The planning phase describes planning of the SLR, formulation of research questions, scope definition, and development of inclusion and exclusion criteria as recommended by Kitchenham’s methodology ( Kitchenham et al., 2007 ).

The identification of attributes for hyper-personalization is necessary for telcos to understand what products and services would benefit most to the customers. This paper searches for attributes discussed in bibliographic journal articles that describe customer preferences. The search method itself is based on an SLR guideline ( Kitchenham et al., 2007 ). The guidelines ensure that the SLR is conducted carefully so that all information related to a subject is duly considered.

The specific research questions are:

1) What are the primary digital personality attributes that have been identified in existing literature?

2) What digital personality attributes relevant for hyper-personalization in telecommunications be developed through a SLR process?

3) How can a thematic analysis of the reviews be conducted to identify themes within the digital personality attributes amenable for hyper-personalization in the telecommunications industry?

To address Research Question 1, several databases were used to determine full text articles that had some bearing on the attributes pertinent to personalization or hyper-personalization. To address Research Question 2, the scope of the articles pertinent to the telecommunications industry was considered and only those related to the telecommunications were advocated. With respect to Research Question 3, a simple numerical “groundedness” was used to evaluate the frequency of occurrence of the attribute. This frequency of occurrence would give an indication of the importance of the attribute ( Keele, 2007 ).

The research questions above are based on three viewpoints:

1) The population, that is pertinent to the intervention, i.e., the telecommunications users.

2) The intervention, that is the transition from personalization to hyper-personalization.

3) The outcome, of hyper-personalization factors that will be used to compare the intervention.

The population referred to here is the group of telecommunications users comprising all levels of digital products and services. The intervention pertinent in the search documents would be associated with the changes from a group customization to individual customization. The outcomes represent the factors that improve a utility function such as reduced cost, reduced time-to-market, or higher customer satisfaction.

The search strategy was aimed at identifying available literature since the required documents are distributed across many databases, it is important to search in multiple database ( Rethlefsen et al., 2021 ). The search process was a manual search of four databases: Scholar, Science Direct, Springer-Link and Wiley. It is recommended that a minimum of two to three databases is used ( MacMillan et al., 2019 ). The database web links are given in Table 1 . The databases were searched one by one over a period of 3 days from 10 October 2023 to 12 October 2023.

The approach to derive the search strategy was developed from the research questions which were divided into three parts, namely Topic, Issue, and Scope as shown in Table 2 .

Table 1. Databases used in the SLR study.

Table 2. Topic, Issue, Scope in keyword search string synthesis.

Such a method utilized the idea of Topic to represent an independent variable (IV), Issue to represent the dependent variable (DV), with the Scope to represent a scope variable (SV). For example, a keyword search string was synthesized as shown in Table 3 .

Table 3. Keyword synonyms.

This method of keyword building provided a highly reproducible search algorithm that could be quickly adapted and adopted to search a large number of keyword search string/phrase using synonyms of the base keywords. The list of words was drawn using synonyms, alternative spelling, and other variations. The search terms were started with the keywords “hyper-personalization” concatenated with “in telecommunications”. The literature was selected for having the keyword phrase “hyper-personalization in telecommunications”. The search phrase synthesis method was further adapted to add pre-text phrases, or post-text phrases that would qualify the search phrase more precisely. This method suggested by the author ensures that the sample frame of documents in the search foci is highly repeatable and reproducible. The Boolean synthesis of search phrase is shown in Table 4 .

The search process was a manual search of peer reviewed journal papers and conference proceedings since 2023. Each paper and proceeding were reviewed for the presence of the keyword. Literature that was related to the keywords was identified as potentially relevant. Using the search syntax Row 1, Table 4 in the search results are shown in Figure 1 .

Table 4. Boolean synthesis of search phrase.

Using the search syntax of Row 6, Table 4 the search results are reduced to 86 as shown in Figure 2 . The search syntax when used in this way provided a very effective method of identifying documents highly relevant to the study.

By continuously restricting the search syntax, the number of documents was reduced as shown in Table 5 .

