The goal of the WP2 ‘Methodology’ workpackage of the Smart Survey Implementation project (SSI) is to find out what general methodological elements trusted smart surveys should have so that they can be used in statistical production by European NSIs. Each task focuses on either an ‘opportunity’ or ‘threat’ that was identified in the TSS I framework and pilot recommendations for smart surveys. The four subtasks are: 1. The successful recruitment of participants for smart surveys. 2. Using machine learning to improve Human-Computer Interaction in smart surveys. 3. Usability and Human-Computer Interaction in smart surveys. 4. Integrating smart surveys with traditional survey methods by estimating the mode effect. We refer to deliverable M6 (Review stage) for a discussion of learnings from past findings on projects conducted in the context of the European Statistical System and the wider academic context with regards to these four key challenges, and deliverable M14 (smart baseline stage) for an overview of how the central questions from deliverable M14 resulted in a series of smaller and larger field tests that were carried out between 2024 and 2025. In this deliverable, findings from the M6 and M14 deliverables are in places summarized when this is necessary to understand the current deliverable, but details in the design and motivations of our study are here not explained in detail. The goal of the current deliverable is concentrated on how to carry out a smart survey; the methodology of recruiting respondents, how to deal with sensor data, design the app and user Interaction in the app, and how to integrate smart surveys. The goal was to develop an end-to-end methodology for smart surveys. The two use cases that we concentrated on were the Household Budget Study (HBS) and Time Use Survey (TUS), whch both are surveys that are part of the European Statistical System (ESS). Of central importance are several large and small field tests. The large tests aim to answer the question of how respondents can be successfully recruited into smart surveys (task 2.1) and how to integrate smart surveys with traditional surveys (task 2.4). The large field tests were conducted in Norway (HBS), France (TUS & HBS), Belgium (TUS), and Germany (HBS). Norway and France used a smart survey app which was self-developed, and Germany and Belgium used the MOTUS platform as developed by Hbits. All countries used the general population as the target population and drew fresh samples to conduct the field test following a general design, where some key elements of the field tests are shared across the countries. Respondents are recruited using an offline method (e.g., recruitment via interviewers or postal mail). This allows for the comparison of country-level differences in, for example, the success of particular recruitment strategies. Apart from common elements to the fieldwork design, countryspecific variations on the design were also used to tailor tests to local circumstances, and to test specific design elements related to recruitment (task 2.1) or mode measurement effects (task 2.4). Deliverable M14 explains these choices in detail. Chapter 2 discusses the outcomes of these tests for recruitment, and concludes that it is difficult to recruit respondents from fresh-probability samples. The use of interviewers in a tailored invitation strategy can really benefit the success of recruitment for smart surveys. Within the project, additional tests around different ways to use interviewers were scheduled for Italy and the Netherland, but these tests did not happen within the timeframe of this project. In the case of Italy issues around the conclusion of a DPIA, and in the Netherlands issues in the IT-system at Statistics Netherlands precluded results from these tests to be included in this deliverable. These will be published separately after the conclusion of the project. Chapter 3 discusses how sensor data that are included as smart elements in a smart survey should be processed. For the HBS pictures of shopping receipts formed the basis of the smart data, whereas for TUS these wer geolocation data. In particular, this chapter focuses on how to guarantee that processed sensor data are of sufficient quality to be used in practice. Geolocation data are processed to generate a pre-filled time-diary of travel-episodes and non-travel episodes. In HBS, pictures of receipts are used to extract relevant product lines, read in the products and their prices, and subsequently link these to standard Coicop codes of products. This chapter shows how Machine Learning can be used to process sensor data, but also illustrates some of the difficulties there are on relying on Machine Learning only. In some cases, the quality of processed sensor data is insufficient; a human-in-the-loop may be necessary to further improve the quality of the data. Chapter 4 directly follows on the chapter on Machine Learning and studies how to integrate process sensor data in a smartphone app, and design the Human-Computer interaction between the respondent and app. At the core of this chapter are a series of small tests conducted in every country throughout the project. The goal of these smaller experiments is to technically test some of the microservices developed in workpackage 3 that process the sensor data, test the Machine Learning standards developed in task 2.2 and integrated in the microservice, and finally to test the Human- Computer Interaction features of smart surveys. This chapter describes the aspects of smart surveys for two microservices that process smart data in detail. It documents aspects of smart surveys that work well and highlights issues respondents face in practice in interacting with steps in the response process. From this chapter follow specific recommendations on how to improve the smart surveys related to TUS and HBS, but also for smart surveys in general. Chapter 5 investigates how outcome statistics change when moving from a traditional diary study in the context of Household Budget or Time Use, towards a smart survey. One of the main reasons to move to a smart survey is to decrease respondent burden and improve measurement quality. The chapter finds that indeed measurement quality changes when moving to a smart survey. For Time Use, these changes can be relatively large, and consist of changes due to increasing missing data problems, and improved measurement. Suggestions are given to potentially reduce measurement differences between survey modes, for example by tackling the issue of missing data in Time Use diaries, and how to estimate the size of the mode measurement effect in detail using statistical modeling techniques. This deliverable is concluded by chapter 6. This chapter integrates findings from the different chapters, and establishes a framework for how to design a smart survey methodology. We conclude that there is not one way to design a smart survey methodology. The specific design of a smart survey should be topic- and country-specific. Some of the country-specific considerations for how to design a smart survey may depend on aspects that extend beyond methodological issues, such as legal/ethical considerations (see WP5 of this project), the platform and IT infrastructure used (see WP 3), and most importantly organisational and business processes (See WP4). Still, we believe there is a common and fundamental choice to make on how smart a smart survey should be, and that a choice early on the design process can help guide all other choices. Finally, this deliverable contains a large section of appendices. These appendices are all designed as stand-alone documents that can be read when interested in more detailed results. These include country-specific results from the large field tests (Appendix A), details on what database to use to enrich geolocation data with the purpose of a location visits in the context of TUS (Appendix B), country reports from the usability tests (Appendix C), detailed results from mode-effect studies carried out in France (Appendix D) and a report on a feasibility study to develop a smart survey around energy use (Appendix E). We hope this deliverable will provide an impetus for further developing smart surveys within official statistics and beyond in the next years. This deliverable aims to show how smart surveys can be conducted successfully. It however also provides lots of specific areas where further research and tests are needed. We hope this deliverable will therefore both serve as a basis, but also as inspiration for the further development of smart surveys in years to come.