison Although our review does not involve a direct ScienceDirect was excluded due to its very restrictive term,
comparison between different interventions, we aim to and boolean limit in advanced search [19]. Web of Science
identify and assess the variety and outcome of the inter- was excluded as its scope of coverage vary significantly based
ventions applied to the population. on many interwound factors within institutional subscription
Outcome The focus of this study is on outcomes that con- package components [19]. The refined list, using the termi-
tribute considerable value to platform ecosystems. nology defined in [19], consisted of Scopus and ACM Digital
Context Platform ecosystems in business settings. Library as principal sources, in addition to IEEE XPLORE as
a supplementary source.
With the above structure in mind, we worded the research
3) Time Frame: To determine the time frame for this
question as follows:
review, we began with a preliminary search to identify early
RQ1 What analytics and data-driven methods and techniques publications discussing the population of this study. Our search
are used for monitoring or improving outcomes within revealed two heavily cited publications: ”Platform Leadership:
platform ecosystems in a business environment? How Intel, Microsoft, and Cisco Drive Industry Innovation”
by Gawer and Cusumano [21], published in 2002; and ”Two-
B. Search Strategy Sided Network Effects: A Theory of Information Product
1) Search Query: For each section of the research question Design” by Parker and Van Alstyne [22], published in 2005.
structure, search terms were curated as shown in table I. The We then conducted a dual-route backward citation search,
search queries were then formed by joining the bracketed which led us to the article ”Systems Competition and Network
groups of search terms using AND and OR booleans, then Effects” by Katz and Shapiro [23], which was published in
applying adaptations to match the syntax requirements of 1994. Further backward citation searches indicated articles
databases and search engines. Those forms vary only syntac- published before 1994 had significantly less relevancy. Thus,
tically with no alteration to the search query and terms. the time frame for this search was set to start from that year.
TABLE I
SEARCH QUERY TERMS C. Inclusion and Exclusion Criteria
In order to ensure adherence to the review scope, the
”software ecosystem*” OR ”platform ecosystem*”
Population OR ”digital ecosystem*” OR ”digital platform*” inclusion and exclusion criteria shown in table II were curated
OR ”software platform*” to be applied to the initial search results. IC refers to inclusion
criteria, and EC refers to exclusion criteria. These criteria were
”quantitative” OR ”driven” OR ”analy*” OR designed using the guidance in Kitchenham and Charters [17],
Intervention ”machine learning” OR ”deep learning” OR and Gough et al. [18]. The inclusion and exclusion criteria
”neural” OR ”reinforcement” OR ”predict*” OR are to be applied to the search results by two researchers. In
”mining” OR ”metrics” OR ”continuous” OR
Comparison ”adapt*” OR ”simulat*” OR ”network” OR case of disagreement, opinions of the other two researchers
”graph” OR ”visual*” OR ”agent” are considered to reach a resolution.

”cluster*” OR ”assess*” OR ”impact” OR D. Quality Assessment


”recommend*” OR ”decision” OR ”health” OR
Outcome ”generativ*” OR ”requirement*” OR We explored the checklists and frameworks for quality as-
”performance” OR ”outcome” OR ”improve” OR sessment of primary studies recommended in [16, 18]. Among
”grow*” OR ”support” OR ”feedback” the most applicable frameworks for this study were: Critical
Appraisal Skills Programme (CASP)[24]; and Standard Qual-
Context
”business*” OR ”industry” OR ”market” OR ity Assessment Criteria (SQAC)[25]. The SQAC criteria was
”customer” OR ”user” adopted, as it has a subset designed for quantitative studies.
2) Databases and Resources: We confined our search to As the criteria were mainly oriented towards medical studies
certain databases that cover topics of relevance to the research such as randomized controlled trials, we followed the guidance
question. We curated an initial list that included: Google in [26, 27] to adapt it our scope by adjusting the wording of
Scholar; Scopus; Web of Science; IEEE Xplore; ACM Digital some criteria, and eliminating irrelevant criteria. The criteria
Library; DBLP; Wiley Online Library; ScienceDirect; and are scored on a 3-point Likert scale, with zero points for unmet
CrossRef. That initial list was further refined to address criteria, one point for partially met, and two points for met
reproducibility and search functionality issues. Google Scholar criteria. The quality score of each individual study is calculated
TABLE II tem evolution involves prediction as the mechanism applied to
I NCLUSION AND EXCLUSION CRITERIA ecosystem evolution as the factor.
