Exploring the role of decision support systems in promoting healthier and more sustainable online food shopping: A card sorting study

,


Introduction
Climate change and the rising prevalence of obesity are considered major threats to the world population (Edwards & Roberts, 2009). Obesity has been reckoned a global epidemic for decades, with the current number of overweight people worldwide lying around 2 billion, of which 650 million are obese (WHO, 2021a). Climate change, i.e., global warming, has also been reported long ago, but effects have accelerated over the past years (UN, 2020). When it comes to this growing prevalence of obesity and the exacerbation of climate change, the food environment plays a critical role (Notarnicola et al., 2017;WHO, 2021b).
In today's food system, commercial interests drive offerings (Swinburn et al., 2019) and increasingly more unhealthy and ultra-processed foods are offered to generate profits (Baker et al., 2020;Monteiro et al., 2013). On the demand side, mainly accessibility, affordability, convenience, and desirability affect consumer choice (Turner et al., 2018), meaning that healthiness and sustainability of food choices are not necessarily prioritized by consumers (White et al., 2019). In response to the challenges faced by the food system, efforts have been made to promote healthier and more sustainable food choices (Pais et al., 2023). These efforts involve a combination of policy changes and bottom-up approaches that aim to influence consumer behavior with respect to food choices (BEUC, 2015). In recent years, the importance of online grocery shopping has increased, prompting a shift in focus towards approaches that can effectively shape consumer behavior in the online food environment (Granheim et al., 2022). To illustrate, in 2015, one-quarter of the world population said they order grocery products online and around 55% indicated to be willing to shop for groceries online in the future (Nielsen, 2015, pp. 1-35). The COVID-19 pandemic further accelerated this trend of online grocery shopping (Baarsma & Groenewegen, 2021). In 2019, e-commerce accounted for only 4.9% of total retail sales, but this number jumped to 6.5% in 2020 (Statista, 2022). Given the rising popularity of online grocery shopping, the online choice environment has become an important platform to change behavior. While there are various obstacles in the online setting that can prevent consumers from making healthier or more sustainable choices, such as choice overload and consequent suboptimal choices due to the availability of many options (Grandi & Cardinali, 2021;Mick et al., 2004;Wang et al., 2021), the digital platform also has the potential to bring about changes in grocery shopping behavior.
In modifying food choice behavior, various techniques can be used such as social norm messages, setting better default choices, providing incentives, or highlighting the health or sustainability of products (Münscher et al., 2016;Valenčič et al., 2022). Such nudging techniques are particularly effective in online choice environments because they can be adjusted based on collected consumer data, such as clickstreams or browsing history, and implemented in real-time (Bucklin et al., 2002;Shin et al., 2020). Real-time interventions can be tailored to context and product choices (Shin et al., 2020;van der Laan & Orcholska, 2022) and have the advantage of targeting individuals when they are most receptive to change and vulnerable to persuasion (Nahum-Shani et al., 2015). An example of a nudge to intervene in real-time is suggesting consumers alternative product choices based on their initial choices or previous shopping and purchase behavior (Schafer et al., 1999). As a result, retailers can gain more insights in consumer behavior and shoppers are more likely to become engaged with the shopping experience (Bradlow et al., 2017).
In the current paper, a system that provides decision-making support to consumers in the online food shopping environment is referred to as a Food Shopping Support System (FSSS). FSSSs are widely used by online retailers to optimize the shopping process and consumer experience. This approach appears not only to increase the speed of shopping, but also the ease of the choice process for consumers (Abbu et al., 2021). By collecting consumer data online, it is possible to infer consumers' preferences more accurately and guide them towards choices that are consistent with those inferred preferences (Lee & Geistfeld, 1998), consequently increasing consumer satisfaction (Bechwati & Xia, 2003). Nonetheless, research and applications of FSSS techniques mainly focus on improving the consumer choice process to ultimately increase sales and profits of retailers. Often it is not investigated how to stimulate the better options, from the consumer's perspective, such as the healthier and more sustainable options in assortments.
Several recent studies provide first insights into the effects of interventions such as providing real-time feedback on choices (De Bauw et al., 2021;Jansen et al., 2021;Koutoukidis et al., 2019;Shin et al., 2022) and offering the ability to sort by health score (Shin et al., 2020). Results are mixed but show only small effects on consumer choices (Ducrot et al., 2016;Slapø et al., 2021). These studies are often done during short testing periods and in hypothetical contexts. Deeper insights into both the effective elements of consumer interventions and digital possibilities have been viewed in a limited and unintegrated manner.
If we improve our understanding of the possibilities of FSSS techniques to steer consumers' choices in a healthier and more sustainable direction, and of what consumers appreciate and experience in this area, then this knowledge can be used in further development of FSSS techniques. To gain understanding of what hinders and facilitates further development and application of healthy sustainable FSSS in practice, it is essential to involve both experts and consumers from an early stage. Usually, the premise is that experts work on the development of support systems and test whether these systems can help in making better food choices (Elsweiler et al., 2022). Nevertheless, user experiences with new technologies might not fully align with expert views (Webster et al., 2010) and many scholars argue that early consumer input on their perceptions and preferences helps to improve the development and implementation of new products or processes (Prahalad & Ramaswamy, 2000;Van Kleef et al., 2005). In particular, the way of presenting support is important for positive user experiences (Knijnenburg et al., 2012), which in turn determines the performance of support systems (Konstan & Riedl, 2012).
The current study employs an interdisciplinary approach to explore how a sustainable healthy FSSS 1 could be configured to optimally influence consumer decision making in the online shopping environment. It is aimed to achieve an optimal match between the consumer (decision making process to be supported) and technical side (development and implementation of FSSS), as this match is considered a key factor for success of support systems (Wierenga et al., 1999). We address the following research question: "How could a food shopping support system (FSSS) for healthier and more sustainable shopping be configured to optimally influence consumer decision making in the online food shopping environment?". The main question will be answered by addressing the following sub questions: (1) what are consumers perceptions, experiences, and preferences regarding digital support strategies? (2) what are key factors for optimal development and implementation of FSSS according to experts and (3) to consumers? We aim to obtain expert and consumer insights on important factors as well as opportunities and barriers in developing and implementing FSSS for healthier and more sustainable food choices and, consequently, offer a comprehensive agenda for future research. For this purpose, a semi-structured interview protocol including a card sorting task and several open-ended questions is used.

