Blockchain-Based Traceability System From the Users’ Perspective: A Case Study of Thai Coffee Supply Chain

We explored the use of blockchain technology for traceability to improve the safety and value of food, focusing on the coffee supply chain as a case study. The main goal was to evaluate the feasibility in terms of design, perceived benefits, and challenges of applying blockchain and traceability from the users’ perspective. We implemented a prototype using a user-centered iterative interface design. Then we used the prototype to answer our research questions in mixed-method research, including in-depth interviews (10 participants) and a survey (350 participants) with stakeholders in the coffee supply chain in Thailand. The results showed that timeline-based design was preferred over map-based or text-based design for the visualization of traceability information and that blockchain was a promising technology, as 67% of the survey participants saw a positive influence of blockchain on the adoption of applications. The most notable benefits were origin checking and increasing product trustworthiness. The most notable challenges were inaccurate or incomplete information and the disclosure of trade secrets. More work is required to address the challenges for everyone in the supply chain ecosystem to adopt the proposed traceability system, including (1) providing trustworthiness and completeness of information by cross-checking with third parties or other users, (2) protecting sensitive information by aligning users’ interests or allowing control of information disclosure, and (3) educating and giving producers the motivation for the difficulty and the extra work.

Blockchain is an innovative combination of computer tech-94 nology to create a distributed database management system. 95 It is designed for the transparency and integrity of the infor-96 mation, allowing multiple parties to transact without a cen-97 tral authority. Blockchain started in 2008 as the technology 98 behind the first successful cryptocurrency, bitcoin [14]. The 99 main purpose of bitcoin was to be a peer-to-peer cryptocur-100 rency transfer application. The blockchain technology behind 101 bitcoin stores the data in a block-based data structure, chained 102 to the past blocks to create a linked history of data. The peer-103 to-peer network of bitcoin computer nodes reaches consensus 104 using the Proof-of-Work algorithm. 105 Although bitcoin has proven to be very successful and has 106 operated for more than a decade with millions of users, the 107 original blockchain technology was not suitable for scaling 108 to other applications. The Bitcoin blockchain has many limi-109 tations, including low throughput, high energy consumption, 110 security issues, and privacy of data. Therefore, blockchain 111 technology has become an active research area, with many 112 researchers aiming to improve various technical aspects of 113 blockchain, such as scalability [15], security [16], scalable 114 integration with other technologies [17]. The technology is 115 still in its infancy and remains an active area of research to 116 date. 117 In terms of application domains, blockchain started in the 118 financial sector but later expanded to other sectors, includ-119 ing agriculture and supply chain. Several researchers have 120 worked on the topic of blockchain and traceability, such 121 as supply chain traceability systems with blockchain and 122 RFID technology in China [18], applications of blockchain 123 or distributed storage technology for agricultural commodity 124 markets [19], and others [20], [21], [22]. Trace Thai is another 125 blockchain-based traceability system for Thai produce 126 [23], [24]. However, Trace Thai was designed for generic 127 products, which may not be optimized for coffee traceability 128 in Thailand.