Existence Criterion Absence Outcome types are categorized using terminology suggested
IC1 Publication date within the defined time frame. EC1 by Engel and Ebel [28], which categorized data-driven service
A primary study, not a secondary or a tertiary innovation outcomes using an analytics maturity model that
IC2 EC2 starts with explorative capabilities, then adds validative, and
study.
Published in a peer-reviewed journal, generative capabilities as the maturity level increases. Figure
IC3
conference proceeding, or book chapter not in
EC3 1 shows the possible analytics outcomes according to Engel
grey literature reports, blog posts, or
non-academic publications.
and Ebel [28]. Explorative outcomes result from ”the use
of data and analytics for discovering opportunities, such as
IC4 Published in English. EC4
needs, trends, ideas, or design options” [28, p. 9], and employs
Not a duplicate or a version of another
IC5
included study.
EC5 techniques such as visualization and exploratory data analysis.
Discusses platform ecosystems or have a
Validative outcomes result from using data and analytics
IC6 significant component related to platform EC6 for concept testing, hypothesis validation, or understanding
ecosystems. future events using historical data. Generative outcomes result
Discusses analytics and data-driven methods from using data and analytics to generate new value in
IC7 with the objective of providing valuable EC7 the form of designs, artifacts, recommendations, decisions,
outcome to a platform ecosystem.
among others[28]. The three maturity stages can be mapped to
IC8 Focuses on business contexts. EC8
corresponding stages in common analytics maturity models. A
by summing its scores in each criterion, dividing the sum by fundamental model is Gartner’s analytics maturity model [29],
the maximum possible score for all criteria combined, and then which describes four levels of analytics maturity:
multiplying the result by 100. Studies with quality scores less • Descriptive analytics.
than 60% will be excluded from this review. The assessment • Diagnostic analytics.
criteria used for this review are listed in table III [26]. The • Predictive analytics.
quality assessment of the search results is to be carried out • Prescriptive analytics.
by two researchers. In case of disagreement, opinions of the
The first two stages in Gartner’s model can be mapped to the
other two researchers are considered to reach a resolution.
explorative analytics outcomes. The second and third stages
TABLE III
Q UALITY A SSESSMENT C RITERIA in Gartner’s model correspond to validative and generative
analytics outcomes, respectively.
QC1 Objective sufficiently described?
QC2 Study design evident and appropriate?
QC3 Data sources described and appropriate?
QC4 Subject characteristics sufficiently described?
QC5 Data size appropriate?
QC6 Analytics methods described/justified and appropriate?
QC7 Means of outcome assessment reported?
QC8 Results reported in sufficient detail?
QC9 Conclusions supported by the results?

E. Data Extraction Fig. 1. Possible analytics outcomes [28].

The data extraction process was designed to provide enough F. Data Analysis and Synthesis
relevant data for the data synthesis step. In parallel, the outline
of the coding scheme is laid out during the data extraction Descriptive and thematic analyses and syntheses will be
process in order to ensure it was grounded in the literature. conducted to present insights from the extracted data. A coding
The following data categories and the underlying attributes scheme was created for themes in each of the following
will be extracted: dimensions: software ecosystem types; analytics methods;
analytics objectives; types of outcomes; and factors affected
• Identifiers: study title; authors; publication year; abstract.
by the outcomes. Tabular forms and graphs will be utilized to
• Context: ecosystem type; industry; business model.
display the synthesis results.
• Method: main techniques; data sources; objectives.