Methods
Important factors as well as opportunities and barriers in developing a support system for sustainable healthy food choices in an online grocery environment were identified by means of exploratory one-on-one expert interviews and consumer focus groups. Focus groups were chosen to generate consumer information because participants were expected to benefit from group interaction (Krueger, 2014). The first author conducted expert interviews in May, June, and July 2022, online via MS Teams using Miro as whiteboard software. Focus groups with the first author as interviewer were held in June 2022. Two were conducted in Wageningen (The Netherlands), one in Didam (The Netherlands) and a fourth focus group was held with people from international backgrounds (online via MS Teams using Qualtrics to capture consumer insights). These areas were chosen based on the anticipated difference in sustainability and health orientation among their inhabitants, with Wageningen being more focused on sustainability and health than Didam. To increase diversity of perspectives and enhance richness of data, an international background group was also included. For all interviews and focus groups, both automatic transcription through MS Teams as well as audio-recordings were made. During the on-site focus groups, an observer was present to monitor group dynamics and speaking order, as a back-up in case audio-recordings would be unclear. Interview language was Dutch or English, depending on the native language and/or preference of the participant(s).
1 Sustainable Healthy FSSS typically refers to sustainability as well as healthiness of foods, meaning that foods promote all dimensions of individuals' health and wellbeing; have low environmental pressure and impact; are accessible, affordable, safe, and equitable; and are culturally acceptable (FAO & WHO, 2019, p.9). In this research, FSSS can promote food choices that are sustainable as well as healthy, but also either sustainable or healthy.

Consumer participants
People were eligible for focus groups if they had actively shopped online before, either for groceries or for other products, and were at least partly responsible for grocery shopping in their household. Furthermore, we included people 18 years of age or older, who did not require a language translator to take part in the discussion and were not considered an expert as defined in section 2.1.2. Purposive sampling was used to recruit a diverse group of people based on gender, age, nationality (for international focus groups), household composition, and educational level to get a broader perspective of consumers. Purposive sampling is a widely adopted practice in qualitative research to achieve an in-depth understanding of a phenomenon (Etikan et al., 2016).
We aimed for four to six participants per focus group and a total of four focus groups to be conducted, as smaller group sizes are recommended to allow people to share their insights and observations thoroughly (Krueger, 2014). After approval was provided by the Social Sciences Ethics Committee (SEC) of Wageningen University & Research, participants were recruited via posting on social networks (LinkedIn), flyers in supermarkets, panel mailing list, and the personal network of the research team. People were informed that they could participate in a group discussion of 90 min (it was expected that focus groups would take a bit longer than expert interviews, see section 2.2 Procedure). Focus group participants, henceforth consumer participants, were financially reimbursed for their time and travel expenses with a Bol.com or Amazon gift card worth €30.

Experts
Experts were defined as persons with specific knowledge in a certain field of action who hold a certain status or exercise a function by which the expert's knowledge has an influence on practice (Bogner et al., 2009). Experts were chosen from different fields to ensure a representation of experts along the stages of the recommendation delivery process that include understanding the consumer, delivering guidance, as well as determining the impact of guidance on the consumer and retailer (Li & Karahanna, 2015). While some overlap of expertise was allowed, it was avoided to include experts with the exact same area of expertise.
To select experts with different areas of expertise, well-known academics in the field of recommender systems, decision aids, digital marketing, and consumer decision making were identified via paper publications. Practitioners working in the field of sustainable and healthy food choices were sought via websites of NGO's and retailers. Additionally, recommendations from the researcher's network and from potential interviewees were chosen for expert selection (Bogner et al., 2009). The area of expertise was the determining factor for eligibility, but it was also intended to recruit experts internationally. Experts were invited until a point of data saturation, i.e., the point at which no new information or themes are observed in the data, was reached (Guest et al., 2006).
Email and LinkedIn messages were used to send experts an invitation to participate in a one-on-one 60-min screen-recorded interview on Teams. The invitation explained the main aim of the study and that they were approached as experts in a specific area of expertise to provide their perspective on different ways in which an online supermarket can support consumer decision making. Potential experts were contacted a second time as a reminder if no response was received within two weeks. To thank expert participants for their time and effort, they were offered a gift card worth €20.

Procedure
A protocol for semi-structured interviews that lasted about 60 (expert interviews) or 90 (focus groups) minutes was used to facilitate the interviews (see Table 1 for the protocol). The full interview guideline is included in Appendix A. To ensure that questions were understandable and correctly perceived, five individuals tested focus group questions (1 on-site and 4 online) and two individuals piloted the expert interview protocol (online). Pilot study data was discarded and not used for further analysis. Informed (oral) consent for being screen recorded and using answers for further analysis and potential publication was asked at the start of the expert interviews and focus groups.
Following the research protocol (Table 1), consumer participants were given a pre-study task to complete at home. The interviews started with a brief introduction on the topic. The interviewer described a supermarket which helps consumers to make food choices by implementing some changes in the online store after it has generated (personal) data. To familiarize with the topic, a word association task and a plenary discussion of why participants thought of these words/ phrases followed. The next task was a card sorting task. The cards described a total of 17 different ways in which a supermarket can support consumers in the online store. The different types of cards are more thoroughly explained in section 2.3 and are shown in Appendix B. The 17 cards were divided over five dimensions, namely (1) positioning, (2) moment of exposure, (3) optimization criteria, (4) required data, and (5) advanced support. After explaining the dimension, participants were asked to rank cards from best to worst according to their own choice criteria. Structured 'why' questions were used to implicitly retrieve the perceived opportunities and barriers associated with the different types of support, either plenary (focus group) or individually (expert interviews). This process was repeated for each of the five dimensions resulting in five rounds. As a follow-up, consumer participants were asked to provide perceived key benefits and concerns of decision support, while experts had to pick the most and least successful to implement and challenging to develop support systems from the full set of cards. Then, all participants were asked to describe or sketch an ideal FSSS for sustainable and/ or healthy food choices, containing elements from previously shown cards or self-invented elements. The interviews ended with the opportunity to provide additional comments and consumer participants were also asked to answer questions regarding demographics and background variables.

