When Engagement Became Social: A 60-Day Network Case Study

At a Thailand-based enterprise IT services company, an in-person workshop lifted activity immediately. A rotating three-person network helped sustain the change: weekday activity +39.6%, weekly reach +35.8%, and 504 new pairing opportunities in two cycles.

Behavior-change programs often succeed in the room and fade at the desk. A Thailand-based enterprise IT services company serving automotive businesses gave us a useful natural experiment in what happens when an in-person activation is followed by a deliberately designed peer network.

On June 24, the company held an in-person workshop. Six days later, Happily launched Triad: a rotating group of three colleagues who can see one another's daily progress, prompt one another, and build a shared streak. We studied the 60 days surrounding those events to ask two questions: did the engagement increase last, and did the new network provide a credible explanation for why?

During the 20 complete workdays after Triad launched, weekday activity was 39.6% above its weekday-matched baseline—and above expectation on 18 of 20 days.

The short answer

The workshop appears to have created the initial activation. Triad then introduced 504 fresh peer connections across the company, creating a credible mechanism for keeping the behavior visible and socially reinforced. The combined intervention likely contributed to the sustained uplift, but the data cannot assign a separate causal share to each part.

Study at a glance

Company: Anonymized enterprise IT services company, 248 active employees
Study window: May 29–July 27, 2026
Interventions: Workshop June 24; Triad June 30
Primary outcome: Weekday daily active employees
Comparison: Same-weekday baseline plus 19 other qualified production Triad customers
Network evidence: Two Triad rotations and 1,387 check-ins

Finding one

The lift began in the workshop—and held after launch

The company averaged 65.9 active employees per baseline workday. Activity jumped immediately after the workshop, averaging 86.3 across the four workdays before Triad launched. Once Triad was live, the average increased again to 91.7 and remained elevated through four complete weeks.

Figure 1Weekday active employees at the anonymized case company. The stronger line is a five-workday moving average. July 28–29 are excluded because source ingestion was not fully reconciled.

Activity and response behavior by phase
Period	Workdays	Daily active	Roster participation	Daily responders
Pre-workshop	17	65.9	26.6%	58.8
Workshop only	4	86.3	34.8%	83.5
Triad live	20	91.7	37.0%	88.0

The matched post-launch expectation was 65.7 active employees per workday. The observed 91.7 represents a 39.6% lift, with a day-level 95% confidence interval from +26.1% to +53.0%. The first four post-workshop days were also higher, but that estimate is much less precise because the window is short.

The change reached more people, not just the same enthusiasts

Complete post-launch weeks averaged 153.3 active employees, compared with 112.9 across eight complete pre-workshop weeks—a 35.8% increase in weekly reach. The longer context makes the shift easier to see: the eight earlier weeks ranged from 99 to 131, the workshop week reached 161, and every complete Triad-live week remained above the extended baseline mean.

Figure 2Distinct active employees across eight complete pre-workshop weeks, the workshop week, and four complete Triad-live weeks. The workshop and incomplete final week are not used in the comparison means.

The product mechanism

Triad turns private intention into visible peer progress

Every two weeks, Triad creates a new group of three colleagues—occasionally four—chosen to create a useful bridge across the organization. For ten workdays, members see one another's progress, receive peer-aware prompts, and build a shared streak. The matching logic avoids recent repeats and, where possible, direct manager-report relationships.

Product previewA designed simulation based on the live Triad flow. It contains no employee data and is intended to explain the interaction, not reproduce every screen pixel-for-pixel.

This design matters because social reinforcement is not the same as broadcast communication. Laboratory and field research shows that behavior spreads more reliably when it is reinforced by multiple contacts, and that the diversity of those contacts can matter more than the raw number of messages. Triad gives each person a small, legible social context, then rotates it before the group becomes another fixed silo.

Finding two

Two rotations connected the whole active company

The first rotation produced 82 small components—mostly isolated groups of three. The second used entirely new pairings. When both cycles are combined, every one of the company's 248 active employees sits in a single connected network.

Figure 3An anonymized rendering of actual Triad membership. A line means two employees shared a Triad; it does not imply a friendship or a direct message.

The network Triad created
Network property	Observed	Why it matters
Active roster covered	248 / 248	The intervention reached the whole active company.
Unique pairings	504	Each edge was a new opportunity for reinforcement.
Repeated pairings	0	The second cycle expanded rather than recycled the network.
Cross-team edges	51.2%	More than half of the bridges crossed a team boundary.
Manager-report edges	0.2%	The structure was peer-led rather than hierarchical.
Largest connected component	100%	After two cycles, a path connected any two active employees.

A connected network does not mean everyone directly interacted with everyone else. It means social reinforcement can travel across bridges rather than stopping at team boundaries. This is the structural basis for the causal explanation: the workshop supplied energy; the rotations supplied new pathways through which the behavior could remain visible.

Finding three

A visible peer signal was associated with follow-through

Across 1,387 Triad check-ins, employees checked in on 30.9% of opportunities after a peer had already made progress visible, compared with 26.9% when no peer signal was present. That is a +4.0 percentage-point association.

Figure 4Member-day check-in rate, grouped by whether another Triad member had already checked in that day. This is an association, not a randomized causal effect.

Once a group-day started, 50.6% reached at least two members, and the median time from the first check-in to the next was 215 minutes. The pattern was also stable across rotations: 1.12 check-ins per group-day in cycle one and 1.13 through the first five workdays of cycle two.