• Outcome: types; mechanisms; factors affected. IV. I MPLEMENTATION
”Mechanisms” refer to the specific tasks that analytics must
A. Search Execution
perform to achieve the desired outcome. ”Factors” refer to
the attributes and components of the ecosystem to which the The search was executed in the databases and search sys-
mechanisms will be applied. For instance, predicting ecosys- tems mentioned in section III-B2. The search was carried out
using the search terms defined in section III-B1. The search supplement the required details. The process was carried out
in Scopus resulted in 577 hits, while ACM Digital Library by two researchers, and the results were cross-validated. Dis-
and IEEE XPLORE provided 77 and 100 hits, respectively. crepancies between the two researchers were resolved through
The metadata of the 754 studies were retrieved from the discussion and consensus. In cases where a consensus could
corresponding sources. not be reached, the other two researchers were consulted to
reach a resolution. Once the data extraction was completed, the
B. Application of Inclusion and Exclusion Criteria collected data were compiled into a single sheet, using headers
The inclusion and exclusion criteria defined in table II were corresponding to the specific data points. The extracted data
applied to the search results. The inclusion and exclusion were made available in [30].
decisions were made based on the title and the abstract of each TABLE VI
S UMMARY OF THE R EVIEW P ROCESS
search result. In cases where a clear inclusion or exclusion
decision could not be reached using the title and the abstract, Number of studies per Decision
Stage
the full text was retrieved, and a full-text inspection was Proceed Exclude
carried out. Table IV shows the number of exclusion decisions Search Execution 754 0
that each criterion was the most prominent basis for. Some of Inclusion and Exclusion 66 688
Quality Assessment 56 10
the exclusion decisions that were made based on EC6 were
Reviewed 56
due to the fact that platform ecosystems were discussed as the
intervention not the population of the respective studies. The
application of the inclusion and exclusion criteria resulted in V. R ESULTS
the exclusion of 688 search results. The remaining 66 studies
A. Descriptive Synthesis
were included for further review.
TABLE IV The reviewed studies span from 2006 to 2023, with 60%
N UMBER OF EXCLUSION DECISIONS PER CRITERION published in or after 2018. Regarding platform ecosystems
Criterion EC1 EC2 EC3 EC4 EC5 EC6 EC7 EC8 types, 27 studies did not have a specific focus on a certain
Exclusions 0 27 3 2 147 212 295 2 type of platform ecosystems. Table VII lists the distribution
of platform ecosystems types across the reviewed studies.
C. Application of Quality Assessment Criteria While the studied platform ecosystems were spread across
A quality assessment of the 66 studies included in our various industries, 18 studies did not have a specific focus
review was conducted according to the protocol outlined on a certain industry. Platform ecosystems operating in the
in section III-D. After applying the adapted SQAC criteria, software industry were the subject of 12 studies. Platforms
quality scores were assigned to each study. The average quality in the research industry came second as the subject of five
score of the assessed studies was 71.3%. The distribution of studies, and manufacturing came third with two studies only.
quality scores within various percentile ranges is shown in Other industries, such as tourism, virtual reality, and finance,
table V. The quality assessment process identified 10 studies appeared sparsely. While 17 studies did not focus on platform
with quality scores below the 60% threshold. These studies ecosystems of a specific business model, 39 studies discussed
were excluded from the review due to concerns about the specific ones. Open-source and freemium business models
validity and reliability of their findings. The remaining 56 were discussed in 15 and 12 studies respectively. Transaction-
studies were deemed to have met a satisfactory level of based business models were the focus of seven studies. Other
methodological rigor and were thus included in later review business models such as PaaS and advertising-based were
steps. The average quality score of the included studies after discussed in a few number of studies.
removing the disqualified studies was 77.1%. TABLE VII
N UMBER OF REVIEWED STUDIES PER PLATFORM ECOSYSTEM TYPE
TABLE V
N UMBER OF STUDIES PER QUALITY PERCENTILE RANGE Platform Ecosystem Type Number of Studies
Two-sided platform ecosystems 19
Range 0 - 60% 60 - 70% 70 - 80% 80 - 90% 90 - 100% Innovation platform ecosystems 8
10 14 22 16 4 IoT platform ecosystems 2
Regarding analytics and data-driven methods, the reviewed
D. Data Extraction Process
studies employed 63 different methods, algorithms, and tech-
The full texts of the 56 studies were retrieved from their niques from 13 domains to devise their analytics approaches.