Stimuli
Visuals of different ways in which a supermarket can support consumers in decision making were presented as cards (A6 format, see Appendix B). Due to the amount of information on the cards used in this study and the limited capacity of participants to sort cards (typically 20 to 30 cards, according to Rugg & McGeorge, 2005), the card sorting task was restricted to 17 cards divided into five dimensions. Each of the cards presented text and/or an image describing the layout of an online grocery store, each with a different way of helping consumers in decision-making. Products were collected from www.ah.nl and www. colruyt.be, and their corresponding Eco-Scores and Nutri-Scores were retrieved from www.openfoodfacts.org and www.colruyt.be. The hypothetical initial product choice was Liga Evergreen Forest Fruit, meaning that the hypothetical expected user preference was a forest fruit flavored cookie. An example is shown in Fig. 1.
We selected a range of extensively researched and implemented support systems, as well as some more futuristic elements to support consumers in decision making at the supermarket. Inspiration was drawn from large online platforms such as Zalando (www.zalando.nl) and Bol.com (www.bol.com). Personalized product suggestions are commonly offered at the product or category page, or during checkout. Alternate products can be presented below the initial choice, above the fold (Nadeem & Mahfooz, 2019) or in a pop-up message (Diao & Sundar, 2004). To personalize the support, e-commerce companies can track and collect a lot of data. Data can be collected explicitly (provided intentionally) or implicitly (gathered by tracking), both bringing about privacy concerns and trust issues (Zhang et al., 2014). Futuristic card features such as interactivity or multi-objective product suggestions were opted in previous literature (e.g., He et al., 2016;Schäfer et al., 2017;Trattner & Elsweiler, 2017) or were the result from brainstorming between the authors.

Data analysis
MS Teams recordings were transcribed automatically and checked for accuracy using a transcription guideline for consistency (Humble, 2015). To ensure anonymity, participants' names were replaced with non-informative numbers. Interviews conducted in Dutch were transcribed in Dutch but coded using the English codebook. Transcripts were analyzed using ATLAS.ti software, with two researchers independently coding the data for increased reliability (Elliott, 2018). The two coders reached a consensus on a set of 81 codes, which were then rearranged into key themes. To identify themes and categories for each card set, both inductive and deductive content analysis approaches were used (Burnard et al., 2008;Hsieh & Shannon, 2005). While some codes and themes were predefined based on previous research (a priori), most of them emerged from the data (Elliott, 2018). Data was coded on a sentence or paragraph level (DeCuir-Gunby et al., 2011). A semantic approach (Braun & Clarke, 2006) was used and quotations selected that represented the views of both experts and consumers were selected.

Results
After characterization of both the consumer participants and expert sample, the more generic consumer and expert perceptions of digital support strategies in the online supermarket are sequentially reported. Next, results are organized around the card sets (i.e., positioningmoment of exposureoptimization criteriadata neededadvanced support) including a representation of different themes that emerged from analysis of the card ranking. A distinction is made between consumer participants and types of experts.

Consumer participants
A total of 19 persons agreed to participate in a focus group interview  Note. *Answer options that were not chosen are excluded from the table: 'nonbinary/third gender, other …, and prefer not to say' for gender, 'elementary school or low secondary education' for education, and 'other … ' for household composition.
(n = 4 groups) in June 2022. Focus groups ranged in size from four to six participants. As per the pre-requisites of this study, on-site consumer participants lived in the Netherlands, either in the Didam (n = 5) or (the surrounding area of) Wageningen (n = 10). Online consumer participants were from Spain (n = 1), Germany (n = 1), Australia (n = 1), and Italy (n = 1). Consumer participants from all age and household composition categories took part in the focus groups (see Table 2). Slightly more females compared to males participated (12/7 ratio), and all had at least high school (n = 5) or secondary education (n = 14). Of all the consumer respondents, 10 were solely responsible for grocery shopping in their household, while 9 had a shared responsibility. All reported having at least 3 years of experience with online shopping (starting year ranged from 2000 to 2019). Almost all had an interest in health (all individual scores above 5, M = 6.03), while only around three third had an interest in sustainability (14 out of 19 individual scores above 5, M = 5.22).

Experts
In total, 56 experts were invited to participate in this study, of which 20 experts agreed to be interviewed in the period between May 17 and July 6, 2022. Of the 20 interviewed experts, 9 had a background in social sciences, whereas 11 were considered technical experts (for more detailed information see Appendix C). Experts had a background in behavioral sciences, persuasive technologies, marketing (nudging), decision aid research, system development, interface design, privacy issues, and/or had an interest in health and/or sustainability. Most experts had a research background (i.e., 1, 3, 7-17; Appendix C), whereas some had a practice-based perspective (i.e., 4-6; Appendix C), or combination of both (i.e., 2, 18-20; Appendix C). Experts from different parts of the world were interviewed, i.e., the Netherlands (n = 10), United States (n = 3), Germany (n = 2), Hong Kong (n = 2), Italy (n = 1), Spain (n = 1), and Belgium (n = 1).

Consumer participants
Consumer participants were open to decision support in the online supermarket, because it saves time and makes shopping easier, provided that the retailer understands what consumers are looking for. Decision support can remind consumers of forgotten products or products that would go well with currently chosen products. Most consumer participants considered exploring new products and receiving support to change dietary behavior as the most important benefits of decision support: "They can help me make a healthier choice. You will be motivated to look at and possibly try products that you do not know." Almost all participants emphasized the importance of providing product suggestions to consumers. Options mentioned included guidance in finding healthier, more sustainable, cheaper, new ('category new'), complementary ('tastes good with … '), or related products ('others also bought … '). A few indicated they wanted suggestions for a complete meal ('try this dish'). Participants also described the importance of decision tools such as filtering (on ingredients or allergies), sorting, tagging products, consumer reviews, and a default order list with previously bought items ('this is what you usually buy'). Clear and transparent product information, discount promotions, and personalization were mentioned as strategies to make suggestions more attractive. A decision support system should know what a consumer is looking for, as illustrated by the following quote: "I think that an intelligent system, a smart system, must in a certain way know what our preferences are, what we are searching for. (…) So, a system could tell me by proposing already the products that I'm searching for." Support being steering, interfering, and too pushy were mentioned as concerns. Some participants expected inspiration ("It can help that you get alternatives, otherwise you keep buying the same thing."), whereas others worried that such support would be condescending, diminish control over their own purchases, and complicate the decision-making process. Moreover, credibility of support was questioned as needs from supermarkets and suppliers in terms of making profit might be prioritized: "That the supermarket is helping me buy what's best for them, rather than me". Furthermore, worries existed about privacy issues ("they know everything about you") and required effort, both in terms of providing data and using decision support: "I do not want to put in additional effort to see what else is available". Nevertheless, (the effort of) providing data was also perceived necessary to get useful decision support. One participant described it as follows: "I kind of hate the idea of organizations like supermarkets knowing like personal data. But by the same token, it's also annoying getting recommended stuff that you have no interest in buying".