130
Coffee is an agricultural product with a high market value 131 (US$3,185.00m in 2022 for Thailand) and is likely to con-132 tinue to expand annually by 11.99% [25]. In 2018, Thailand 133 had 57 new companies registered for coffee production [26], 134 and in 2022, Thai people consumed an average of 0.59 L of 135 coffee per person [25].
as well as expected benefits of the machine, such as dividends  In addition to Bager et al. [40] mentioned in the previous 201 subsection, we found only one work that studied the adoption 202 of blockchain-based coffee traceability systems in business. 203 Thiruchelvam et al. [41] analyzed comments from stakehold-204 ers and proposed the Technology Acceptance Model (TAM) 205 of blockchain technology in the Burundi coffee industry. 206 In a more general context, several works studied the adop-207 tion of blockchain technology in agri-food supply chain trace-208 ability. Chen et al. [10] conducted a thematic analysis of the 209 literature to identify the processes, benefits, and challenges 210 of the adoption of blockchain technologies in food supply 211 chains. Similar works studied the adoption in the literature 212 using content analysis [11] or analyzed review articles [42]. 213 As a key factor in technology adoption is prospective users, 214 works such as Behnke and Janssen [8], Yadav et al. [43], and 215 Saurabh and Dey [44] involved users to identify the bound-216 aries or challenges to be addressed before the technology 217 can be put into practical use. However, as seen in Table 1, 218 which summarizes their method and notable findings, the 219 key factors for technology adoption could differ depending 220 on the type of product and the country in which the study 221 was conducted. The study of technology adoption in the Thai 222 coffee supply chain is lacking and therefore our focus. 223 This paper investigates the adoption of technology by 224 involving prospective users through mixed-method research. 225 Furthermore, we formulate questions in a way that made 226 participants aware of 'blockchain' since the opinion of the 227 system might lie in the digitized supply chain rather than 228 the blockchain [40]. We discuss how our findings con-229 form or vary from other works in Section VII. Additionally, 230 we include reports of the possible solutions from users' per-231 spectives, which have not been studied sufficiently globally. 232 233 We adopt mixed-method research to investigate the benefits, 234 challenges, and suggestions in adopting a blockchain-based 235 traceability system. Following an exploratory sequential 236 design [45], we first conducted semi-structured interviews 237 with coffee producers and customers to collect qualitative 238 data, which informed subsequent quantitative data collection 239 using a questionnaire survey. The interviews and survey were 240 reviewed and approved by the Institutional Review Board, 241 and informed consent was obtained at the beginning of the 242 studies. To ensure the feasibility of the procedure and the 243 questions, we tested both the interview and the survey with 244 one pilot participant each. The pilot results were not included 245 in the final result but served as guidelines to make the final 246 refinement of the questions.

247
The interview and survey questions were formulated to 248 answer the research questions: (1) what factors influence 249 the adoption of our application and (2) does blockchain 250 affect their opinions about providing the data or adopting 251 the system. For the survey, we also included (3) how infor-252 mation disclosure affects their opinions about adopting the 253 system. The last research question was used to address one 254 VOLUME 10, 2022 TABLE 1. Summary of selected related works. Note that some works did not specify where their study was conducted, so we assume the country from the authors' affiliation.
of the main concerns of the participants we found during the 255 interview.
256 Figure 1 provides an overview of our methodology. As a 257 blockchain-based traceability system is relatively new in 258 Thailand, we included a system prototype in the interviews 259 and survey to ensure participants understand the system. 260 We first briefly explain how we came up with the prototype 261 and then explained the interview and survey procedure.

263
We used a user-centered iterative design to develop our sys-264 tem prototype. We first designed the system using infor-265 mation from literature, used the design to conduct user 266 interviews, then used findings from the interviews to refine 267 the design. Each phase involved a few sub-iterations and 268 3 -5 design alternatives. We interviewed 10 participants in 269 total. There were also pilot studies before the formal stud-270 ies. The main questions in the interviews included (1) what 271 information users were interested in and (2) how to present 272 the information to users. By involving users when designing 273 the system, we reduce the effect of poor designs on the 274 willingness to adopt the system.

276
The interview was semi-structured and mostly done online 277 due to the Covid-19 situation. As it was difficult to let 278 participants try our application, we used a recorded video 279 describing the application and shared our screen when we 280 clicked through each design. The questions included partici-281 pant information and their opinions after seeing the demon-282 strations, as listed in Appendix A. 283 We used video-conferencing applications that the par-284 ticipant was familiar with. The interviews were conducted 285 in Thai. Participants could turn off their video if showing 286 their video was uncomfortable or inconvenient. Each session 287 involved one or two researchers and one or two participants. 288 Each interview took about 30 -120 minutes. We recorded the 289 conversations for later analysis and compensated participants 290 for their time.