respective sources. A manual inspection was carried out to Analytics were applied using standalone methods in 19 studies,
extract the data points outlined in section III-E into a stan- the other 37 (66%) of the reviewed studies used combina-
dardized data extraction template sheet. During the inspection tions of methods in the form of ensembles or pipelines. A
process, we encountered instances of incomplete or ambiguous breakdown of the analytics methods used in the reviewed
information related to terminology, entities, or organizations. literature will be provided in section VI. Concerning data
In these cases, we conducted a thorough investigation using sources, 28 studies utilized batch data. APIs, either alone or
other literature sources and search engines to verify and in combination with web scraping, were used in 23 studies.
Web scraping was the sole data source in five studies. As for domains provided over 60% of the analytics toolbox, and
analytics objectives, decision support was an objective in 34 featured in 32 (57%) of the reviewed studies.
studies. Health evaluation was the objective in eight studies. 8) Networks and Graph Algorithms: Nearly 40% of the
Activity prediction, stakeholders clustering, and component reviewed studies employed methods centered on the analysis
recommendation each was the objective in five studies. of networks and their underlying structures. For example,
The analytics outcomes for the studied platform ecosystem graph algorithms were employed in 16 studies, social network
spanned the three dimensions mentioned in section III-E. Re- analysis in eight studies, and complex network analysis in
garding types of outcomes, explorative outcomes were present seven studies.
in 27 studies, compared to 25 with validative outcomes, and 13 9) Scarcity of Embedded Analytics: Only five (9%) of
with generative outcomes. Various mechanisms for realizing the reviewed studies presented an approach where analytics
outcomes were observed in the reviewed literature. Prediction were embedded within platform ecosystems as an integral
was the prevalent mechanism, applied in 24 studies. Evaluation component or service. This is also reflected through the rare
was the next most common mechanism, appearing in 11 application of methods that handle data streams, such as online
studies. Other mechanisms, such as monitoring, clustering, learning techniques, which were applied in two studies only.
and optimization, were less frequently mentioned. As for the
10) Reliance on Time-lagged Data Fetching: Only 16%
factors that these mechanisms were applied to, ecosystem
(9) of the reviewed studies leveraged streaming-capable data
evolution was the most commonly discussed factors, appearing
fetching methods, such as dedicated API endpoints. In con-
in 16 studies. Ecosystem stakeholders, activity, generativity,
trast, nearly 60% (33) of the reviewed studies relied on time-
and health were present in around 10 studies each. Other
lagged methods, such as batch processing or web scraping.
factors, such as ecosystem performance, architecture, and
A quarter of the reviewed studies relied on a combination of
stability, each were main factors in less than five studies.
both APIs and web scraping for data fetching.
B. Thematic Synthesis 11) Imbalanced Analytics Objective Focus: Over 64% (36)
of the reviewed studies had decision support as the analytics
1) Limited and Imbalanced Platform Type Focus: The objective or motivation. The next most common objective,
reviewed studies focused only on three types of platform health evaluation, appeared in 8 studies only.
ecosystems. While 27 studies did not focus on a specific 12) Limited Progression to Generative Analytics: Over
type, 19 (65%) of the other 29 studies focused on two-sided 75% (43) of the reviewed studies could achieve explorative
platform ecosystems—which connect two distinct user groups or validative analytics outcomes. However, only 13 studies
to interact and transact directly. could break through generative-level analytics outcomes.
2) Variety of Industries: The reviewed literature studied
platform ecosystems within 16 different industries with vary- VI. D ISCUSSION
ing emphasis. This is consistent with the frequently stated
A. Discussion of the Research Question
observation that platform ecosystems are penetrating new
industry domains in recent years [5]. We embarked on this systematic literature review aiming to
3) Limited Business Model Focus: Only six business mod- answer the following research question:
els were represented in the reviewed studies, with advertising- RQ1 What analytics and data-driven methods and techniques
based and PaaS business models mentioned only once each. are used for monitoring or improving outcomes within
4) Multi-domain Toolbox: A total of 63 methods, algo- platform ecosystems in a business environment?
rithms, and techniques from 13 domains were used in the To answer RQ1, table VIII provides a breakdown of the
reviewed studies. More than 85% of the reviewed studies used analytics toolbox used within platform ecosystems based on
a multi-domain analytics approach. the results of this systematic literature review.