Experts
Experts noted that decision support can be very useful for consumers since the decision-making process becomes less difficult and it is easier to process information. To support healthy and sustainable choices, experts stated that retailers could limit the proportion of unhealthy and unsustainable products in the assortment (choice editing), although this might lead to a loss of customers. Also, giving healthy or sustainable products the most prominent spots on the page, making those options the default, or offering alternative products (product swaps) were opted strategies. One participant stated: "Definitely provide visible alternatives and substitution products for consumers online. To find a way in which they can be encouraged, tempted, and supported to choose something more sustainable, to explore different options." (Expert 20).
Personalization was seen as essential to avoid unwanted suggestions and, as such, it was deemed necessary to experimentally investigate how consumers respond to this. Moreover, experts emphasized that consumers should have the autonomy to decide what type of support they want to receive (for example, show healthier options, but not more sustainable options) and what data they want to share. Additionally, experts indicated that effectiveness of support depends on transparency about the how and why of providing support (see Table 3): it should be easy to compare initial products with alternative options, clear that consumers are being supported, clear what type of support is given, and why.
Even though providing food choice support was expected to be both effective and not too difficult in terms of implementation and development ("None of those other things [types of support] seem like they're impossible to implement.", Expert 7), experts raised some concerns related to privacy, support being annoying or requiring too much effort, lack of consumer knowledge, and required data (see Table 3 for expert views on successfulness and challenges). When personalizing support, consumer privacy issues arise ("What else is going to happen to that data once the supermarket has it?", Expert 7) and experts wondered whether consumers would be willing to go through the effort of providing data explicitly. Consumers usually make quick decisions, and it can be risky to disturb them too much (e.g., with a pop-up message) or demand too much involvement and effort from consumers. Additionally, the general feeling among experts was that consumers lack essential knowledge to fully understand product labelling (e.g., Nutri-Score), which is intensified by the lack of harmonized labels for healthiness and sustainability. A final important issue was the dynamic nature of data, meaning that data behind decision support might not be always up to date.

Positioning
where to present decision support?

Card sort results
To rank the cards in terms of positioning, social experts considered visibility of support very important, whereas technical experts and consumer participants focused on annoyance in terms of disrupting the shopping experience. Consumers also considered the effort to look for or to get rid of guidance. As such, Fig. 2 shows a low ranking for the disruptive pop-up message and a higher ranking for less disturbing, easily visible suggestions presented below or above the original product. Participants also mentioned that suggestions could be presented on the right side of the initial product and corresponding information. The best way of positioning might depend on the retailer's goal, as illustrated by this quote: "This really depends on the intention of the website, whether they want to make customers happy or they want to just force them for alternative products." (Expert 8).

Key themes in the perceptions of expert and consumer participants
Effectiveness to change choices -In reflecting on the positioning of suggestions, effectiveness was linked to presenting decision support in a clearly visible way: "Purchase behavior (….) is not a conscious process anymore and the more you are able to trigger the attention of the consumer (…), the more impact you probably have." (Expert 1). Nonetheless, most participants expected that presenting guidance in a too visible way would annoy consumers because it disrupts the shopping flow and/or requires effort. A pop-up message was expected to be most persuasive, but also most disturbing. This trade-off was illustrated by one expert: "I might go actually with the pop up at the first place, because it has more of a push element to it … But that might also be considered annoying by consumers, I'm going to put that on the third place." (Expert 20). Both experts and consumer participants realized that there is an optimal frequency to display suggestions, particularly related to healthiness and sustainability of products. Too little reduces effectiveness and too many can lead to consumer overload and indecision.
Preventing confusion -Both expert and consumer participants referred to the shopping flow when discussing the positioning of suggestions. Ideally, the consumer first sees the initial product and then receives support. Presenting support above the initially picked product was perceived as confusing by various participants: "If you read quickly, you will probably read it incorrectly and there is a much greater risk that you click on the wrong element." (Consumer participant).

Card sort results
For moment of exposure, experts and consumer participants mainly based their ranking on the support's potential impact on the decisionmaking process, perceived level of freedom, and transparency of suggestions. As a result, displaying support at the checkout page was rated relatively low, whereas support at the product page and personalizing each category page were almost equally preferred, showing highest ranking for the product page (Fig. 3).

Key themes in the perceptions of expert and consumer participants
Effectiveness to change choices -Participants considered support at the product page as effective, as it allows consumers to easily compare suggested alternatives to the initial choice: "If you wanted to make a change, you easily make it here" (Expert 13). Moreover, presenting support immediately (e.g., personalized category page card) was considered useful by a substantial part of participants because it presents an overview of the product options, allows for product discovery, and influences the decision making of consumers early. Yet, participants complained that immediate help with preference-based product suggestions might be steering and restrict consumers in exploring other options ("The metaphor is that if I walk into a clothing store and they come to you, 'can I help you?' No, I want to see for myself first", Consumer participant). Also, multiple experts criticized the lack of transparency that suggestions were being given ("There's not really much space for understanding the fact that you're being advised and provided a substitute.", Expert 3).
Almost all participants considered suggestions at the checkout page too late to effectively change choices, because it was expected to be too unlikely that people would change their choices once they have made up their mind. Moreover, consumer participants perceived it as annoying to be offered good alternatives after making product choices. Still, some social experts considered the checkout page a good moment to provide feedback in terms of a total healthiness or sustainability score, or to suggest alternatives for the most unhealthy and unsustainable product choices, as illustrated with this quote: "I think that there is a moment of reflection and if there are really good product suggestions, then you might want to switch. Especially if you [as retailer] would provide the total [healthiness] score of the basket." (Expert 5).
Combine moments of exposure -Several experts mentioned that all options ( Fig. 3: product page, personalized category page, and checkout page) are ways to change behavior and could be combined. For instance, it was opted that a healthier or more sustainable version could be shown on the product page, and then the most unhealthy or unsustainable products in the basket could be identified for which alternative products could be suggested. While experts seemed to be enthusiastic about combining different moments of exposure, consumer participants were more hesitant. On the one hand, it was perceived helpful to be offered alternatives or to be reminded about potentially forgotten products, or even be inspired to buy better food options. On the other hand, consumer participants feared they might feel overloaded with all the guidance and that it would lengthen the grocery shopping process. One participant for instance stated: "You keep being distracted from your own choice when you are directed to a different choice each time. You will be in doubt all the time. I find that quite annoying."