Table 6 shows a sample of output counts for the final search string “identifying predictive digital demography attributes for hyper-personalization in telecommunications using artificial intelligence for customer satisfaction from: 2023”.

The search process using the same or similar keywords was extended to the remaining databases. Although the databases had different search templates, the goal was to retrieve relevant documents. The total number of documents retrieved for the different databases is compared in Table 7 and shown graphically in Figure 3 . It is notable that Google Scholar extracted the largest number of documents for

Table 5. Search string refinement for zooming into relevant documents.

Table 6. SLR documents retrieved from database.

Table 7. SLR documents retrieved from database.

a given search string.

The output from the keyword search was then subjected to the following inclusion and exclusion criteria. The inclusion criteria required that the document had a complete research process with defined research questions, data collection, data analysis and conclusion. The exclusion criteria are that the document had no defined research objective, unclear research process, or abstract. In this format, the inclusion and exclusion criteria could be easily added, edited, or deleted. as shown in Table 8 .

Table 8. Selection criteria.

The selection criteria are an important aspect of conducting an SLR to ensure that articles are selected according to a set criterion ( Kitchenham et al., 2007 ). When the search outputs were available, the articles were:

1) Noted for the search database used. The search databases are listed Table 6 .

2) Keyword search was conducted. All articles were checked to ensure that the keywords were related to the topic with only two options available as shown in Table 8 .

3) The title was deliberated to ensure that it was related to the topic. The role of the correct search sentence was important to return a relevant list of articles. Title mismatch, duplicate titles, or outdated articles were excluded. Articles could be excluded for one of the reasons possible as shown in Table 8 .

4) The abstract was read carefully to ensure that the article was closely related to the intent of the SLR being conducted. Articles could be excluded for one of the several reasons possible as shown in Table 8 .

5) The full text was reviewed in detail to ensure that the detailed contents of the article were pertinent to the topic of SLR being conducted. Articles could be excluded for one of the several reasons possible as shown in Table 8 .

The selection process identified the articles necessary for the SLR of the topic as shown in Table 11 . In the author’s opinion, based on this progression of keyword, title, abstract, and full text selections, the excerpts selection must be considered as a continuation to the next level of selection.

The study selection was a 5-stage process as shown in Table 9 . Where studies could be easily associated with the inclusion criteria, the study was selected objectively.

1) In Stage 1, articles with matching keywords were included. The initial keywords were broad-based that returned a large number of articles.

2) In Stage 2, titles were easily scanned for context and irrelevant documents were excluded. For example, a search on hyper-personalization could result in some articles on disease prevention. Only articles closely related to hyper-personalization for telecommunications were accepted.

3) In Stage 3, the abstracts were studied to understand the contents for matching with the requirements described in the objectives.

4) In Stage 4, the full texts of the documents were studied and sufficient comprehension was necessary for selection. The full text articles were selected on the various aspects of hyper-personalization discussed with the presence of attributes sought in the objectives of the research.

5) In Stage 5, the document was searched for attributes that were sought as the ultimate requirements of the SLR. The article was read thoroughly to extract excerpts related to hyper-personalization.

Table 9. Stages of activity.

Of the 86 Google Scholar articles retrieved from the search phrase, only 19 articles met the inclusion criteria for Title as shown in Table 10 .

Similarly, the initial search phrases resulted in Science Direct articles (4), Springer-Link articles (7), and Wiley article (1) as shown in Table 11 .

An electronic data extraction form using Microsoft Excel was created for data collection following the SLR guidelines ( Kitchenham, 2004 ). The form was necessary to ensure accurate recording of the documents obtained from the search. The data extracted from each article was: the source of the article, the author information, year of publication, keywords, title, abstract, and the pdf (portable

Table 10. Partial output of articles from Google Scholar search.

document format) document or the doi (digital object identifier, when only the online version was available).

Next, the list of 33 articles was checked for the Title. If the title was acceptable, the article was retained and marked Title OK, else, the reason for exclusion was noted, as Title mismatch, Duplicate, or Outdated as indicated in Table 8 . Excluded articles were removed from further consideration. Subsequently, the Abstract and Full Text were searched with the available criteria as indicated in Table 8 . Additionally, the pdf documents or the web-link were recorded to easily allow retrieval of document.