5) Ensembles and Pipelines: While 7 studies used a single
standalone analytics method, 49 (87.5%) of the reviewed B. Analytics Maturity Roadmap for Practitioners
studies used combinations of methods in ensemble or pipeline
In addition to the aforementioned descriptive and the-
form as their main analytics approach.
matic syntheses, aspects related to data engineering, analytics
6) Analogy with Biological Ecosystems: Methods from methodology, and outcomes in the data collected throughout
computational systems biology domain were employed in 15% the review process were further synthesized using grounded
of the reviewed studies, making it the fourth most utilized theory. The synthesized data were mapped to the types of
domain. This is consistent with the frequently drawn analogy outcomes feasible as the analytics and data engineering in-
between platform ecosystems and natural ecosystems regard- frastructure advance through maturity levels. The resulting
ing dynamism, evolvability, and interactivity aspects [31]. mapping was outlined in a roadmap spanning those maturity
7) Data Science and Artificial Intelligence: Out of the 13 levels as shown in table IX. The aim of that roadmap is to
domains that the analytics methods belonged to, nine were provide practitioners with a foundational understanding of the
associated with artificial intelligence and data science. These steps and phases that will be encountered in order to unlock the
TABLE VIII business models, such as Data-as-a-Service, were not present.
P LATFORM ECOSYSTEMS ANALYTICS TOOLBOX The prevalence of open-source business model in the reviewed
Data Preprocessing Data Mining, Data Integration, Text Mining,
studies highlights the need for addressing analytics applica-
and Feature Exploratory Data Analysis, Mining Software tions in proprietary enterprise businesses in future research.
Engineering Repositories
3) Expanding Methodological Analogies: The reviewed
Correlation Analysis, Association Rule studies effectively drew methodological analogies, especially
Association and Pattern
Learning, Apriori Algorithm, Formal Concept
Recognition with natural ecosystems, to achieve analytics objectives.
Analysis, Concept Lattice
Topic Modeling, Latent Dirichlet Allocation Methodological analogies to other fields, such as economics
Natural Language and social sciences, remain an underexplored area.
(LDA), Sentiment Analysis, Adaptive
Processing (NLP)
Grammar, Multimodal Grammar Inference
4) Advancing Network and Graph Approaches: Despite
Ontology-based Clustering, K-means
Clustering Algorithms Clustering, Graph Community Detection,
being integral to the analytics approaches throughout the
Graph Clustering reviewed literature, and being ”the most common way of rep-
Multi-sided Recommender Systems, resenting platforms and their complements” [32, p. 12], only
Recommender Systems
Collaborative Filtering, Content-based Filtering a limited portion of the spectrum of network and graph-based
Multi-variate Regression, Ordinary Multivariate techniques was explored. Notably, an advanced technique such
Regression and Least Square, Time-series Analysis, Gray as graph neural networks appeared in one study only.
Prediction Prediction, Vector Autoregressive Model with
Exogenous Variables (VARX) 5) Embedded and Real-time Analytics: Real-time analyt-
Sequence Modeling and Multi-layer Hidden Markov Models, Markov ics techniques, such as online learning, and real-time data
Analysis Decision Processes, Event Sequence Analysis processing techniques, such as stream processing, were rarely
Graph Neural Networks, Fuzzy Cognitive present in the reviewed literature. Only a few studies presented
Neural Networks and Maps, Deep Neural Networks, Online Learning, an approach where analytics were embedded and integrated
Deep Learning Transformer Models, Encoder-Decoder Models,
Attention Mechanisms
into the platform ecosystem. As advanced analytics pipelines
and infrastructure are essential for advanced outcomes, future
Optimization and Data Envelopment Analysis, Projection Pursuit,
Efficiency Analysis Genetic Algorithms research should focus on this underexplored area.