Card sort results
Social experts considered sustainability goals important, whereas technical experts and consumers prioritized health goals (Fig. 4). The importance of retailers putting forward concerns for health and sustainability goals was emphasized: "I think that it's also a very soft push measure on the side of retailers (…) it normalizes the fact that you would expect the consumer to want to make more sustainable choices." (Expert 20).

Fig. 2.
Ranking of positioning of the decision support: Number of times each card (above the product, below the product, pop-up message) is ranked #1, #2, or #3 by either consumers (n = 19), social experts (n = 9), or technical experts (n = 11). Note. Simplified versions of the cards are shown in this figure. See Appendix B for the stimuli used during interviews and focus groups.

Key themes in the perceptions of expert and consumer participants
Effectiveness to change choices -Participants pointed out that healthy eating is often a goal for consumers, making them more open to suggestions based on health than sustainability. On the contrary, sustainability aspects might be less known to consumers and suggestions can increase awareness in this area: "I do not know how a cookie is Fig. 4. Ranking of optimization criteria: Number of times each card (healthy, preferences, sustainable, and related) is ranked #1, #2, #3, or #4 by either consumers (n = 19), social experts (n = 9), or technical experts (n = 11). Note that vertically the amount for technical experts does not add up to n = 11 (#4) or exceeds n = 11 (#1 and #2) due to some experts ranking two cards equally (healthy and sustainable as equal, either both as #1 or #2). Note. Simplified version of the card for healthy choices is shown in this figure. The other three cards had the same layout, but a different goal (preferences, sustainable, related). See Appendix B for the stimuli used during interviews and focus groups. produced and what it contains, how much CO2, water it all costs. So, I would appreciate seeing 'That cookie is more sustainable', and then I might think 'Oh well, maybe I will choose that one." (Consumer participant). Consumer participants felt that they do not need support finding preference-based items, whereas experts stated that suggestions should (also) be preference-based to match consumer needs and wishes. Related products from the same product category were not considered helpful by participants ("That's sort of the 'hey, I already bought battery, saying now you're gonna recommend me 92 other kinds of batteries.", Expert 16), but complementary products might be. For all suggestions, regardless of their goal, it was deemed important they fall into the product category of the initial product (e.g., substitute cookie with cookie).
Transparency -The main challenge of suggesting healthier or more sustainable options identified by participants was the absence of a clear, uniform definition of what constitutes a healthy or sustainable food product. A few experts linked this to the lack of a definition from EU policy makers: "They have a lot of trouble there to really straighten that out and to make sure everyone agrees." (Expert 19). Moreover, the standard way of providing healthy or sustainable options might not work for everyone because what is healthy and sustainable might differ across people: "People buy into different definitions of sustainability, or they are willing to make different tradeoffs around it." (Expert 16). Thus, many stated that justifying decision support is needed, for instance by providing more details on the criteria used to justify why a suggestion is healthier or more sustainable: "Explain why something is healthier or more sustainable (…), maybe in terms of CO2, water usage in numbers or something, those sorts of things." (Consumer participant). Support only for engaged consumers -Sustainable and healthy suggestions were considered helpful in educating and changing choices of the unaware ("Because some people do not know", Expert 3) and indifferent consumers ("People who just don't think about it that much and they see 'Oh, this is somewhat healthier', they might think 'Oh yeah, I can also choose that.", Consumer participant). Nevertheless, a substantial part of (mainly social) experts worried that suggestions would help only consumers already committed to healthy or sustainable eating. This worry was also emphasized by one consumer participant: "I think that everyone says that they want healthy or sustainable options, but I'm not sure that people actually shop that way in real life". Another concern among participants was consumers' inability to buy more sustainable or healthier options due to for instance, dietary or budgetary restrictions. Two consumer participants expressed their frustrations, as illustrated by this quote: "It's the moral thing that frustrates me a lot because I would like to buy it, but (…) I don't have enough money to be a more sustainable person." In terms of solutions, experts believed that consumers should be incentivized to choose the healthy or sustainable option, especially to overcome initial resistance: "I think because there are still a lot of customers who are not that intrinsically motivated that (…) there must be some kind of short-term reward that makes them like, 'Hey, I'm going to make that choice anyway'. And once they get used to it, (…) it has become their buying behavior." (Expert 4). Various experts stressed the importance of understanding the drivers of sustainable and healthy food choices to adjust incentives accordingly. As an option, various experts stated to minimize money restrictions with price discounts ("Or you say that A and B products [based on Eco-or Nutri-Score] are discounted by 10% this week.", Expert 5).
Combine or indicate relevance of goals -Linking one's preferences to a health or sustainability goal in one product suggestion, or framing decision support based on one's goals (e.g., 'this one is healthier' for health interested people) was perceived promising to increase likeability of suggestions: "Because that is crystal clear: the more personalized, the greater the chance that it will be chosen." (Expert 2). Personalization seemed to be most important when offering healthier alternatives since what is healthy for one person, might not be healthy for another. For sustainability, two experts advised to combine preference and sustainability goals but not mention that it is more sustainable, because consumers often do not care about sustainability, nor grasp the idea of its meaning: "The consumer does not think about sustainability at all (…) they do not really understand it either" (Expert 2).
For consumer participants, it was not so much about combining different goals in one suggestion, but rather the possibility to indicate what type of decision support would be helpful for them, for example in a user profile ("If the retailer offers alternatives, you have the option to indicate what you want to see there.", Consumer participant). Various experts shared this view that autonomy is important: "You can provide some controllability, you know, in terms of what gets recommended for what reasons." (Expert 7). Related to this, some experts opted that goal selection would not be needed if multiple rows of suggestions would be presented ("I think they have to make sure that all those options are available at once.", Expert 13), leaving room for consumers to choose themselves what they are interested in, and providing the possibility to compare various options.

