happily.ai research All research
Case study · Customer A

All six human skills moved up in 11 months at one customer

In 11 months on Happily, a 160-person cohort at a Thai professional services firm added +0.74 rating points across the six human skills, with 91% of measurements moving in the positive direction. The gain held at every starting level, including the top quartile, which is the test that rules out regression to the mean as the explanation. The cohort represents 54% of the company.

+0.74
Mean rating change in 11 months (3.48 to 4.22)
91%
Of person-skill measurements moved up
2.5×
The cross-customer average gain

The companion longitudinal study across 2,630 individuals at 80 organisations gives the broader prior: human skill ratings rise with use, by an average of +0.29 points over a roughly year-long window, and rise more for heavier users. This case study zooms in on one organisation where the trajectory was steeper than the average and faster than the median.

The focal customer is a 295-person Thai professional services firm. 160 of their employees, 54% of the workforce, used Happily actively enough to clear the cohort bar for the cross-customer study. In a median analysis window of 337 days, the cohort's average rating across the six human skills (also called soft skills, or Power Skills inside Happily) rose from 3.48 to 4.22. Every one of the six skills moved up. Even people who started in the top quartile gained, which is the answer to the obvious sceptical question.

The headline that this article can support, after the corrections that the cross-customer study taught us to make: in this organisation, in this window, every starting level gained, and the magnitude of the gain was about 2.5 times the cross-customer average. The pattern is the same shape we see across customers. The size is larger.

The customer's lowest-starting quartile gained +1.31 points over the window. The top-quartile starters still gained +0.30 points. The dose-response held at every starting level.
Why this matters

An aggregate gain at one company is interesting. A gain that holds at every starting level is the answer to the sceptical question of whether the result is just regression to the mean. With 160 individuals, the within-company stratified check is statistically meaningful in a way it is not at the very smallest customers, and it comes out positive.

Methodology
Organisation
A Thai professional services firm, 295 employees. Anonymized at the customer's request; referred to as Customer A throughout.
Cohort
160 individuals (54% of the workforce) who accumulated ≥6 ratings on the same skill over ≥180 days, with ≥2 ratings in each half of their window.
Time window
Per-person window spans first to last rating. Mean 324 days, median 337 days (10.8 to 11.1 months). Range 182 to 405 days.
Outcome
Per-cell mean rating in the first half of the window vs the second half, derived from the AI-evaluated quality (rating) dimension of each Power Skill expression in written feedback. 539 person-skill cells.
RTM correction
Within-company stratification: cells split into quartiles by first-half-mean rating; mean delta computed per quartile. If the gain were entirely regression to the mean, the top quartile would not gain at all.
Baseline
The cross-customer cohort: 2,630 individuals at 80 organisations, mean Δ = +0.29 over a comparable window.

The pattern: every skill, the same direction

Six skills, six positive shifts. The smallest gain (Empathy, which started highest at Customer A at 3.63) is +0.37. The largest (Self-awareness, which started lowest at 3.28) is +0.91. The order is exactly what you would expect: skills with more headroom move further.

All six skills moved up at Customer A, in under a year First-half vs second-half mean rating. 160 people, 539 cells, median 11-month window. 1.0 2.0 3.0 4.0 5.0 Rating (1 = low, 5 = high) First half Second half Self-awareness n=90 +0.91 Initiative n=70 +0.85 Critical Thinking n=160 +0.84 Leadership n=53 +0.81 Optimism n=101 +0.59 Empathy n=65 +0.37 Cohort-wide mean: 3.48 to 4.22 (+0.74). 91% of cells positive. 160 individuals (54% of the 295-person company). Source: Happily People Science, May 2026. 539 person-skill cells across 160 individuals. Median span 11 months.
Figure 1 Each pair of dots is one skill's journey at Customer A. Light dot = first-half mean rating, dark dot = second-half. Every pair moves right. The size of the move ranges from +0.37 (Empathy, which started highest) to +0.91 (Self-awareness, which started lowest). Self-awareness, Initiative, Critical Thinking, and Leadership all gained more than +0.80 points in 11 months.

The test that rules out regression to the mean

An average gain across a cohort can come from two places. Some of it is real movement. Some of it is statistical, the artefact that always shows up when you measure the same thing twice and the low starters happen to have more room to rise. The way to separate them is to look at the people who had nowhere mechanical to go. If they gained, the result is not just an artefact.

The customer's top-quartile starters had an average first-half rating of 4.34 on a 0.5 to 5 scale. They had very little upward room. They gained +0.30 points anyway. 85% of cells in that group still moved positively. The dose-response held, at every level.