Social Network Analysis, Complex Network 6) Widening the Scope of Analytics Objectives and Out-
Network and Graph Analysis, Graph Algorithms, Topological comes: The imbalanced focus on decision support as the pri-
Analysis Analysis, Knowledge Network Analysis, Graph
Clustering, Graph Community Detection mary analytics objective of the reviewed literature is consistent
with the prevalence of explorative and validative analytics
Domain-Specific Languages, Multi-agent
Modelling and Simulation, Cellular Automaton Simulation, outcomes. Thus, a virtuous loop can be initiated by giving
Simulation Monte-Carlo Simulation, Game Theory-based more research focus to generative-level analytics, which, in
Simulation, Complex Network Simulation turn, will unlock a wider scope of analytics objectives.
Co-expression Genetic Networks, Real-coded
Computational Systems Genetic Algorithms, Metabolic Network
Biology Analysis, Flux Balance Analysis, Extreme
Pathway Analysis D. Threats to Validity
API Quality Metrics, Health Metrics,
Quantitative Metrics
Generativity Metrics, Maturity Models 1) Threats to Internal Validity: While the search strategy
was designed to be comprehensive, it is possible that some
full potential outcome of their platforms’ data. It is meant to relevant studies were not included due to the limitations of the
be a starting point offering the advantage of hindsight provided search terms used or the databases searched. The exclusion
through the current literature on platform ecosystems analytics. of studies not indexed in the three databases used could
This roadmap has a limited scope and does not cover aspects introduce a selection bias. In addition, despite adhering to
such as cultural and organizational change which were neither quantifiable metrics, the quality assessment process relied on
obtainable nor within the focus of this review. the researchers’ judgments. Although measures were taken to
ensure consistency, there may be some degree of variability in
C. Underexplored Research Areas
the assessment of the quality of the reviewed studies.
1) Diversification of Platform Type Focus: Two-sided plat- 2) Threats to External validity: The review offers an in-
form ecosystems have been the focus of a significant propor- depth exploration of analytics and data-driven methods within
tion of the reviewed studies, leaving other types underexplored. platform ecosystems, focusing on a specific set of studies
Future research could, therefore, benefit from diversifying the identified through targeted databases and search terms. Con-
types of platform ecosystems under study, such as multi-sided, sequently, the findings are particularly relevant to this context.
innovation, and integrated platforms. However, this focus might limit the generalizability of the
2) Diversification of Business Model Focus: The limited results to different ecosystems or related fields. The potential
range of business models represented in the reviewed studies for selection or publication bias, inherent in the process of
points to another underexplored area. Advertising-based and study selection, could further constrain the applicability of the
Platform-as-a-Service models were rarely represented, other findings beyond the specific context of this review.
TABLE IX
P LATFORM E COSYSTEMS A NALYTICS M ATURITY ROADMAP (PEAMR)
Generative ∪ Validative ∪ Explorative

Collect data points of direct relevance only. Collect all available data points. Link collected data.
Fetch/store data from/to storage silos. Setup data lakes and warehouses. Support high-depth graph databases.
Clear data licensing and access issues. Enable data querying and slicing. Enable graph queries and traversals. data
Validative ∪ Explorative

Infer data schema to adapt data processing. Integration layer for schema conversion. Enrich data using internal/external sources.
Preprocess data using custom scripts. Adaptation protocols for each source. Enable in-memory computing.
Explorative

Enable localized analytics. Enable holistic analytics. Enable self-service analytics.


Set-up ad-hoc analytics pipelines. Set-up standard analytics services. Integrated analytics-as-a-service. analytics
Update model and techniques per request. Periodic model and techniques update. Real-time models and techniques update.
Local accessibility within a group. Accessible across the organization. Integrated and continuously accessible.
Monitoring ecosystem metrics and state. Prediction of ecosystem metrics and state. Platform performance optimization.
Insights on patterns and trends Evaluation of decisions Generation of designs/artifacts and decisions. outcome
In the form of dashboards and reports Validation of hypotheses Recommendation of artifacts/actors.
Available through APIs. Integrated in user interfaces.