The data extraction form was piloted on a sample of preliminary articles to assess both the technical issues (e.g., completeness of the form, usability) and the sequence of information collected. Since only the researcher used the form for recording, the pilot study was assessed by the researcher. The form was then used during the data collection phase. Note that in Table 11 , the objective was to identify all document from the internet that met the selection criteria. Only a short part of the title was reported for clarity.

Table 11. Documents selected for determining hyper-personalization attributes.

Documents collected during the search are systematically selected by the inclusion criteria and deselected by the exclusion criteria. The numbers of documents included and excluded are shown in Table 12 and the PRISMA flow chart is shown in Figure 4 . The PRISMA is the Preferred Reporting Items for Systematic Reviews

Table 12. Searching specific documents.

Table 13. Documents selected for determining hyper-personalization attributes.

and Meta-Analyses. This analysis was done using an Excel template which counted the number of inclusions and exclusions. The selection of the articles based on the numbers of documents is shown in Figure 4 . A summary of the documents retrieved, excluded, and included is shown in Table 13 .

Each article selected for the SLR was evaluated for quality assessment ( Kitchenham et al., 2007 ) using the criteria specified by the Database of Abstracts of Reviews of Effects (DARE) criteria ( University of York, 2014 ).

1) Were the inclusion and exclusion criteria applied on all the documents?

The inclusion and exclusion criteria defined in Table 8 were applied on all the studies.

2) Is the literature search exhaustive to have included all relevant studies pertinent to the objectives of this study?

Four digital databases ( Table 1 ) and another source of study articles were included in the study.

3) Were selected articles reviewed to form a synthesis?

The selected articles ( Table 14 ) were analyzed for excerpts to form a synthesis of attributes for hyper-personalization.

4) Was the quality of the included studies assessed?

The quality of the studies was evaluated by the defined quality criteria and used in the extraction of the study articles.

5) Were selected articles reviewed to have clear purpose, methodology, findings, and conclusions?

The selected articles were reviews for completeness of purpose, methodology, findings, and conclusion in the full text search stage.

Of the 33 studies articles ( Table 11 ) only 10 articles meeting all the inclusion criteria on title, abstract, and full text were identified and referenced as shown in Table 14 . These articles were then used for excerpts collection in the data extraction stage.

Table 14. Documents selected for determining hyper-personalization attributes.

Once a document was deemed to be suitable, the full text data was studied to extract ideas that represented the attributes sought for the hyper-personalization. The ideas were recorded in a word pool and aligned with other ideas. This ideation step was to create a list of content describing customer needs by mapping words through word translation, localization, and content synonym. Thus, customer sentiments described in the article were translated to implied meaning and then to wanted attribute as shown in Table 15 .

Table 15. Translation of reworded data.

The data extraction was conducted on each of the selected article with a schema as shown in Figure 5 . The data extraction was based on a numeric count of occurrences of each attribute and the aggregate of the word group. The count of items would give an indication of the importance of a given attribute and so would the group count.

A simple data extraction form was used for this purpose. Since the data extraction was conducted by the author alone, a simple pilot study for convenience of use was conducted. Since the data was extracted at the moment of reviewing the document, the selection of attributes was exploratory in nature, and the list of attributes was built simultaneously. Complete data extraction was achieved when the data showed a saturation of attributes.

Selected documents were analyzed through a detailed study for any excerpt that resembled an interest towards hyper-personalization. The documents identified by the SLR were studied one-by-one while looking for phrases or sentences that represented customer words with implied meaning related to a wanted attribute.

As the excerpts were identified and collected into meaningful groups named as an attribute suggestive of a variable that could be related to hyper-personalization. Many synonyms of excerpts were possible for an attribute. A collection of attributes implying a similar concept was called a parameter. For example, closely related attributes such as family and age were named Demography.