291
After summarizing the recordings into text, we analyzed 292 the findings for the coffee production information flow and 293 visualization, then used them to improve the design. The 294 first author performed an inductive thematic analysis [46] 295 of the interview content to identify the willingness to adopt 296 our system, benefits, challenges, and suggestions. The find-297 ings from the interviews informed our questionnaire design. 298 appeared in the questionnaire as choices to confirm and inves-300 tigate the degree of the findings. Note that the suggestion 301 categories reported in this paper were later revised to match 302 with coding from the survey to ease the discussion.

304
The survey used a questionnaire with questions simi-305 lar to those of the interviews, as listed in Appendix B. 306 We distributed the questionnaire to the National Science and   Finally, we determined the reliability of our codes. 326 We adopted Krippendorff's alpha [47], which provides a 327 higher adequacy for opinion research [49]. As one answer 328 might have multiple codes, we calculated the agreement using 329 the Measuring Agreement on Set-valued Items (MASI) [50] 330 with the NLTK toolkit [51]. The codes and number of answers 331 tagged with each code that two coders agreed on are reported 332 in Section VI. 334 We designed how users can record, pass, and see transac-335 tions. The designs were informed by literature reviews [27], 336 [28], [30], documentaries [52], [53], other online resources 337 [54], [55], [56], and user interviews.

338
When users logged in to our system, they would see a menu 339 according to the processing steps: planting, picking, process-340 ing, milling, grading, roasting, and packaging. In each step,  Our design included three interesting features.

349
First, users could mix beans from multiple sources, allow-350 ing the mixing of coffee beans from multiple farmers as 351 well as the blending of multiple coffee species. Information 352 from the previous step was listed with the checkbox, so users 353 could select one or multiple sources to continue, as shown in 354 Figure 2a. 355 Second, we allowed partial usage of beans. Users could 356 specify the quantity to be used, as shown in Figure 2b, and the 357 system would automatically calculate the remaining quantity 358 to be used later. This feature provided flexibility, for example, 359 for roasters when they blend coffee beans. 360 Third, users could use either a QR code or progress to the 361 next step without a QR code, as shown in Figure 2c, so one 362 person could have multiple roles. This feature might not be 363 thought about without the interview, as other studies of the 364 coffee supply chain seemed to suggest a dedicated role for 365 each user, while the interview revealed otherwise for Thai 366 coffee producers.

367
There were also minor design features. For example, 368 we did not directly ask about the chemical used. Instead, 369 we used the term ''remark'' to allow organic farmers to 370 explain their specialty while not discouraging normal farm-371 ers. Users could optionally add a taste note after the pack-372 aging step, as all participants mentioned the taste of coffee 373 during the interview. This information should be useful for 374 supporting purchasing decisions.

375
For the visualization of traceability information, the 376 designs were mainly informed from our previous work [42]. 377 We analyzed 14 blockchain-based agri-food traceabil-378 ity applications and 10 non-blockchain-based agri-food 379 traceability applications, then classified ways to visualize 380 tractability information into 5 approaches: text, table, time-381 line, graph, and map. Since the preferred approach remains 382 an open question, we decided to explore all alternatives by 383 coming up with a design for each visualization type. Each 384 design presented similar information but featured different 385 pieces of information. In all designs, the users could click on 386 an icon after the information to see additional information, 387 such as certificates or map location, in a pop-up window.

388
Allowing mixing beans surprisingly made the visualization 389 complicated. The scalability of the design was a major chal-390 lenge, especially on mobile applications. The issue existed 391 even when we used a simple case where we blended ara-392 bica and robusta from four farmers. Thus, we eliminated the 393 table-based and graph-based design because of their poor 394 scalability and limited benefits. 395 We continued with text-based, timeline-based, and map-396 based design. We found that the map-based design was the 397 most intuitive to scale, as the visibility of places already 398 depends on the zoom level. The timeline-based was the most 399 problematic since both lines and text created visual clutter. 400 In the end, we decided to use collapsible interaction for 401 both text-based and timeline-based design, to highlight the 402 most important information (i.e., the blending of arabica and 403 robusta).

