This study’s primary objective was to assess the efficacy of financial incentive strategies for achieving person-level loss of ≥5% of baseline weight. The trial spanned three 6-month phases: phase I, incentivized initiation with 13 biweekly group sessions; phase II, incentivized maintenance with 3 monthly group sessions plus 5 monthly individual calls; and phase III, a 26-week, nonincentivized maintenance program with 3 bimonthly calls. Group sessions and calls were led by a registered dietitian.

Participants with obesity were recruited in Madison, WI, and Durham, NC, USA. The inclusion and exclusion criteria are detailed in ClinicalTrials.gov NCT04770909 and published [16]. Related to the Log2Lose technology, participants met requirements if they were able to speak and read English; able to download and use the Fitbit and MyFitnessPal smartphone apps daily; owned a smartphone with a data and texting plan; had reliable internet access; had the ability to participate in video calls via a smartphone, tablet, or computer; and as described below, later in the trial agreed to create a Gmail account after Google acquired Fitbit.

Participants were randomized to one of the 4 arms: incentives for both dietary self-monitoring and weight loss, with partial incentives if only 1 criterion was met (arm 1, combined); incentives for dietary self-monitoring only (arm 2, diet only); incentives for weight loss only (arm 3, weight only); or no financial incentives (arm 4, control) [16].

All participants received a BodyTrace cellular scale and were encouraged to weigh themselves regularly: twice weekly in phase I (weight loss) and once weekly in phase II (maintenance). Participants were instructed to keep the scale on a stable surface and restrict use by others. Participants were asked to track daily food and beverage by using a smartphone app and at 6-month intervals, to wear an activity tracker for 7 days.

All participants received automated motivational text messages twice weekly for 78 weeks (phases I-III), reinforcing educational themes and behavioral skills from group sessions and counseling calls. In phases I and II, participants in incentivized arms received weekly text messages indicating whether they had earned an incentive. Messages were sent between approximately 8 AM and 11 AM local time, with no adjustment for travel across time zones. Due to the COVID-19 pandemic, remote weights replaced in-person measurements as the primary outcome. Prior research supports the validity of remote monitoring compared with in-person weighing, with the added convenience of avoiding travel [17].

Control-arm participants received the same devices and platform access but no financial incentives. All participants received the same frequency and timing of motivational text messages, which were standardized prompts rather than tailored feedback. No additional behavioral counseling was provided via the platform.

This study reports the outcomes for the total number of text messages received by day of the week and message type, the delivery success rate, the total number of weights transmitted, the average number of daily weigh-ins per participant, and the total number of calorie entries received. It summarizes the number of incentive payments processed automatically by the software compared to those processed manually. Finally, we report common technical and troubleshooting examples.

Participants were randomized equally to one of the 4 arms by using a computer-generated block randomization sequence accessible only by the study statisticians. Participants were blinded to the incentive criterion of other arms. Sample size and power calculations were conducted using the methodology for multi-arm tests of proportions in PASS 2019. Further details are available in the published protocol paper [16].

Study conduct was overseen by the University of Wisconsin-Madison institutional review board (approval 2020-1693) and a Data Safety Monitoring Board that met at least annually throughout the study period for ethics review and approval. All participants provided informed consent and could opt out of the trial at any time. People completed a web-based prescreening survey, followed by telephone screening and then final in-person eligibility screening and informed consent. Data were stored in secure password-protected databases only accessible to institutional review board–approved study staff. Compensation was provided for assessment completion at weeks 26 (US $40), 52 (US $25), and 78 (US $35).

The Log2Lose platform was developed to automate data collection, incentive processing, and monitoring procedures in support of the randomized trial. The system integrated weight and dietary data from consumer devices and applications, applied prespecified algorithms to determine incentive eligibility, and delivered feedback to participants via text messages. Research staff had access to a dashboard that allowed monitoring of data completeness, device connectivity, and message delivery. These features were designed to maintain intervention fidelity and to provide safeguards against data anomalies or technical failures. Importantly, the platform was developed to be adaptable, enabling adjustments to accommodate vendor-driven changes such as the discontinuation of the MyFitnessPal–Fitbit application programming interface (API) and shifting the US Federal Communications Commission text messaging regulations, which might otherwise have compromised study integrity.