Flexibility of arranging and rearranging excerpts, attributes, and parameter was the principle that made this Affinity Diagram a qualitative method related to thematic analysis. Affinity diagram ( Widjaja & Takahashi, 2016 ) is similar to the thematic analysis ( Braun & Clarke, 2006 ) and grounded theory where the excerpts are extracted from a document and assigned as codes, constructs, or categories ( Figure 6 ). The schema for the Affinity Diagram is shown in Figure 6 .

The qualitative analysis is highly relevant here for three reasons.

1) Affinity diagram (thematic analysis, or grounded theory) is suited to making claims about how individuals interpret reality.

2) There is no established theory on the process of seeking attributes for hyper-personalization ( Creswell, 2014 ).

3) The qualitative analysis provides a guideline for research data interpretation ( Cullen & Brennan, 2021 ).

The qualitative analysis adopted in this research is based on both the inductive and deductive approaches. The inductive coding allowed new codes to emerge from the excerpts. The deductive coding was deduced from the developing codes. All the codes were driven towards answering the research questions and the research framework adapted for hyper-personalization ( Wicks, 2017 ).

The data extraction process was a cyclic process ( Figure 7 ) with codes emerging into a recognizable theme associated with the research purpose ( Williams & Moser, 2019 ). The constant comparison method was used in the data organizing and refining activity until the code converged to saturation.

To facilitate a dynamic method of recording during the cyclic coding process, an Excel format as shown in Table 16 was used. An example of excerpts from full text and the attribute implied is shown Table 16 using page 1141.

Table 16. Coding via data extraction.

The articles were reviewed using Affinity Diagram. Affinity analysis is a brainstorming technique to organize ideas and concepts. The technique can be represented as a diagram to visualize the organization ( Widjaja & Takahashi, 2016 ). As each article was evaluated, the article was identified and recorded. During the review process, attributes discussed within the article related to the user’s digital information were recorded. Initially, some of the attributes were added based on the “definite” types of attributes. The number of times an attribute was used within the article was recorded. Attributes were added, combined with other attributes, or deleted as each article contributed to the information accumulated. When a number of attributes developed into a meaningful parameter, the attributes were reorganized under the parameter. This process was repeated until a saturated state was obtained.

The Affinity Diagram was used to organize innovative insights and ideas to create new patterns of understanding. The data was collected through the excerpts analysis from full text documents. As each document was read, variables related to hyper-personalization were identified, noted, and recorded as shown in Table 17 . Data collected was arranged and rearranged to form meaningful derivatives and collections of ideas ( Gupta et al., 2016 ). The collection of ideas converged on a concept which was regarded as an attribute for hyper-personalization. In this method a pool of words is considered and each observation (word or phrase) is used to find relationships between the observation and identifying a creative connection through grouping and regrouping of a derived idea to form attributes relevant to hyper-personalization. The affinity diagramming technique is visualized as a mind map ( Iqbal et al., 2022 ) as shown in Figure 8 .

At least four approaches were used to reduce the risk of bias in the study. One, the review process used multiple databases to provide a broad coverage of journals, conferences, and sources. This increases the likelihood of identifying a comprehensive range of documents relevant to the research while reducing the risk of missing studies that might be excluded due to the database indexing policy. Hence, multiple databases were used in this study.

Two, search engines use different policies for ranking algorithms to flag the requested document in terms of relevance, citation numbers, and content. Some content may be seen to be irrelevant to the search engine. For example, a medical database may systematically exclude engineering documents. Hence, multiple search engines were used.

Three, some search engines had their own search templates resulting in a different approach to finding relevant documents ( Rethlefsen et al., 2021 ). This bias was reduced by using multiple databases and multiple search engines.

Four, some seemingly relevant documents were only available behind multiple barriers of signing up, institutional collaboration, library permissions, expensive payment, or requesting the authors, for the full text article. Strict compliance to all these requirements is impractical for the researcher. Consequently, open-access articles tended to be in the selected search list. Nevertheless, this bias was reduced by a more detailed study of the abstract and searching for alternative sources of the document.