404
Text-based Design, as shown in Figure 3a and 3b, placed 405 information in collapsible panels. The information was first 406 grouped by coffee species and then by the owner of the infor-407 mation. The header of each panel included (1) the species 408 VOLUME 10, 2022    The design collapsed information of each species into one 417 item, used a big circle to indicate the blending, and used an 418 arrow in a circle to indicate expansible information. We used 419 color to highlight the difference in species. Important infor-420 mation about each item appeared in bold font size while other 421 information appeared in smaller font sizes.

446
The POC was used at the beginning of the interview.

447
However, since our main goal was to explore the designs 448 and most users did not focus on technical details, we iterated

461
During the iterative design process, we conducted interviews 462 with 10 participants. The main objectives were to investigate 463 the prototype issues and to gain insight into the willingness 464 to adopt the system, the benefits, the challenges, and the 465 suggestions to ease the adoption. We used the insights to 466 inform the questionnaire design, which is explained in the 467 next section.

469
Our participants included five producers (P1 -P5) and five 470 consumers (C1 -C5), aged between 21 and 59. Table 2 lists 471 the profile of our participants. Most of the participants had 472 basic knowledge of blockchain technology, but only C1, C2, 473 and P4 knew about traceability systems. Note that C1 -C3 474 and P1 saw the earlier design than the rest. P4 and P5 were 475 managers in a coffee-production-related department in the 476 same company. They were interviewed in the same session, 477 as per the participant's request.

479
We asked participants whether the system reflected the practi-480 cal production process. Most suggestions about coffee infor-481 mation flow in early design came from C2, who studied coffee 482 production. We incorporated them into our design, which was 483 later confirmed by other producers. There are three notable 484 findings from the interview, which did not stand out when we 485 reviewed other materials. 486 First, mixing or blending beans from multiple sources is 487 crucial. Farmers in the same area might sell to intermediaries 488 or companies as a group (C3), in large quantities in some 489 cases (P5). The blending is still needed even for single-origin 490 coffee, as roasters might mix beans from multiple batches to 491 control the taste (C2). Second, the role of stakeholders in the 492 Thai coffee supply chain may be overlapped. Finally, there 493 are a lot of details in coffee processing, and some of them do 494 not have standards widely adopted by Thai coffee producers. 495 For instance, there is no standard grading, and roasting level 496 could differ from roaster to roaster (C2). P5 suggested that 497 the current form was fine and could leave other details in one 498 optional remark text box. 499 VOLUME 10, 2022 We asked participants what design(s) they prefer. Overall, 500 participants preferred the timeline-based design, while hav-501 ing a map provided additional information. Meanwhile, P2 502 did not like any designs as they were hard to understand.

503
He suggested that all information should be in one picture 504 without too many clicks. Other comments about each design 505 are as follows.

518
Participants had different opinions about this design. P3 liked 519 its ease to understand. P5 deemed the design interesting 520 and modern. The map allowed an overview of the overall 521 process (C2) but required many clicks to see details (C3).

522
It was deemed confusing (P5) and too much clutter (C5). C1 523 complained that some icons did not reflect the production 524 process and suggested using animation to illustrate the time 525 order. A clearer illustration of time order was also suggested 526 by C2. C1 and C3 suggested that it should appear as additional 527 information on other designs.

529
We asked participants whether they would use the system 530 if the system works. The answers generally depended on 531 their views on the benefits and challenges. Most customer 532 participants would try our application as it looked useful (C3, 533 C4, C5) or fun (C1). C3 said he would continue using it 534 if the information could reflect the quality. C2 commented 535 that if everyone in the supply chain agreed to disclose 536 the information, it would be good and beneficial to the 537 community. However, producers' willingness was deemed 538 a challenge.