Card sort results
In terms of data requirements, there is often a trade-off between the level of personalization versus consumer autonomy and privacy. While experts mostly focused on the usefulness of collected data, consumers were more concerned about their privacy. Consequently, social experts favored tracking required data, whereas non-personal decision support was mostly chosen by technical experts and consumer participants (Fig. 5).

Key themes in the perceptions of expert and consumer participants
Effectiveness to change choices -While social experts considered tracked search and choice behavior the most useful data, followed by consumers explicitly indicating what they want the retailer to know ("At least I know that I consider that data relevant to be taken into consideration.", Expert 5), technical experts believed that even in the non-personal situation, retailers have some information from purchase behavior to personalize support. In addition, experts thought that stimulating healthy and sustainable food choices can be done without personal information from the consumer. The need for this was stressed by one expert: "We have to make sure that everyone eats somewhat more sustainable, so you don't really need any personal information for that." (Expert 18).
Effectiveness might be hampered by complexities related to data collection. First, experts emphasized it is important, but difficult, to have a large enough product database to be able to provide support (e.g., more sustainable as well as healthier version). Second, various experts worried that retailers might be too dependent on consumer data to provide decision support, which was perceived especially risky if consumers should explicitly provide data. Due to the perceived premise among experts that more data leads to better decision support, it was considered complex to provide accurate support to both new ("But for them, it is usually low hit", Expert 12) and regular customers ("I think supermarkets know still way too little information about the customer to know about my preferences.", Expert 10). Assuming that consumer data is available, experts still expected challenges to integrate data optimally and provide highly accurate and satisfying product suggestions: "If I tell I have this allergy (…), I think the tolerance for false positives is going to be really low, right?" (Expert 7). This inaccuracy of support was also a concern among consumer participants, as illustrated by this quote: "You bought something once and then it is at the top of page all year (…), while it was just a one-time buy because my parents came over for dinner." Autonomy versus effort -Expert's preferred way of collecting data depended a lot on the balance between consumer autonomy and effort: "There are things that you can infer from purchase behavior (…) it's a question of would I rather have you figure this out, or would I rather tell you." (Expert 7). According to experts, explicitly sharing information is preferred over tracking information, because consumers are aware of what they share and can opt-out. Nonetheless, consumer participants mainly critiqued effort of explicitly providing personal data if asked every shopping trip ("I do not want to answer the same question over and over again.") but saving information in a profile was perceived as an acceptable way to collect data ("And at the profile everything is adjustable, adaptable, and regularly updated."), or by some even considered better than tracking data.
Willingness to share information -Consumers were expected to experience a trade-off between the usefulness of decision support and protecting their privacy when asked for their data, as illustrated by one expert's quote: "You want to save time by your computer already showing you preselected stuff based on your preferences or you want to spend a little bit more time but protect or have the illusion that you're protecting your privacy a little bit." (Expert 15). Privacy concerns were mainly related to what data would be collected, what would happen to the data, and whether third parties would have access. While experts favored consciously sharing information over tracking ("It gives a sense of ownership to the consumer, and they feel that they are willingly sharing information.", Expert 20), only a subset of consumer participants agreed. Most consumer participants favored non-personal data collection for reasons related to privacy (and low required effort), but some also considered tracking purchase data acceptable to personalize offers.
While most consumer participants were not in favor of explicitly providing information (mainly due to required effort), many would be willing to explicitly share data that is considered crucial for providing food choice support. This led to relatively high willingness to share information on allergies and diet preferences, whereas opinions on providing information on age and gender were more divided, and participants felt reluctant to share information on personal details such as income, weight, and height (Appendix C). Experts indicated to see most relevance in asking for allergies and diet preferences: "And they [retailer] can just sort of filter out those products and you [consumer] only see the things that you can eat. I think that could be really valuable for people. (Expert 7). One expert emphasized that willingness to share information might depend on the efforts from the retailer to convince.

Card sort results
Experts rated the more advanced options for providing support from the perspective of a consumer who needs to be able to quickly understand the suggestion and be motivated to respond. As can be seen in Fig. 6, visually attractive labels therefore scored relatively high and putting the initiative with the consumer (in 'ask another' and 'ask support') relatively low. Consumer participants somewhat liked autonomy (in 'ask support'), but it was deemed more important that decision support was easily understood and explained. This led to the relatively high ranking for labelling.