The gain held at every starting level at Customer A Mean rating change by starting-rating quartile. n=539 cells across 160 people. 0 +0.50 +1.00 +1.40 Mean rating change (Δ) Starting bottom ≤3.08, n=135 +1.31 Starting Q2 3.09–3.50, n=137 +0.79 Starting Q3 3.51–4.00, n=144 +0.56 Starting top >4.00, n=123 +0.30 Why this matters: people who started near the ceiling still gained, which means the result is not just regression to the mean. Source: Happily People Science, May 2026. Within-company stratified analysis. 539 cells, 160 individuals, median 11-month window.
Figure 2 Mean rating change at Customer A, split into quartiles by where each person-skill cell started. The bottom-quartile group had the most room to grow and gained the most (+1.31). The top-quartile group had the least room and gained the least (+0.30). All four groups gained.

How fast Customer A got there

The median analysis window for Customer A is 337 days, slightly under a year. The cross-customer median, for comparison, is 452 days. Customer A's cohort reached its second-half ratings substantially faster than the average customer in the longitudinal study reached theirs, and with a larger magnitude of change. That combination, faster and larger, is what makes this case study a useful reference point for what is achievable on Happily.

Engagement, measured as response-active months during the window, was around the cross-customer median (13 months at Customer A, 14 across the broader cohort). The cohort was not hyper-engaged compared to the broader study. The pace was a property of the organisation, not of an unusual amount of activity.

How Customer A compares

Customer A vs cross-customer baseline (2,630 individuals, 80 organisations)
MetricCustomer ACross-customerDifference
Mean rating change (Δ)+0.74+0.29+0.45
% of measurements positive91%67%+24 pp
% with strong positive change (≥0.5)68%41%+27 pp
Median analysis span (days)337452−115
Top-quartile starters mean Δ+0.30~0.00+0.30

The most informative row is the last one. In the cross-customer analysis, top-quartile starters showed essentially zero gain on average, because most organisations did not have enough engagement or enough time to overcome the ceiling effect. Customer A's top-quartile starters gained +0.30 points anyway. That is the differentiator.

Corroborating evidence at smaller scale

The same pattern shows up at Customer B, a smaller Thai professional services firm (53 employees, 22 in the cohort). At Customer B, every one of the six skills also moved up, with an average gain of +1.00 points over a similar window. Different organisation, different size, same shape: all six skills, the same direction, sustained engagement producing visible movement. Two organisations point the same way, independent observations of the same effect at different scales.

What this case study shows

This is an observational study of one customer, with the cross-customer cohort as comparison. It supports three clear claims:

  1. At Customer A, in this window, every one of the six human skills moved up.
  2. The gain was about 2.5 times the cross-customer average.
  3. The gain held at every starting level, including people who started in the top quartile, which rules out a pure regression-to-mean explanation.

Read it for what it is: one organisation's trajectory, scoped to the 160 employees who used Happily sustainedly. Other customers moved less, and the magnitude here is an example of what is achievable when an organisation leans in, not a number every team will hit. The dose-response pattern is the strongest evidence an observational study can offer that the loop contributes to the movement. The experiment that would measure that contribution directly, a randomised comparison with engagement as the only difference between matched groups, is the natural next step for this line of research.

Limitations

  • Observational design. A randomised comparison would isolate the loop's contribution from everything else changing in a person's work life, and is the experiment that would turn this association into a measured effect.
  • Selection. The 160 in the cohort are the engaged majority of Customer A (54%). The other 135 employees are not in the analysis, and their trajectory is unknown.
  • The AI rating model that produces these ratings evolved over the analysis window. Some of the observed gain may reflect model drift rather than skill change. A future iteration of this study should freeze the model version across the window.
  • The within-company RTM correction is meaningful at n=160 but still narrower than the cross-customer 4×4 stratification. The cross-customer study remains the cleaner test of dose-response after RTM.
  • One time window. Customer A's future trajectory may differ from what this window shows.

References

  1. Happily Research (2026). Human skills rise with practice: a 2,630-person study. Cross-customer longitudinal analysis. The broader cohort context for this case study.
  2. Happily Research (2026). The six skills that show up where engagement does. Framework defense for the six Power Skills.
  3. Happily Research (2026). All six human skills also moved up at a smaller customer. Corroborating case study at a 53-person Thai professional services firm.
  4. Campbell, D. T., & Kenny, D. A. (1999). A Primer on Regression Artifacts. Guilford Press. The standard reference on regression to the mean and how to control for it.
  5. Happily Research (2026). Customer A cohort dataset (anonymized Thai professional services firm). Internal analysis, 160 individuals, 539 person-skill cells, May 2026.
See what 11 months on Happily could look like for your team

The picture in this case study (all six skills moving, the gain holding at every starting level, the speed) is the report Happily can produce for any team with sustained usage. Engagement levels and time-in-window do most of the work.

Get in touch
Free pilot for qualifying teams