Reporting bias is influenced by search engine operations and characteristics of the included studies. Search engines tend to highlight studies that are significant and deemphasize insignificant studies. Search engines also tend to highlight studies with higher number of citations. This leads to a bias favoring highly cited studies and thus not being representative of the literature ( Pautasso, May 2010 ). Often documents with more citations tend to receive even more citations. Studies that did not support an established theory are less likely to be published. Language bias is also possible as many studies are not published in the researcher’s native English language. Language bias may also be associated with regional bias ( Briner & Denyer, 2012 ).

The mitigation to reporting bias in this study includes a comprehensive search to identify both published and unpublished studies, e.g., Google Scholar identified 1290 documents ( Table 7 ) which were reduced to 86 documents ( Table 10 ), and finally to 7 documents ( Table 13 ).

The parameters identified through the affinity diagramming include Demography, Usage Patterns, Performance, Affordability, Device Usage, Purchase decision, Online activity, and Customer Interactions. The attributes for hyper-personalization are listed in Table 17 . Frequency indicates the number of occurrences of the attribute while Total indicates the number of occurrences of the attributes within the parameter. Based on this research, eight parameters were identified ( Table 17 , Figure 9 ) with a total of 32 attributes most associated with hyper-personalization ( Table 18 ).

Based on the thematic analysis, a structured and comprehensive understanding

Table 17. Results of the study.

of the attributes were identified and classified as parameters shown in Figure 10 . The parameters represent excerpts collated into meaning categories and provide a visualization of the attribute groups that highlight the consumer’s digital demography as elucidated in this study. The attribute groups provide an easy reference to the collections of attributes identified in this study. The parameters collectively determine the customers’ demography, their usage patterns, performance expectations, affordability of products and services, device types used, purchase decisions, their online activities, and interactions necessary.

From the results of the study, the excerpts were translated into the attributes associated with hyper-personalization as shown in Table 18 . These attributes represent the digital demography identified as most pertinent to predict the user’s purchasing behavior essential for effective hyper-personalization strategies.

Table 18. Sample attributes identified by SLR.

The first objective of this study was to use an SLR based on the Kitchenham recommendations. The methodology was adapted to identify articles pertaining to hyper-personalization from several databases. An unbiased search was adopted to select the most pertinent articles. The selection of articles was listed in a table that was used to draw the PRISMA flow chart.

The second objective of this research was to apply a Boolean search process to discover primary studies pertinent to hyper-personalization. The systematic search procedure used in this study ensured that the number of articles was progressively reduced to be more selective as permitted in the inclusion and exclusion criteria. This method facilitated the creation of search results that were replicable and reproducible and thus representing a crucial step to reduce the bias in the search method. This was accomplished by having a series of filters that progressively eliminated articles in the sequence Keyword > Title > Abstracts > Full Text.

The third objective was to perform a thematic analysis to gather data that could be applied to hyper-personalization. Carefully selected articles were examined to extract excerpts related to the attributes associated to hyper-personalization. The coding process involved open coding, axial coding, and selective coding. As a result of the thematic analysis, eight parameters consisting of 32 attributes were identified for hyper-personalization. The parameters identified include Demography, Usage Patterns, Performance, Affordability, Device Usage, Purchase decision, Online activity, and Customer Interactions.

Another benefit of this study is that it provides a template for conducting an SLR without the need for specialized software. The study demonstrates a self-reliant methodology that can be easily traced by an individual researcher.

First, the keywords for the SLR were constructed by using permutations of words listed in a table and identified as topic, issues, and scope. This method entailed a phrase that had components of an independent variable, a dependent variable, and a scope variable.

Second, the initial keyword or phrase was developed into a more specific phrase by attaching another word (or phrase) to the anterior or posterior of the current phrase. This ensured that the search space was not changed but trimmed progressively. Such a method ensured that search results did not swing between searches due to the sample space foci.

Third, Boolean concepts of And, Or, and Not were also used in the construction of the keyword and phrase.

Fourth, a table of search output was used to list all the document properties in a tabular format. Articles were downloaded for subsequent inclusion or exclusion. Where the document was not available as a file, the URL for the document was recorded.