Key themes in the perceptions of expert and consumer participants
Autonomy versus pro-active system -A large part of experts and consumer participants felt that putting the initiative with consumers in terms of them asking for decision support is user friendly, but the likelihood of consumers making use of this option was expected to be low due to the high required effort ("It's an extra step in the process of doing something. (…) they [consumers] would be slowed down and their mental process would be disrupted, which usually has a negative impact.", Expert 8). Ideally, a pro-active system would include some element of autonomy. For example, consumers could have the option to switch support (receive another suggestion, Fig. 6: ask another), as mentioned by consumer participants ("I like 'give me another option', then you can make a kind of game out of it.") and experts ("The first choice is not necessarily an alternative for them, so it's nice to have the option to indicate, 'I do want a healthy choice, but not specifically this one, maybe there is something else.'", Expert 5).
(ways to) Justify product suggestions -Experts and consumer participants considered providing a reason why a specific product is recommended very useful in terms of transparency and educating consumers to enable informed decision-making: "There might be an increasing slice of consumers that says 'OK, you're telling me this product recommended is healthier or more sustainable, says who? Which are the parameters and so on?'" (Expert 20). Providing information on various Ranking of the required data: Number of times each card (non-personal, tracking, and asking) is ranked #1, #2, or #3 by either consumers (n = 19), social experts (n = 9), or technical experts (n = 11). Note that vertically the amount for technical experts does not add up to n = 11 (#3) or exceeds n = 11 (#1) due to some experts ranking two cards equally (non-personal and tracking or asking and tracking as #1).
criteria (e.g., health and sustainability level of products) allows consumers to compare options and choose themselves: "On the one hand, it might show that sustainable is not automatically healthy and vice versa. On the other hand, it is also possible that they match and confirm the choice." (Expert 19).
Overall, labels were expected to be more salient, convincing, and interpretive for consumers ("The signal colors, I think that works well.", Expert 18) than providing a textual explanation. Additionally, the thirdparty verification of a label was expected to increase trustworthiness of decision support: "Then you also give people the agency to make that decision instead of saying 'you should pick this one because it's healthier, because I say so.'" (Expert 17). Nevertheless, experts and some consumer participants doubted whether consumers would understand the meaning behind a label, and therefore preferred a factual, textual justification (e.g., 'lower in sugar' instead of 'healthier'). Furthermore, some social experts indicated that consumers might feel overloaded by the many different labels out there ("If there's a lot of them, transparency is not ensured.", Expert 20), which was perceived especially worrisome for sustainability because there is not one harmonized and regulated sustainability label at the EU level.
More advanced decision support methods -Various participants opted that providing a justification for product suggestions could be done in a friendly and playful way. Different options were mentioned, of which a button that re-directs to an information page was most often brought up. This information page can then provide an explanation for why a particular product is being suggested ("Possibly an option that a consumer can click on and go and find out the justification of why the supermarket proposed that specific product or alternative as more sustainable.",Expert 20) or what the label means ("Maybe have an information page where consumers can click to if they see a label to get information what that label means.", Consumer participant). In doing this, retailers allow consumers to increase their awareness and knowledge and to make more informed choices during their next shopping trip.
Moreover, experts cited various other types of advanced decision support that could be useful in convincing consumers to accept support from the retailer. Examples of such strategies were the use of norms (show alternatives that peers or role models have chosen), information banners for informed decision making ("These are personalized results for you", Expert 7), enhancing feelings of commitment at moment of choice by goal setting ("How much of your basket do you want to be sustainable?", Expert 1), and gamifying. A gamified ecosystem was expected to encourage purchases of sustainable and healthier products because of the reward system (e.g., save points for free products) and the possibility to compare results with other people. This reward system was also expected to be important for commitment: "Because the effect is stronger when you do that publicly, so suppose you can show it to others or your shopping cart looks different or your profile picture has an edge, (…) that you make that commitment visible in some way and maybe even to others." (Expert 1).