539
All producers seemed reluctant to adopt our system, mostly 540 because of a lack of perceived benefits or motivation for 541 them (P1, P4, P5) and the lack of functionality they need 542 (P3). P2 would not adopt as he did not want to disclose 543 information, though he agreed that our system would be good 544 for customers. Similarly, P5 commented that their company 545 would not use this detailed information but agreed that our 546 system was good for recording information. C2 would not 547 98790 VOLUME 10, 2022 adopt our system since the current system already has all the 548 needed information.   thus could increase income and sales. 578 We were particularly interested in whether tractability 579 could add financial value to the product. Hence, we specif-580 ically asked the customer participants about the additional 581 price they could afford. C3 thought the traceability could add 582 20 -30% value to premium Thai coffee and 10% to imported 583 coffee. He normally would not drink expensive coffee, but 584 the system could ensure the origin, so he might try expensive 585 coffee occasionally. However, C2 thought the traceability 586 itself would not add financial value, but a good system and 587 certificate would. C1 would buy a cheaper one first but might 588 try the traceable one if the price difference is less than 20%. 589 We observed that the participants seemed reluctant to pay  The product would look more authentic (C1). Our system 597 could help ensure the product origin that the seller claimed 598 (C3). C5 believed it was good that the information could not 599 be edited. Additionally, it would be fair, as the blockchain has 600 no central party. 601

602
Our system could help customers make the decision (C1) and 603 get the coffee they want (C4). C5 said he would use the system 604 because he liked finding new coffee. The system could help 605 him understand and be more confident in the product.

E. PERCEIVED CHALLENGES OF THE SYSTEM
607 While C5 thought there should not be any challenge in adopt-608 ing our system, many participants raised issues that originated 609 from producers.  This problem particularly burdened farmers. Coffee farms 618 might be in areas with no electricity and internet (C2, P4). 619 Most farmers might not carefully keep a record, such as the 620 age of each tree (C2). Farmers might be illiterate or not 621 know how to use technology (C2, P2, P4). Furthermore, P3 622 also concerned an older generation who may have issues 623 in recording data, although he had no concern as a young 624 farmer.

625
2) DISCLOSURE OF TRADE SECRETS 626 P1, P3, and P5 agreed that they could disclose the information 627 that the system asked for. P2 explained that he could input 628 the publicly available information such as the date. However, 629 information such as his supply chain and blending methods, 630 are sensitive and would never be disclosed. P2 explained that 631 poor-quality substances are released into the mass market. 632 This is a problem, but no one wants to solve it, since it is 633 deemed a mutual benefit. The producers could make more 634 money while the customers could get affordable coffee. Sim-635 ilarly, P5 would not disclose price information. The farmers 636 might not want to disclose substances in soil (C2) or be 637 traced back (P2). C4 also mentioned the system disclosing 638 seller information, which might cause conflicts, cutting price 639 competition, and benefit loss. However, he thought it could 640 be a good competition.

642
P2 commented that if customers knew the type of beans the 643 roasters used for blending, they would think the final price is 644 more expensive than the cost of the beans, without thinking 645 that there would be other costs besides the beans. Thus, 646 disclosing sensitive information could affect sales and the 647 relationship between people in the supply chain. C3 worried 648 VOLUME 10, 2022 that bad reviews might ruin farmers, especially when the 649 review could not be deleted.  C1 suggested that the system should be more useful if user 702 identity is verified. A taste note should be recorded by certi-703 fied Q graders (C5), as there were differences between trained 704 and untrained tasters (P2). P4 suggested that our system 705 should have people who are trustworthy to verify or approve 706 the information. C1 commented that the information would 707 look more reliable if everybody, including competitors, had 708 that information.

4) SUGGESTIONS RELATED TO BUSINESS AND MOTIVATION 710
P4 and P5 commented that a company in a mass market would 711 not adopt the system. However, he recognized that small and 712 medium enterprises (SMEs) and large enterprises might have 713 different interests and practices. Our system could have dual 714 models for each type of business. Additionally, the system 715 should be suitable for small coffee shops that promote single-716 origin coffee. Aiming at companies that own the completed 717 supply chain (from planting to packaging) first should ease 718 the system development and deployment.