The platform applied prescheduled weekly algorithms to determine incentive eligibility (Figure 1). It operated on a Heroku platform-as-a-service cloud server, utilizing a Ruby on Rails application linked to a PostgreSQL database. Data from BodyTrace scales (weights), Fitbit activity trackers (steps per day, minutes active), and Fitbit and MyFitnessPal dietary apps (calories logged per day) were securely imported into the platform using a web API. The BodyTrace scale measured weight and had an embedded cellular chip using the global system for the mobile communications network, which transmitted weight data to the BodyTrace servers. Each scale’s international mobile equipment identity number was recorded at enrollment, linked to the participant ID, and used to retrieve data via API.

During enrollment, participants created a Fitbit account and received an authorization link from the platform. Upon authorization, the platform linked the Fitbit account to the participant ID, enabling import of daily calories, steps, and activity minutes via API. MyFitnessPal did not offer an API to pull data into the platform. Therefore, participants also linked their Fitbit account to the MyFitnessPal app so that calorie data from MyFitnessPal would transfer over to Fitbit.

Participant information (eg, enrollment, surveys) from the Research Electronic Data Capture (REDCap) system, a secure online platform for data capture and storage, was integrated into Log2Lose via API for analysis and study management [18,19]. This integration also included automated notifications for participants and staff.

We summarized the total number of messages received by day of the week and by type of message (incentive or motivational), as well as the delivery success rate. We also summarized the total number of weights transmitted, the average number of daily weigh-ins per participant, and the total number of calorie entries received. We described the number of incentive payments processed automatically by the software versus those processed manually. Descriptive statistics were used to summarize study participation, message distribution, weight recordings, calorie entries, and incentive payments. Frequencies and percentages were calculated for categorical outcomes, including message type, delivery success, and incentive processing mode, while means, standard deviations, and ranges were used for continuous outcomes such as daily message counts, number of weight recordings per participant, and calorie entries. Temporal patterns in logging behavior were described narratively in relation to technical disruptions, and rare events such as delivery failures and manual incentive processing were reported descriptively given their low frequency.

By the end of this study, 126,283 messages were sent, with a daily maximum of 1414. Message distribution was balanced between sites: 62,276 to 348 participants in Wisconsin and 64,007 to 357 participants in North Carolina. One participant withdrew before receiving the first message; thus, 705 participants received messages.

Incentive messages were sent primarily on Mondays (n=26,111) and to a lesser extent on Tuesdays (n=300). Motivational messages were sent on Tuesdays (n=50,352) and Fridays (n=49,319). An additional 201 messages were sent manually on a Wednesday to report a technical error.

Overall, >99% of the messages were delivered successfully. Delivery failures typically resulted from international travel, prolonged service loss, or extended periods with phones turned off. The most common failure type involved messages flagged as spam by mobile carriers despite Twilio’s trusted-source agreements. Possible triggers included absence of sender identification, use of dollar signs or shorthand characters, excessive exclamation marks, and repetitive content, although carriers did not disclose specific criteria.

By study end, 706 participants recorded a total of 181,285 weight measurements from the cellular scales. Of these, 94.9% (172,096/181,285) were accepted. The highest number of weights transmitted in a single day was 349. On weighing days, participants recorded an average of 1.24 (SD 0.316) measurements, reflecting multiple consecutive weigh-ins.

There were 114,144 daily calorie entries from 694 participants, with a single day maximum of 256. Logging declined sharply around the 90-day mark, partly due to MyFitnessPal–Fitbit API issues in December 2021. Initially, Fitbit data were retrieved via MyFitnessPal for cohort 1, but persistent API problems prompted a switch to Fitbit-only logging. This improved reliability by removing third-party syncing, though participants reported lower satisfaction with Fitbit’s interface. Google’s acquisition of Fitbit required all users to have Google accounts by the end of 2025. In July 2024, Fitbit consolidated its dashboard into the mobile app, ending web access and affecting our final cohort.

The platform automatically calculated and queued 35,187 incentives during phases I and II. Only 222 incentives (0.6%) across both sites required manual processing, mainly due to technical or behavioral complexities (Figure 2; Table 1).

Novel strategies are needed to support both the initiation and maintenance of weight loss. Because weight management requires frequent, repeated behaviors, technologies that minimize barriers to tracking may be particularly effective. Mobile technologies allow for continuous monitoring of progress, and with appropriate technical support (Table 1), Log2Lose demonstrates that near real-time, large-scale data collection, and feedback are feasible across time zones and diverse settings. Our experience suggests that automation is scalable when paired with pragmatic, responsive support.

