Data collection and analysis

Figure 18: The process of semi-structured interviews (Hirsjärvi & Hurme 2000, 67).

This study follows the process that Hirsjärvi & Hurme (2000) have proposed as the interview themes are formed from the theory presented earlier on this thesis and from the findings of the content analysis. Although the themes were the same on each interview, the questions varied among interviewees to gain comprehensive picture on how the case company has employed service recovery on their social media channels and how service failures and customer complaints are seen from the company’s point of view. The frame for each interview is presented in Appendix 3.

4.2. Data collection and analysis

The empirical data on this study was collected between April and September 2016 first by executing a content analysis on social media and then by conducting three semi-structured interviews with the case company’s personnel. As the case company has customer service on both Facebook and Twitter, these social media channels became the objects of the research. Even though the company is active also on Instagram, YouTube and LinkedIn, these sites are not part of this study as customers rarely communicate on these mediums.

In addition, the case company has two separate pages on Facebook. To gain the most valuable information related to the services the case company offers its private customers, their main Facebook channel was chosen as the data collection source over the other channel that focuses on only one service.

On this study the unit of analysis (Elo & Kyngäs 2008, 109) is a message from the case company’s private customer voicing disappointment towards their operations. In other words, all written messages on the case company’s Facebook and Twitter pages are obtained, whereas only the messages including direct complaints, improvement suggestions and inquiries about postal items were collected (hereafter these messages are referred as original complaints). Next, their Facebook and Twitter pages are explored to gain an overall understanding of the data also providing the headlines for the matrix where part of the data is collected (e.g. gender of complainer, how often the failure has occurred and how fast the case company replies) as Elo & Kyngäs (2008, 109) suggested. During exploring it became apparent that the period of one month is enough to collect the data as it started to saturate. Hirsjärvi et al. (2008, 177) pointed out that saturation is a concept which is used in qualitative research, meaning that the researcher collects data continuously as long as it gives new information related to the research question. Next data from period of one month is collected to Word sheets to make notes whenever an interesting or relevant data is discovered as Elo & Kyngäs (2008, 109-110) and Ye & Tripathi (2016, 3841) pointed out.

Alongside the notes, the relevant data is collected to Excel matrix to analyze it (see examples in Table 4 and Table 5). Next, notes are first divided into groups and then categorized to compare observations to find both similarities and dissimilarities. (Dey 1993;

sited in Elo & Kyngäs 2008, 110) Whereas after the data is collected to the matrix, it is analyzed by theming the topics found from the material and by calculating the facts including for example the number of complaints and how long it takes before the customer receives a reply. This is transformed to figures, tables and diagrams to understand the whole picture of the subject and to further analyze and report the results.

Table 4: Example of data collection matrix: complainer

Table 5: Example of data collection matrix: company

To gain a holistic view on the case company’s service recovery practices from the personnel point of view, three interviews were conducted. Semi-structured interviews included interviews with both the customer service manager and employee along with communications manager. Next figure (Figure 19) adapted from the case company’s annual report demonstrates the area of semi-structured interviews.

Figure 19: Semi-structured interviews at the case company (adapted from Posti Group 2015, 92).

First interview took place at the case company’s office in Helsinki while both second and third interview were done on telephone (see more detailed information of the interviews in Table 6). The themes and questions for each interview are presented in Appendix 3.

Table 6: Conducted interviews

Interviewee Position Date Place

Interviewee 1 manager, customer service August 26th 2016 Helsinki, head office Interviewee 2 manager, communications September 2nd 2016 On the phone Interviewee 3 employee, social media

customer service

September 6th 2016 On the phone

Researchers that have applied qualitative research method into their study, tend to use discretionary sample as the aim is not to generalize the results statistically but to gain knowledge on a certain event or phenomenon. Especially when conducting a case study it is possible to gain significant information through only a few interviews as the amount of observations may still be vast, meaning that the data is qualitatively rich. (Hirsjärvi & Hurme 2000, 58-59) For these reasons it is justified to use a small discretionary sample (n=3) including personnel from various hierarchy levels from different departments. To obtain the most knowledgeable answers related to the subject of the study, the interviewees were selected based on the case company recommendations.

Each interview was recorded and took from less than half an hour up to an hour. By recording interviews the researcher was able not only to listen attentively but also to concentrate on the interview as a whole. As the aim was not to explore linguistic characteristics, interviews were transcribed subsequently leaving out repeated words, non-lexical conversation sounds and pauses. Altogether 29 pages of transcribed material was received from the interviews which were analyzed using inductive approach by categorizing the data. According to Saunders et al. (2009, 492-493) and Valli & Aaltola (2015, 115) data categorizing requires two steps: first categories are created based on either the terms used by the interviewees or both the existing literature and theory and then the data is arranged into the created categories. (Ibid.) This study has adopted conventional content analysis meaning that the codes emerge during the analysis (Hsieh & Shannon 2005, 1286).

First the data that has emerged as a result of transcription is read through several times to achieve a comprehensive understanding. It is then divided to segments by content which is then simplified (Table 7) enabling the researcher to find the key elements. (Hsieh &

Shannon 2005, 1279; Valli & Aaltola 2015, 116)

Table 7: Examples of data reduction

Original sentence Simplified content

To reply customers’ questions the customer service seeks information from our in-house experts to obtain the best possible answer.

Using in-house experts to solve issues.

Customer privacy is one of the top priorities on social media and to keep it that way from time to time communication needs to move from public to private to respect that privacy.

To respect privacy not all problem solving should be public.

After the data has been thoroughly read and the key content has emerged, the concepts were divided into subcategories and then to categories to form the overall theme (see Table 8). (Hsieh & Shannon 2005, 1279; Valli & Aaltola 2015, 116)

Table 8: Examples of data classifying and theming

Concept Subcategory Category Theme

In-house experts help

Next, to prepare a solid phase for reporting, subcategories and categories have been organized as Hsieh & Shannon (2005) suggested. Following chapter provides an overview on reliability, validity and ethics where methodological aspects and choices are thoroughly evaluated.