719
C4 and P5 suggested that producers might use it if they see 720 the benefit, so we should educate producers and give them 721 motivation. Additionally, P2 commented that if we could 722 bring the government to talk with roasters and draft the policy 723 together, the system would be completed and good for the 724 Thai coffee community.

726
We confirmed and expanded the findings from the inter-727 view by surveying users for their opinions using a ques-728 tionnaire. From the interview, one major challenge was the 729 disclosure of sensitive information. We deemed the need to 730 address this challenge before further system development 731 since the measure could affect the architecture and user expe-732 rience. The selected proposal was providing an option for 733 users to control information access by themselves, which 734 could compromise the transparency property of blockchain. 735 To investigate our proposal, we added related questions 736 to the questionnaire. The notable additional question was 737 ''How does allowing relevant parties to choose to disclose 738 information by themselves affect your decision to use the 739 system?'' 740 A. PARTICIPANTS 741 We received answers from 350 participants. Modal values of 742 the participant's profile included female (56%), age group 743 between 40 -49 (38%), with master's degree (46%), and 744 using mobile applications more than 5 hours per day (37%). 745 Figure 5 reports the participant statistics. Six participants did 746 not specify their gender. Most participants did not know of or 747 knew but never used traceability applications or blockchain 748 applications, as illustrated in Figure 6. 749 Table 3 reports participants' experience. The major-750 ity of the participants (79%) had only consumption 751 98792 VOLUME 10, 2022  After learning about our application, 181 participants (52%) 770 thought our application was following the current practice. 771 However, 119 participants were unsure about the process and 772 50 participants indicated the need to improve the flow. One 773 reason was that most of the participants were consumers with 774 limited knowledge about the coffee supply chain. In addition, 775 some participants had issues with the survey (e.g., unclear 776 questions or unclear demonstration).

777
Most participants preferred timeline-based visualization, 778 as illustrated in Figure 7. For others, four participants sug-779 gested an option to choose visualization, while two suggested 780 a mixture of visualizations. One professional producer and 781 one participant, who had experience in planting, suggested 782 more details in the visualization. One participant suggested 783 an overview first, then detail on demand. One participant, 784 who had more than ten years of experience in all produc-785 tion processes, indicated that all visualizations were not 786 understandable.  Figure 9 shows the challenges in the adoption of the system in 830 general. The most concerning issue was inaccurate or incom-831 plete information (71%). Participants were also concerned 832 about disclosure of trade secrets (52%), difficulty in infor-833 mation filling (49%), and problems between people in the 834 supply chain (28%), respectively. Problems with the Revenue 835 Department were the least concerned issue (15%). A small 836 number of participants were also concerned that the appli-837 cation might increase the cost or price and that there could 838 be problems related to infrastructure such as the readiness of 839 equipment.