3) Threats to Construct Validity: The systematic approach R EFERENCES


to data extraction, descriptive synthesis, and thematic synthesis
was designed to minimize bias and error. However, the inter- [1] M. de Reuver, C. Sørensen, and R. Basole, “The Digital Platform: A
pretation of study findings inherently involves some level of Research Agenda,” Journal of Information Technology, vol. 33, Apr.
subjectivity that affects the synthesis processes, which should 2017.
[2] E. Laine, O. Alhava, A. Peltokorpi, and O. Seppänen, “Platform ecosys-
be considered when interpreting the results of this review. tems: Unlocking the subcontractors’ business model opportunities,” in
4) Threats to Temporal Validity: The findings of this review IGLC - Proc. Annu. Conf. Int. Group Lean Constr., Brilakis I., Walsh
K., and Sacks R., Eds. The International Group for Lean Construction,
are based on the existing body of literature at the time of the 2017, pp. 177–184.
study, but the relevance of the findings may diminish over [3] A. Ghazawneh and O. Henfridsson, “Balancing platform control and ex-
time as further research is conducted and newer studies are ternal contribution in third-party development: The boundary resources
model,” Information Systems Journal, vol. 23, no. 2, pp. 173–192, 2013.
published. The identification of underexplored research areas, [4] A. Tiwana, “Evolutionary Competition in Platform Ecosystems,” Infor-
however, provides a forward-looking perspective, helping to mation Systems Research, vol. 26, no. 2, pp. 266–281, Jun. 2015.
mitigate this potential limitation. [5] ——, Platform Ecosystems: Aligning Architecture, Governance, and
Strategy. Newnes, Nov. 2013.
[6] “App Store.” [Online]. Available: https://www.apple.com/app-store/
[7] “Android and Google Play statistics, development resources and
intelligence.” [Online]. Available: https://www.appbrain.com/stats
VII. C ONCLUSION [8] F. Schüler and D. Petrik, “Objectives of Platform Research: A Co-
citation and Systematic Literature Review Analysis,” in Management
Digitaler Plattformen, ser. ZfbF-Sonderheft, M. Seiter, L. Grünert, and
A. Steur, Eds. Wiesbaden: Springer Fachmedien, 2021, pp. 1–33.
The findings of this systematic literature review provide [9] T. H. Davenport and J. G. Harris, Competing on Analytics: The New
a comprehensive survey of analytics and data-driven meth- Science of Winning. Harvard Business Press, 2007.
[10] O. Barbosa, R. P. dos Santos, C. Alves, C. Werner, and S. Jansen,
ods applications within platform ecosystems. Those findings “A systematic mapping study on software ecosystems from a three-
were distilled through reviewing relevant studies on platform dimensional perspective,” Software Ecosystems, pp. 59–81, Apr. 2013.
ecosystems analytics. The select group of studies were the [11] N. Staub, K. Haki, S. Aier, and R. Winter, “Governance Mechanisms
in Digital Platform Ecosystems: Addressing the Generativity-Control
result of an advanced search through three different databases Tension,” Communications of the Association for Information Systems,
using a comprehensive set of search terms. The initial search vol. 51, no. 1, pp. 906–939, 2022.
results were further evaluated to exclude false positives and [12] O. Müller, I. Junglas, J. vom Brocke, and S. Debortoli, “Utilizing big
data analytics for information systems research: Challenges, promises
irrelevant results. Following a quality assessment process, a and guidelines,” European Journal of Information Systems, vol. 25, no. 4,
final set of 56 studies underwent data extraction, descriptive pp. 289–302, Jul. 2016.
synthesis, and thematic synthesis processes. The results of [13] C. Tian, R. Cao, H. Zhang, F. Li, W. Ding, and B. Ray, “Service analytics
framework for web-delivered services,” in 2008 IEEE International
these processes enabled us to answer the research question by Conference on Service Operations and Logistics, and Informatics, vol. 1,
providing a practical overview of the analytics and data-driven Oct. 2008, pp. 635–640.
methods used within platform ecosystems. An actionable an- [14] B. Snyder and B. Curtis, “Using Analytics to Guide Improvement during
an Agile–DevOps Transformation,” IEEE Software, vol. 35, no. 1, pp.
alytics maturity roadmap was presented to provide a pathway 78–83, Jan. 2018.
for practitioners to progress through analytics maturity levels [15] P. Malcher, O. Barbosa, D. Viana, and R. Santos, “Software Ecosystems:
within their organizations. The findings included an actionable A Tertiary Study and a Thematic Model,” 2022.