Discussion
Decision support and personalization techniques are widely used in online shopping with the primary aim of boosting sales and profits based on consumer preferences. Yet, the potential of these techniques to encourage consumers to buy healthier and more sustainably is still largely unexplored in research and practice. Our study shows that both experts and consumer participants value and even expect decision support in online supermarkets, also to encourage healthier and more sustainable choices. However, effectiveness herein depends on the delicate balance between guiding certain choices by adapting parts of the assortment, offering inspiration and suggestions, while at the same time avoiding support being perceived as intrusive and thus interfering with a pleasant shopping experience. Positive perceptions of FSSS were linked to visible support presented early in the shopping trip, as well as transparency in informing consumers that they are being advised. This is consistent with findings that attribute the success of interventions to the fact that interventions should be simple and tangible (White et al., Fig. 6. Ranking of the advanced methods: Number of times each card (ask another, labels, text, and ask support) is ranked #1, #2, #3, or #4 by either consumers (n = 19), social experts (n = 9), or technical experts (n = 11). Note. Simplified versions of the cards are shown in this figure. See Appendix B for the stimuli used during interviews and focus groups. 2019), should specify a desired behavior (Kalnikaite et al., 2011), and intervene at product selection (Forwood et al., 2015;Nahum-Shani et al., 2015).
The findings demonstrate that personalized healthy and sustainable recommendations are highly valued as they can be tailored to individual differences in nutritional needs, preferences, and consumer motivation to receive support when making choices that are either healthier, more sustainable or both. This is in line with previous research on targeted marketing and support (e.g., Grbovic et al., 2015). Even though previous research already emphasized the increasing consumer expectation of personalization in food environments (Lyzwinski et al., 2018;Stewart-Knox et al., 2013) and the need for matching sustainable and healthy food offers to people's self-interest and ego-centric motives (Griskevicius et al., 2012;Verain et al., 2022), personalization in online shopping is still limited in the field of sustainable and healthy choices, providing great opportunities for future experimental research and practitioners.
Even though personalization is increasingly expected by consumers, it does not automatically result in a willingness to share personal data due to the required effort and concerns about privacy and commercial interests. As a result, consumers favored online retailers to not collect personal data, but did show a willingness to provide personal information when considered useful for food choice support, such as information on diet preferences and allergies. Experts stressed the usefulness of available previous purchase data in providing support, but worried about not having up to date consumer and product information to provide accurate support. Future work may consider using various levels of personalization in offering healthier or more sustainable food options and testing the effect on food choices and consumer satisfaction.
Relatedly, some concerns were raised about consumer's use of FSSS, especially because healthier or more sustainable suggestions may only be effective for engaged consumers who are already committed to eating healthy and sustainable. Experts indicated the relevance of incentives such as providing discounts, norm messages, and gamifying in stimulating better food choices. Another option mentioned was to provide personalized sustainable food suggestions but refrain from indicating that it is a more sustainable option. It was expected that the absence of sustainability framing would make certain consumers more likely to buy sustainable choices, as previous research already indicated (e.g., Demartini et al., 2022;Krpan & Houtsma, 2020). Nonetheless, this way consumers are not educated (made more aware) about sustainability of food choices, which does not strike with participant's perceived benefit of educating consumers using advanced decision support such as labelling or informative text.
Both experts and consumers place great importance on transparent and substantiated decision support using labels or informative text to enable understanding of why a particular product is recommended. However, the criteria for considering a product healthier or more sustainable differ between labeling systems and healthier products are not always sustainable and vice versa (Tilman & Clark, 2014). Understandable explanations about health and sustainability in the decision support process can increase confidence and provide consumers with knowledge about these topics, which they often lack (Breathnach et al., 2021;Camilleri et al., 2019;Dickson-Spillmann et al., 2011;Hartmann et al., 2021). Also, transparency in why advice is given can reduce mistrust in the retailer's improper motives. The relevance of explaining to users the reasoning behind product suggestions has been emphasized in previous research (e.g., Cramer et al., 2008;Felfernig et al., 2007;Nunes & Jannach, 2017). Providing information on the product page seems to be more effective than providing information through external tools (e.g., website links) due to use and usefulness of immediately visible information (Werle et al., 2022). It should be noted that over-reliance on increased education and information provision may enlarge perceived information overload and can be counterproductive (Hoffmann et al., 2022).
The feeling among both experts and consumer participants was that autonomy to choose type of help (e.g., sustainable but not healthy suggestions) and the ability to change suggestions (switch to another product) as well as autonomy to provide data such as through an optional user profile would be appreciated. Nonetheless, negative perceptions of experts and consumer participants on FSSS appeared to be linked to required effort in looking for guidance (e.g., additional clicks) and in terms of data provision (e.g., repeatedly providing data). Experts feared a potential intention-behavior gap, indicating that a positive consumer attitude towards autonomy is not made actionable. Also, limited consumer interest in support might lead to low use. Ideally, one would give users autonomy in their decision-making while using support, since freedom of choice and perceived responsibility of choice have previously been identified as determinants of satisfaction (Botti & McGill, 2006). This assessment of the opportunities and barriers of sustainable healthy FSSS should be interpreted with some limitations in mind. The explorative qualitative nature of this study and the limited sample size, as well as the narrow geographical focus limit generalizability of findings. Moreover, data collection taking place both offline and online might have led to differences in how people express themselves and a lack of spontaneity in online settings (Stewart & Williams, 2005). However, the sample included a diverse group of experts and consumer participants, enabling us to identify a wide range of relevant factors. Also, results could be applicable in other contexts such as meal order websites and the mobile checkout counter in brick-and-mortar stores.
Another limitation is that our card sorting task as a method to elicit responses may have biased results. Despite retailers being able to assist consumers in various ways, such as by filtering, sorting, and offering suggestions (Häubl & Trifts, 2000;Lurie & Wen, 2014), in the current study we had to limit the number of cards used. Our focus on providing suggestions may have limited the generalizability of our findings to other types of decision support, such as filtering. Also, our examples could have limited participants' creativity, although results showed that it was a valuable starting point for discussion and helped participants verbalize their responses.
Experts and consumer participants sometimes differed in their view of optimal support. For example, experts thought that support should be offered as early as possible in the decision-making process, while consumer participants indicated that they would rather receive this support a little later in the process. Future research and applications would greatly benefit from experimental research into actual effects of subtle changes in the choice environment of the online supermarket. Moreover, qualitative research does show consumer preferences and concerns and provides insight into possibilities in terms of data and design, but this does not necessarily translate into different purchasing behavior. Future work could investigate the effect of the level of personalization in offering healthier and sustainable options on food choices and customer satisfaction among a large sample of respondents. This could involve distinguishing participants based on their health and sustainability interest and use these consumer profiles for a more targeted approach in the FSSS. The impact of the FSSS design in terms of each of the five dimensions (positioning, moment of exposure, optimization criteria, required data, and advanced support) can be examined in further quantitative research. For example, the positioning (e.g., present support above or below product), but also the types of explanations or justification for the product suggestions by using for example labelling or textual information. Moreover, it would be interesting to examine the willingness of consumers to disclose personal data for support that is focused on healthiness and sustainability and the types of data they would be willing to share for this purpose.
Preferably, future quantitative work is not limited to framed market experiments (Caputo & Just, 2022) but includes natural field experiments carried out in an actual online supermarket without explicitly informing consumers about the interventions to ensure more reliable responses. By conducting field experiments, researchers could also focus on testing the effectiveness of other types of support such as filtering. Moreover, exploring the potential of decision support systems in other industries could be a valuable avenue for future research. Investigating the use of support systems can help researchers in identifying best practices, challenges, and opportunities that may also be transferable to the context of sustainable and healthy food choices. This can broaden understanding of decision support systems and their potential application in various industries, as well as stimulate improvements in the field of support systems.
In conclusion, our research shows that online FSSSs can be seen as a promising way to optimally support consumers in making healthier and more sustainable food choices. Unlike traditional shopping, online shopping offers the opportunity to steer the consumer early in the process based on previously collected data or behavior during shopping. This presents a world of opportunities, as both consumer and expert participants emphasize the importance of providing support that is desired by the consumer, effortless, and not overwhelming or patronizing. Consumer participants indicate that they want to keep control over support (for example, receiving sustainable suggestions, but not healthier ones) and their data, with the latter being important due to privacy concerns. Personal support for sustainable and healthy food choices would therefore be an effective means of intervention for relatively motivated consumers, although research into the precise implementation of individualized support is limited. Thus, identifying the best ways to provide continuous and optimal personalized support is a promising area for experimental research.

Author contributions
All authors have contributed to the conception of the study. LJ has collected the data. Together with an independent coder, LJ coded the data, discussed the outcomes, and decided on a final coding structure and themes. LJ, EVL, EvK have analyzed and interpreted the coded data. LJ has written the first version of the manuscript and rewritten based on feedback from all co-authors. All authors read and approved the final manuscript.

Ethics approval and consent to participate
Ethical clearance was provided by the institutional Social Sciences Ethics Committee (SEC) of Wageningen University, the Netherlands. All participants provided informed consent prior to taking part in the study.

Consent for publication
Consent from the participants was asked at the beginning of the interviews. We informed the participants that they would get a noninformative participant number and that their name would not appear anywhere, so anonymity would be guaranteed. We asked the participants to agree with the processing of data before the start of the interview.

Availability of data and materials
The data that support the findings of this paper are available upon reasonable request from the corresponding author, LJ.

Funding
This work was funded by the Section Business Science of Wageningen University & Research through the Business Science for Sustainability project, and the Dutch 4TU Federation through the Pride and Prejudice project. The funding providers had no further role in the design of the study, the collection, analysis, and interpretation of data, or in writing the manuscript.

Ethical statement
Ethical clearance was provided by the institutional Social Sciences Ethics Committee (SEC) of Wageningen University, the Netherlands. All participants provided informed consent prior to taking part in the study.

Declaration of competing interest
None.

Data availability
Data will be made available on request.