Causal boundary

The peer comparison is consistent with social reinforcement, but it does not prove that one person's check-in caused the next. A common event or a generally motivated day could influence the whole group. The result is most useful when read together with the sustained company-level lift and the measured network structure.

Wider engagement

Triad's wider production cohort also increased

The first version of this study compared the case company with every sufficiently active company on the platform. That was useful for detecting a platform-wide timing effect, but it did not measure Triad's uplift elsewhere. We rebuilt the comparison around companies that actually launched a June 30 Triad cycle.

Figure 5Aggregate weekday activity across 19 other qualified production customers with a June 30 Triad cycle. Demo, trial, test-drive, inactive, excluded, and Happily internal accounts are removed.

Across those 19 companies, aggregate weekday activity was 10.9% above matched expectation through the first full cycle and 9.7% above expectation through July 27. Thirteen of 19 companies increased. The full-month number is slightly below double digits because the broader uplift moderated after the first cycle; the chart shows both windows rather than selecting only the higher result.

This is supporting evidence that the feature's effect was broader than one company. It is not a clean control for the case study: only two qualified production companies without the June 30 cycle met the same activity thresholds, too few for a credible untreated comparison.

Response behavior rose with activity

At the case company, daily question responders rose 49.9% versus matched expectation, and the employee response-day rate increased by 11.8 points, from 23.7% to 35.5%. Recognition also moved in the expected direction, from 8.4 to 10.1 events per workday, but its uncertainty range crosses zero. We therefore treat recognition as supporting evidence, not a confirmed effect.

How social networks can make behavior change stick

The result fits a broader body of social-network research. Centola's experimental work shows that clustered reinforcement from multiple peers can make a health behavior spread farther than long, sparse connections. Aral and Nicolaides found evidence of exercise contagion in a large natural experiment, with network position shaping the strength of the effect. Ugander and colleagues showed that structurally diverse contacts predict adoption better than contact count alone. Network interventions can therefore work by changing who reinforces whom, not only by increasing communication volume.

Organizational research points in the same direction: access to useful relationships across boundaries is associated with individual and group performance. But network studies also carry an important warning. People who are similar often connect with one another, which can make similarity look like contagion. The designed rotations help the mechanism case because the system introduced new ties deliberately, yet this study still does not randomize the intervention itself.

What this means

Decisions supported by the case study
Decision	Evidence	Practical implication
Pair activation with continuity	The workshop jump appeared immediately; the lift persisted for 20 post-launch workdays.	Do not treat the workshop as the entire behavior-change program.
Design the network	504 new edges, 51.2% cross-team, zero repeats, and one connected company network.	Who is grouped with whom is part of the intervention, not setup detail.
Keep the social unit small	Half of started group-days reached a second member; peer-visible opportunities had a +4.0-point association.	Make progress legible enough that a colleague can respond to it.
Measure outcomes beyond feature use	Daily activity, weekly reach, and response behavior all rose.	Use Triad check-ins to explain the mechanism, not as the only success metric.

Limitations

Every active employee at the case company received Triad, so there is no untreated internal control group.
The workshop and Triad launches were six days apart, preventing a clean separation of their causal contributions.
The broader production cohort removes known demo, trial, test-drive, inactive, excluded, and internal accounts, but companies still differ in size, workforce, configuration, and concurrent programs.
Only two qualified non-Triad production companies met the comparison thresholds, so the study does not use them as a credible untreated control.
The peer follow-through comparison is observational and may reflect shared motivation or common events.
The second rotation contributes only five complete workdays through July 27.
Platform activity is a useful engagement signal, but it does not capture every form of offline engagement or long-term business performance.

Bottom line

The evidence is strongest for a combined workshop-plus-network effect. The anonymized company's engagement rose sharply, remained elevated, spread to more employees, and coincided with a deliberately expanded peer network. The network evidence explains a credible way the change was sustained; it does not convert this observational case into proof that Triad alone caused the full uplift.

Cite this study

Happily Research (2026). When Engagement Became Social: A 60-Day Network Case Study. happily.ai/research/networked-behavior-change/

References

Centola, D. (2010). The spread of behavior in an online social network experiment. Science, 329(5996), 1194–1197.
Aral, S. & Nicolaides, C. (2017). Exercise contagion in a global social network. Nature Communications, 8, 14753.
Fowler, J. H. & Christakis, N. A. (2010). Cooperative behavior cascades in human social networks. PNAS, 107(12), 5334–5338.
Valente, T. W. (2012). Network interventions. Science, 337(6090), 49–53.
Ugander, J., Backstrom, L., Marlow, C. & Kleinberg, J. (2012). Structural diversity in social contagion. PNAS, 109(16), 5962–5966.
Sparrowe, R. T., Liden, R. C., Wayne, S. J. & Kraimer, M. L. (2001). Social networks and the performance of individuals and groups. Academy of Management Journal, 44(2), 316–325.
Cross, R. & Cummings, J. N. (2004). Tie and network correlates of individual performance in knowledge-intensive work. Academy of Management Journal, 47(6), 928–937.
Shalizi, C. R. & Thomas, A. C. (2011). Homophily and contagion are generically confounded in observational social network studies. Sociological Methods & Research, 40(2), 211–239.