Key Findings and Recommendations

Key Findings

1. Smart-device activity and sleep followed recurring weekly patterns

Average activity and sleep varied across the week:

• Sunday: lowest average activity at approximately 6,607 steps

• Sunday: highest average sleep duration at approximately 7.55 hours

• Saturday: highest average activity at approximately 7,752 steps

• Thursday: lowest average sleep duration at approximately 6.69 hours

Sunday therefore appeared to function as a lower-activity, higher-sleep day within this sample, while Saturday represented the week's highest activity level.

These patterns are useful for thinking about the timing of wellness content, but they should remain descriptive rather than causal. The analysis does not establish that the day of the week itself causes changes in activity or sleep.

2. Sedentary time was more strongly associated with sleep duration than step count

Activity and sleep records were joined by user ID and date, producing 422 matched observations.

Within those matched observations:

• Daily steps vs. sleep duration: r = -0.19

• Sedentary minutes vs. sleep duration: r = -0.53

• Average sleep duration: approximately 420 minutes, or 7.0 hours

Sedentary time therefore had a substantially stronger negative association with sleep duration than total daily steps.

I also tested whether activity on one day was related to sleep recorded the following day. That relationship was even weaker at approximately r = -0.06. This supported a cautious interpretation: activity and sleep clearly vary across the week, but this dataset does not establish a strong direct relationship between taking more steps and sleeping longer.

3. Stress and sleep had the strongest association examined in the supplementary dataset

Within the 374-record Sleep Health and Lifestyle dataset:

• Stress vs. sleep quality: r = -0.90

• Stress vs. sleep duration: r = -0.81

• Stress vs. daily steps: r = 0.19

The stress–sleep quality relationship corresponds to approximately R² = 0.81 in a simple linear relationship, making it considerably stronger than the activity-based relationships identified in the Fitbit data.

This supports testing Bellabeat messaging around stress management and sleep wellness rather than focusing exclusively on step counts.

However, the Sleep Health and Lifestyle dataset is synthetic/compiled. These findings should therefore be treated as hypothesis-generating and illustrative, not as evidence that the same relationships necessarily exist among Bellabeat customers.

4. Descriptive gender differences appeared in stress and sleep, but not physical activity

In the supplementary dataset:

• Average physical activity level: 59.14 for women vs. 59.20 for men

• Average sleep quality: 7.66 for women vs. 6.97 for men

• Average stress level: 4.68 for women vs. 6.08 for men

Physical activity averages were almost identical between the two groups, while larger descriptive differences appeared in stress and sleep quality.

For a women-focused wellness brand such as Bellabeat, this provides a useful question for future research: whether first-party customer data similarly shows that sleep, stress, recovery, or other wellness dimensions provide more useful differentiation than activity metrics alone.

Because these values come from synthetic data and no causal inference was performed, I would not present them as evidence of broad gender differences in the population.

5. Secondary tracking features showed substantially lower engagement

Feature use varied considerably within the Fitbit data.

• The combined daily activity data contained 35 unique user IDs.

• The available sleep data contained 24 unique user IDs.

• Only 13 users logged weight information.

Weight-tracking users averaged approximately 8,150 steps per day, compared with approximately 6,774 steps among users who did not log weight. They also averaged fewer sedentary minutes.

This does not mean that weight logging caused greater activity. A more reasonable interpretation is that users who voluntarily use additional tracking features may already be more engaged with health and fitness behaviors.

From a product perspective, this raises an important question: can Bellabeat improve broader feature adoption by making users more aware of what their device and application can track?

These values and correlations are documented in the SQL Query Log and final presentation. 

Marketing and Engagement Recommendations

Recommendation 1: Position Bellabeat around whole-person wellness, not step counting alone

Activity tracking is already a standard capability across the wearable market. The analysis suggests that Bellabeat could differentiate its products by giving greater visibility to sleep, stress management, recovery, mindfulness, and personalized wellness guidance.

Potential applications include:

• Highlight sleep and stress-related capabilities of Bellabeat devices and the Bellabeat app.

• Present activity tracking as one part of wellness rather than the primary value proposition.

• Use Bellabeat Membership to connect individual metrics with personalized wellness recommendations.

• Test marketing messages centered on recovery, sleep, stress, and balance against more traditional activity-focused messaging.

Because the strongest stress findings came from supplementary synthetic data, I would recommend testing this positioning with first-party Bellabeat customer data before making a large-scale marketing shift.

Recommendation 2: Use weekly behavior patterns to guide the timing of engagement

The Fitbit data showed recurring differences by day of the week. Bellabeat could test whether sending context-specific content at relevant times improves engagement.

Potential examples include:

• Sunday: recovery content, sleep preparation, mindfulness, gentle movement, or hydration

• Thursday: sleep reminders, wind-down prompts, stress check-ins, or breathing exercises

• Saturday: hydration, recovery, activity celebration, and post-exercise wellness content

• Monday or Friday: movement reminders during periods that showed relatively high sedentary time

Rather than sending every user the same notification schedule, these patterns could serve as starting points for A/B testing and eventually be replaced with notifications based on each user's own behavior.

Recommendation 3: Build onboarding around multi-feature adoption and personalization

Not every Fitbit user in the sample used the available sleep or weight-tracking features. This creates a potential product-engagement opportunity.

Bellabeat could:

• Introduce activity, sleep, stress, hydration, and other available tracking capabilities during the first week of onboarding.

• Use in-app prompts to gradually introduce features rather than presenting every capability at once.

• Explain the benefit of providing multiple types of wellness data: better personalization rather than simply “more tracking.”

• Use Bellabeat Membership as the next step for users interested in more individualized guidance.

• Monitor feature adoption, weekly active use, notification engagement, retention, and membership conversion to determine whether the strategy is effective.

A useful future experiment would compare standard onboarding against a structured multi-feature onboarding pathway and measure whether broader feature adoption improves long-term engagement.

Limitations and Next Steps

Several limitations prevent the results from being generalized directly to Bellabeat's customer population.

The primary Fitbit dataset is small, was collected in 2016, and contains no demographic information. The combined files also contained 35 unique user IDs despite the source documentation describing 30 participants, an unresolved source-data discrepancy.

Sleep data were available for a smaller subset of participants, and weight data were especially sparse. The convention used for assigning sleep records to dates was also not completely clear. Testing same-day and next-day activity relationships did not produce a strong enough relationship to resolve that issue.

The supplementary Sleep Health and Lifestyle dataset adds demographic and wellness variables but is synthetic/compiled rather than directly observed, which substantially limits the conclusions that should be drawn from it.

Finally, all relationships in this case study are observational and correlational. They do not demonstrate cause and effect.

Recommended Next Steps

• Analyze current first-party Bellabeat customer data.

• Use a larger and more current wearable dataset with reliable demographic information.

• Study feature adoption and retention longitudinally.

• Test notification timing and onboarding strategies experimentally.

• Investigate Bellabeat-specific wellness variables, including menstrual-cycle and recovery data where appropriate.

• Determine whether using a broader range of Bellabeat features is actually associated with improved engagement or wellness outcomes over time.