841
We specifically asked participants how blockchain affects 842 the decision to use the system. The responses showed that 843 67 percent of participants, including 80 percent of profes-844 sional producers, saw a positive influence of blockchain in 845 application adoption, as illustrated in Figure 10. Table 4 846 shows comments about how blockchain affects their decision 847 to use the system from 202 participants. Note that we also 848 found one case with contradicting answers: the participant 849 indicated a negative influence of blockchain but commented 850 that blockchain increases confidence in the information. The most notable benefits were related to information, 852 which was indicated by 67 participants. The blockchain had 853 a positive influence on the adoption as it could help in 854 origin checking. The information was deemed transparent, 855 verifiable, traceable, authentic, accurate, and/or credible. The 856 system could help prevent fraud, counterfeiting, as well as 857 information alteration. While 48 participants just repeated 858 their answers without further explanation, 44 participants 859 98794 VOLUME 10, 2022 Other benefits included benefits related to product or quality 862 (e.g., help improving or controlling product quality), bene-  We also investigated how our proposal to provide optional 886 disclosure affects the decision to use the system. Figure 11 887 illustrates the willingness to adopt our application, with and 888 without optional information disclosure. Interestingly, the 889 option to select disclosed information only slightly decreased 890 the willingness to adopt the application (3.86 on average 891 or 2% decreasing). 71 percent of participants, including all 892 professional producers, gave the same score regardless of the 893 disclosure option. Two participants dramatically decreased 894 their willingness to adopt the application with the disclosure 895 option (i.e., from 4 or 5 to 1). Unfortunately, only one reason 896 ''the trade secret'' was given without any other explanation. 897   FIGURE 11. The willingness to adopt our application, where 5 was to ''definitely use'' the system. Table 5 shows comments about the adopting our 898 application with optional information disclosure from 899 221 participants. 86 participants saw that optional disclosure 900 had a positive or good influence or was acceptable. Notable 901 benefits of the option were (1) respect for privacy or trade 902 secrets and (2) increasing trustworthiness and confidence. 903 One participant explained that by allowing producers to 904 choose what to disclose, producers could show the strengths 905 that they want consumers to know, which could ease the 906 purchasing decision. 43 participants thought that optional 907 disclosure had little or no influence on adoption, while the 908 opinion of 37 participants was more dependent on the dis-909 closed information. For example, several participants com-910 mented that they had more trust in products that disclose 911 more information. Nevertheless, the disclosure option could 912 have a negative influence, as given by 24 participants. This 913 option could make users unsure about filled-in information, 914 especially in terms of completeness. Some participants saw 915 no use of the application if many producers chose not to 916 disclose the essential information.   Table 6 shows other comments, particularly suggestions,  Table 7 lists the benefits, challenges, and suggestions 953 derived from our studies. We include remarks that discuss 954 whether our findings are similar to or varied from other 955 works. Note that we mainly point to our previous work [42] 956 since the review covered various blockchain-based systems 957 that were applied to agri-food supply chains.

958
In general, our findings support existing knowledge, with 959 some variations related to sales enhancement, problems with 960 the revenue department, suggestions related to commerce, 961 suggestions related to information or verification, and sug-962 gestions related to business and motivation. The order of 963 importance of the challenges also differs from the related 964 works summarized in Table 1. 965 We summarize our findings when developing a blockchain-966 based coffee traceability system into implications for easing 967 the adoption and design implications. While the customer participants could see the benefits of 970 our system, the source of their concern stemmed from the 971 producers. In the interview, the producer participants could 972 see extra work and problems with unclear immediate benefits. 973 However, we observed that farmer participants were willing 974 to adopt our system more than the producers in the middle of 975 the supply chain. Both the farmer participants and customer 976 participants mentioned direct purchasing, while the roaster 977 participants were afraid of losing customers. This conflict 978 could lead to difficulty in the adoption. However, the results 979 from the survey hinted that professional producers were will-980 ing to adopt the application, especially if we could address 981 their concerns about the trustworthiness and completeness 982 of the information, disclosure of sensitive information, and 983 difficulty or extra work in information filling. We discuss 984 suggestions to address these concerns in the following points. 985 The main challenge of the system adoption was ensuring 989 the trustworthiness and completeness of the information, 990 which was not directly related to blockchain in the view of 991 practitioners and our research team. Still, it was mentioned 992 repeatedly in our interviews and survey, apparently when 993 we asked about the blockchain. While blockchain technol-994 ogy ensures the integrity of information, the trustworthi-995 ness and completeness of the information entered remain an 996 unsolved research question, as discussed in many reviews 997 (e.g., [11], [61]). A human can enter wrong data inten-998 tionally or unintentionally. Unlike bitcoin, blockchain-based 999 traceability systems are too complex to design bullet-proof 1000 smart contracts to prevent bad data from entering the sys-1001 tem. Bager et al. [40], for example, suggested using IoT to 1002 mitigate this issue. However, additional sensors could further 1003 increase the cost, and the data could still be manipulated.