Fifth, the search was based on the priority based on Keywords, Title, Abstract, and Full text. This ensured that irrelevant documents were rapidly excluded and a manageable list of documents remained for further analysis. This was an important method when the document selection had to reduce hundreds of documents to a few important ones.

Sixth, the search for the attributes was an extension of the Keywords, Title, Abstract, and Full text. Another spreadsheet format was used to collect the attributes from the excerpts. Attributes were not always found as a keyword but often as a phrase or sentence implying an attribute. Thus, a phrase or sentence was regarded as pointing to an attribute when an underlying attribute was implicated. For example, a sentence or phrase on “provide customized services” could be reworded to “changeable options”. Also, “interact with unhappy guests” could be reworded to “complaint handling”. This approach was necessary to identify synonyms of words phrases with similar tones of meaning.

During the elucidation of factors that are important for hyper-personalization, none of the selected articles were exclusively related to the factors sought. Therefore, the scope of the topics discussed in the paper was taken as an indicative factor. In many cases, the voice of the customer was interpreted to a reworded data, i.e., common idea.

It is notable that for any given search phrase applied in the databases, Google Scholar gave the highest number of documents meeting the search phrase. Search phrases could also be saved in a label. Items in the label could also be saved into a document list as a comma separated value (csv) file. This made it much easier to compile the list of articles from the search.

There are both theoretical and practical implications of the study findings. The theoretical implication is that the research exposed the theoretical underpinning of digital demographic attributes in the context of hyper-personalization for telecommunications. It is a significant contribution to the body of supporting development models that can guide future academic investigations.

The practical implication of the study offers telcos a practical approach to identifying attributes important to implementing effective hyper-personalization strategies. This is achieved by the practical methodology advocated in this study. The experience highlighted through this study can be an enabler to facilitate customer-centric approaches and consequently enhance customer satisfaction, loyalty, and reduced churn.

This study provides a thorough investigation into identifying parameters essential for hyper-personalization in the Telco industry. It describes an objective process of identifying the parameters necessary to achieve matching customer preferences and product or service delivery. These parameters and the associated attributes are intended to form the basis of variables adopted into artificial intelligence prediction models, elevating the degree of product and service matching with customer expectations.

The implications of this research are considered as follows:

This study offers insights into using digital personality attributes to integrate into artificial intelligence applications for more accurate predictions of the customer preferences.

This study contributes to a practical model of how SLR can be conducted in other areas of research. The study will form a basis that guides future research to explore new dimensions of identifying digital personality attributes for hyper-personalization.

This SLR study can provide telcos with a better understanding of the digital personality attributes most relevant to hyper-personalization of their services. This study provides an objective method of determining the attributes in a systematic way rather than a presumed list of attributes.

The findings of this study can be adopted by telcos to tailor their offerings to meet customer needs more effectively and thereby increase customer satisfaction and retention. This will also enable telcos to differentiate themselves by providing hyper-personalized services.

This research also promotes a customer-centric approach by emphasizing the importance identifying attributes important to customers and then adapting hyper-personalization to meeting individual customer needs. This strategy will lead to better customer satisfaction and customer retention.

The search was based on a manual search process using keywords and phrases concatenated by Boolean logic. Since this was limited to certain databases, some articles may be left out. However, this was mitigated to some extent using Google Scholar search which searched more widely across many databases. Also, only the researcher selected the articles and conducted the reviews. The researcher’s interest in the topic could result in some subjective decisions. The subjective decisions were to some extent mitigated by the Study Quality Assessment.

The main limitation in this study was the difficulty to find enough relevant articles that discussed hyper-personalization in the field of telecommunications. SLRs are vulnerable to publication bias as they depend on published literature. Published literature itself is biased with studies that mostly show positive and significant findings while non-significant and negative results are often not published. Furthermore, the quality and relevance of the studies vary widely and some lack the correct scope of hyper-personalization. Heterogeneity of the study methods used in the documents searched also makes it difficult to synthesize or generalize findings. Although these limitations are mitigated with clear inclusion and exclusion criteria, some level of subjectivity still influences the process. Another limitation was the articles were mostly that which were freely available on the internet. As a result, the study mainly included open access articles in this research.