[16] J. P. T. Higgins, J. Thomas, J. Chandler, M. Cumpston, T. Li, M. J.
summary of underexplored research areas to help expand the Page, and V. A. Welch, Cochrane Handbook for Systematic Reviews of
existing body of literature on platform ecosystem analytics. Interventions. John Wiley & Sons, Sep. 2019.
[17] B. Kitchenham and S. Charters, “Guidelines for performing Systematic
Literature Reviews in Software Engineering,” vol. 2, Jan. 2007.
[18] D. Gough, S. Oliver, and J. Thomas, An Introduction to Systematic
Reviews. SAGE, Mar. 2017.
[19] M. Gusenbauer and N. R. Haddaway, “Which academic search systems
are suitable for systematic reviews or meta-analyses? Evaluating retrieval
qualities of Google Scholar, PubMed, and 26 other resources,” Research
Synthesis Methods, vol. 11, no. 2, pp. 181–217, 2020.
[20] M. Visser, N. J. van Eck, and L. Waltman, “Large-scale comparison
of bibliographic data sources: Scopus, Web of Science, Dimensions,
Crossref, and Microsoft Academic,” Quantitative Science Studies, vol. 2,
pp. 1–37, Jan. 2021.
[21] A. Gawer and M. A. Cusumano, Platform Leadership: How Intel,
Microsoft, and Cisco Drive Industry Innovation. Harvard Business
School Press, 2002.
[22] G. G. Parker and M. W. Van Alstyne, “Two-Sided Network Effects: A
Theory of Information Product Design,” Management Science, vol. 51,
no. 10, pp. 1494–1504, Oct. 2005.
[23] M. L. Katz and C. Shapiro, “Systems Competition and Network Effects,”
Journal of Economic Perspectives, vol. 8, no. 2, pp. 93–115, Jun. 1994.
[24] J. Singh, “Critical Appraisal Skills Programme. CASP Appraisal Tools,”
Journal of Pharmacology and Pharmacotherapeutics, vol. 4, p. 76, Jan.
2013.
[25] L. M. Kmet, R. C. Lee, and A. H. F. f. M. Research, Standard Quality
Assessment Criteria for Evaluating Primary Research Papers from a
Variety of Fields. Alberta Heritage Foundation for Medical Research,
2004.
[26] L. Kmet and R. Lee, “Standard Quality Assessment Criteria for Eval-
uating Primary Research Papers from a Variety of FieldsAHFMRHTA
Initiative20040213,” HTA Initiative, vol. 2, Jan. 2004.
[27] B. Kitchenham, “Procedures for Performing Systematic Reviews,” Keele,
UK, Keele Univ., vol. 33, Aug. 2004.
[28] C. Engel and P. Ebel, “Data-Driven Service Innovation: A Systematic
Literature Review and Development of a Research Agenda,” in Proceed-
ings of the 27th European Conference on Information Systems (ECIS),
ser. ECIS ’27, Stockholm & Uppsala, Sweden, Apr. 2019.
[29] N. Rayner and K. Schlegel, “Maturity model overview for business
intelligence and performance management,” Gartner, Stamford, 2008.
[30] S. Hegazy, “Extracted Data from Reviewed Studies for Platform
Ecosystems Analytics SLR,” May 2023. [Online]. Available: https:
//zenodo.org/record/7930786#.ZGLzOqVByUk
[31] D. Dhungana, I. Groher, E. Schludermann, and S. Biffl, “Software
ecosystems vs. natural ecosystems: Learning from the ingenious mind
of nature,” in Proceedings of the Fourth European Conference on
Software Architecture: Companion Volume, ser. ECSA ’10, Copenhagen,
Denmark, Aug. 2010, pp. 96–102.
[32] C. Baldwin and C. J. Woodard, “The Architecture of Platforms: A
Unified View,” Platforms, Markets and Innovation, Sep. 2008.

View publication stats

You might also like