1004
Fortunately, the participants seem to be satisfied with 1005 trustworthiness verification. Identity verification, review, and 1006 rating exist in many applications nowadays. Customers could 1007 consider these factors to justify information from unknown 1008 sellers on online shopping sites, which have trustworthiness 1009 issues similar to our case. Our participants also mentioned 1010 TABLE 7. Benefits, challenges, and suggestions from user interviews and survey. The interview column reports the ID of the participant who mentioned the topic. The asterisks (*) indicate that the participants mentioned the topic when we specifically asked about concerns regarding blockchain. The survey column reports the number of participants who agreed with the topic (for benefits and challenges) or mentioned the topic (for suggestions).  One concern of the producers for adopting our system was 1018 the disclosure of sensitive information, but the customers 1019 recognized this concern. We found that the customer partic-1020 ipants were generally interested in public information, as a 1021 part of package or advertisement messages. For instance, 1022 they were only interested in the final price of the product 1023 and the planting city or province. The cost of raw materials 1024 or farmer names were considered good to know, but not 1025 necessary. The suggestion to record the cost for transparency 1026 by Bager et al.    In this research, we came up with various design alterna-1074 tives that present the same set of data and found that the 1075 timeline-based design was preferred. This preference could 1076 be because time is an important factor in coffee quality.

1077
Though it might not be evaluated against other designs, the 1078 timeline-based design had been implemented by many sys-1079 tems (e.g., eggs [62] and olive oil [63]). However, this design 1080 required careful consideration when the supply chain became 1081 more complex. In our case, we summarized and grouped 1082 items, but it may not be feasible in some other cases.

1083
The map-based design was also a good alternative with 1084 better scalability if the implementation could hide or group 1085 the pinned locations and paths in relation to the zoom level 1086 of the map. To our surprise, many participants also viewed 1087 text-based design positively. Hence, it could be a good, 1088 budgeted, and easy-to-implement alternative for traceability 1089 systems in general. Accuracy and completeness of information were the major 1093 concerns of users. Our designs included certificate icons with 1094 pop-up certificate detail, which could be seen as one method 1095 of trustworthiness verification. However, since we did not 1096 highlight it in the video demonstration, many participants 1097 might not notice this feature and raise concerns about the 1098 verification. Thus, it is important to make verification of 1099 trustworthiness or completeness of information apparent to 1100 users to ease their concerns.

1102
Our study has three main limitations. The first limitation 1103 is the number of participants and their background. Most 1104 of our participants were customers. Also, conducting online 1105 interviews meant we excluded users who were not familiar 1106 with the technology. Designing for people with low digital 1107 literacy is another challenge that should be addressed in the 1108 future. Though we believe there should not be significant 1109 changes to the benefits, challenges, and suggestions, different 1110 user groups could lead to a redesigned information flow and 1111 user interface.

1112
The second limitation is the lack of interaction between 1113 participants and the prototype. We report all findings, includ-1114 ing what participants might miss or misunderstand due to this 1115 limitation, as they could highlight important features or poor 1116 designs from a user perspective. For instance, we suggested 1117 apparent data verification because most participants missed 1118 our certificate icons from the video demonstration.

1119
The last limitation is the implementation. We implemented 1120 our first design as a web-based mobile-friendly prototype 1121 with Hyperledger Fabric as a blockchain layer. However, 1122 findings from the interviews suggested major modifications, 1123 and we decided to use an interactive prototype to quickly 1124 iterate the design to reflect users' suggestions and better 1125 understand stakeholders before going to actual implementa-1126 tion. For instance, our implementation only supported input 1127 from three types of users (farmers, millers, and roasters), and 1128 each user had one role, which is a false assumption in the Thai 1129 coffee supply chain. In the future, we need to implement the 1130 design and convince more stakeholders to participate in the 1131 blockchain.

VIII. CONCLUSION
We developed a blockchain-based traceability system for 1134 coffee in Thailand. The system could record essential steps 1135 in coffee production. The information flow and visualization were informed by user comments. Our design provided flexibility by allowing mixing beans from multiple sources in each step, partial usage of beans, and multiple user roles. 1139 We compared text-based, timeline-based, and map-based