Our findings contribute to the broader literature on incentive-based weight management interventions. Prior studies have shown that financial incentives can support initial weight loss but often face challenges in sustaining long-term adherence [5-7]. More recent work such as Ladapo et al [4] have demonstrated the potential of both outcome-based and behavior-based incentives; yet, questions remain regarding how best to deliver and sustain these approaches in real-world contexts. Log2Lose extends this body of work by demonstrating that incentives can be operationalized at scale through a fully remote, multisite trial supported by consumer technologies. Unlike earlier studies that relied on in-person contact or static incentive structures, our trial highlights the importance of adaptability in the face of shifting technology, regulation, and participant needs. In this way, Log2Lose not only reinforces prior evidence that incentives can be effective but also advances the field by offering a model for integrating automation, flexibility, and responsive technical support into long-term digital health interventions.

The COVID-19 pandemic necessitated a swift pivot to fully remote delivery. Group weight loss sessions were moved to videoconference, nearly all technical support was provided remotely, and primary outcome data were collected with cellular scales instead of during in-person visits. Preparation during our pilot phase positioned us to navigate an increasingly complex technology and regulatory environment during and after the pandemic [11,20].

To avoid reliance on a single application, the platform was designed to be app-agnostic. This flexibility proved essential when the MyFitnessPal–Fitbit API link ended mid-trial. We migrated dietary logging entirely to Fitbit and routed integrations directly through Log2Lose rather than third-party-to-third-party connections. Although this migration required substantial engineering effort, it improved data integrity and reduced opaque synchronization issues. Later vendor changes such as Fitbit’s consolidation after Google’s acquisition necessitated further protocol revisions and staffing adjustments.

The Federal Communications Commission’s introduction of new text messaging regulations added another layer of complexity. Maintaining message delivery required continual problem identification and resolution, often resembling a game of whack-a-mole, especially for issues that threatened intervention integrity. Some messages, particularly those with repetitive content or dollar amounts, were flagged and blocked by telecom carriers. Weekly incentive messages were not personalized with participant-specific information such as names or pronouns, which may have increased the likelihood of being flagged. Although our platform attempts to resend failed messages automatically, evolving security protocols necessitated manual weekly surveillance. The research team reviewed failures by carrier and content, and small adjustments such as removing numerals often restored successful delivery in subsequent weeks. In today’s rapidly evolving landscape of technology, regulations, and proprietary ownership, adjustments are to be expected in multiyear studies. Building flexibility into planning is therefore essential.

Automation processed approximately 99% (126082/126283) of the data and incentives on a fixed weekly schedule, reducing errors and preserving fidelity at scale. Automated systems deliver interventions consistently and without bias, but skilled technical support remains indispensable. Our experience underscores that adequate staffing and technical expertise are critical to the success of large-scale digital health programs.

The platform’s monitoring tools enabled staff to address many technical problems proactively without contacting participants. However, when automated processes failed, whether for email alerts, class reminders, surveys, text messages, or weight data collection, immediate intervention was required. With over 700 participants, resolving issues demanded substantial time and resources. For instance, a scale issue we identified was that placement near a bathroom wall with baseboard molding would create a fulcrum effect, causing inaccurate weight measurements. Daylight savings time caused problems with text message delivery that we did not plan for. In each case, staff investigated root causes through messaging API error codes and implemented targeted solutions to prevent recurrence.

Although more than 85% of US adults own a smartphone [21], generalizability is limited to those with smartphones and data plans and the digital literacy to use the devices, apps, and videoconferencing technology. Message delivery receipts were recorded, but message opens were not tracked because text messaging lacks reliable open tracking functionality. Email open rates were not implemented to protect participant privacy. Data inaccuracies could occur when other household members used the scale, participants underreported dietary intake, or technology literacy was limited. Although such inaccuracies increase variability in efficacy estimates, they are inherent to real-world settings, generalizable to similar interventions, and likely to bias results toward reduced effect sizes rather than inflate estimated effects.

Log2Lose demonstrated the feasibility of a scalable, digitally enabled weight management program. Success depended on adaptable software built on consumer technology, coupled with continuous monitoring and responsive support. By leveraging widely available consumer technologies, the platform adapted to regulatory and technological changes while maintaining intervention fidelity. Although automation allows for broad reach and efficiency, dedicated staffing for technical support is essential. This program model can inform health care providers, payers, and employers implementing obesity interventions, supporting a healthier workforce and potentially reducing health care costs.