The problem
Customer activity could be observed through many individual measures, but those signals did not provide a single view of overall engagement or show clearly whether a customer's relationship with the product was strengthening or declining. Product and Marketing needed a more useful way to understand customer health, identify changing behavior, and determine where outreach, education, or product action might be valuable.
My role
Conceived and designed as an individual-contributor analytical initiative spanning the engagement concept, metric framework, scoring approach, customer-level data model, trend analysis, dashboard experience, and business-use scenarios.
The approach
- Combine more than 40 customer-level behavioral, product-usage, device, and service signals into a structured engagement model.
- Adapt the conceptual approach of a multi-factor sports rating into a normalized customer-health score while preserving visibility into the underlying measures.
- Track engagement longitudinally at the individual-account level so changes over time could be evaluated alongside the current score.
- Design the model to support actionable Product and Marketing questions rather than producing a score for its own sake.
- Provide drill-down visibility into the behaviors contributing to engagement so users could understand why a customer's score was changing.
- Explore how differences between strongly engaged and declining customers could inform retention, education, promotional, and product strategies.
Technology & capabilities
- Tableau analytical dashboards and visual storytelling
- Customer-level analytical data structures
- Longitudinal trend and score history
- Multi-factor metric normalization and scoring
- Behavioral and product-usage analytics
Skills demonstrated
- Customer engagement modeling
- Behavioral analytics
- Metric design
- Composite scoring
- Trend analysis
- Customer segmentation
- Tableau development
- Dashboard design
- Data visualization
- Analytical storytelling
- Product analytics
- Marketing analytics
- Business problem framing
- Customer-level data modeling
- Translating analytics into action
- Scoring model design
- Normalization & calibration
Outcomes
- Created a unified engagement concept from more than 40 otherwise separate customer signals, providing a more holistic view of customer health.
- Added month-over-month account-level trending so users could distinguish current engagement from changes in customer behavior over time.
- Demonstrated how customer behavior could be translated into actionable Product and Marketing segments, including customers showing declining use or underutilization.
- Created an analytical framework for examining which behaviors differentiated more strongly engaged customers from those showing signs of disengagement.
- Produced a prototype that demonstrated the potential of customer-health scoring even though the scorecard itself was not ultimately moved into production.
From scattered signals to customer health
The starting point was not a lack of data. It was the opposite.
Customer activity could be observed through many individual measures: product usage, device activity, service behavior, automation usage, equipment status, and other signals. Each measure said something useful, but no single metric answered the larger question:
How engaged is this customer—and is that engagement getting stronger or weaker?
I wanted to create a more holistic way to describe customer health while still preserving the underlying behaviors that made the score meaningful.
Adapting the NFL passer-rating model
The scoring concept was inspired by the NFL passer-rating formula.
Passer rating does not evaluate a quarterback from a single statistic. It combines four different dimensions of performance—completion rate, yards per attempt, touchdowns per attempt, and interceptions as a negative factor—into a common scoring framework.
I saw a useful parallel in customer engagement.
The Homelife Engagement Score used four corresponding dimensions:
- Login Days — direct interaction with the application.
- Active Activity — customer-initiated actions such as creating rules, changing scenes, or arming the security system.
- Passive Activity — ongoing value produced by the system, such as scheduled rule execution and notifications.
- Offline Days — a negative factor representing periods when the system was unavailable and the condition remained unresolved.
The intent was not to reproduce football mathematics literally. It was to reuse an important idea behind the rating system: different measures with different meanings can be translated onto a common scale and combined into a single indicator while still retaining the component measures underneath it.
Defining what “average” means
The scoring curves were deliberately calibrated around observed customer behavior rather than arbitrary percentages.
An expected or typical level of usage was assigned a component score of approximately 66.67, paralleling the baseline construction of the passer-rating formula. Stronger-than-expected behavior moved higher on the curve, with exceptional usage around 133.33 and an upper bound of approximately 158.3.
The shape of the curve depended on the behavior being measured. Login frequency could be represented reasonably with a linear relationship, while skill usage required a polynomial curve to better reflect how increasing activity related to meaningful engagement.
The important point was not that a customer should achieve 100—or any other particular number. The score established a consistent scale against which behavior could be compared over time.
The direction mattered more than the number
The most useful information in HES was not the absolute score. It was the movement.
A customer could have a bad month just as a quarterback can have a bad game. One period of lower activity did not necessarily mean that the relationship was failing.
What mattered was trajectory.
Because HES was retained at the individual-account level month over month, the analysis could distinguish between customers whose engagement was stable, improving, gradually declining, or changing sharply.
The baseline itself could also evolve. The same thing happened historically in football: performance that once represented an average passer rating became less typical as the game and expectations changed. Customer behavior could evolve in the same way as products, features, and patterns of use changed.
HES therefore was intended as a comparative analytical signal—not an immutable judgment of whether a customer was “good” or “bad.”
More than forty signals, one customer view
Each real-world customer account carried a collection of behavioral and operational measures that could contribute to the engagement picture.
The model brought those measures together at the individual-account level, creating a single analytical view while retaining the ability to examine the components underneath it.
That distinction mattered. A composite score can simplify complexity, but it should not hide it.
Users needed to be able to move from:
“This customer’s engagement is declining.”
to:
“What changed?”
The underlying measures provided that explanation.
Trend mattered as much as the score
A customer’s current score tells only part of the story.
An account with moderate engagement that had been stable for months could represent a very different situation from an account with a higher score that had recently declined sharply.
For that reason, the model preserved month-over-month account-level history and made trajectory a central part of the analysis.
This allowed Product and Marketing teams to think about engagement as a changing relationship rather than a static classification.
Designed for action
The scorecard was intended to support decisions, not simply describe behavior.
For Marketing, declining engagement could identify customers who might benefit from targeted outreach, education, or a promotional offer.
Underutilized equipment provided another example. If a customer owned a capability but rarely used it, the more interesting question was not simply that usage was low. It was:
How are more engaged customers using the same capability differently?
That comparison could inform both customer communication and product strategy.
For Product teams, the model offered a way to examine which behaviors tended to accompany stronger engagement and which patterns might indicate friction, confusion, or declining value.
Connecting behavior with business questions
One of the most useful aspects of the concept was that the engagement score was not meant to exist in isolation from the business.
The model provided a framework for asking questions such as:
- Which customers are becoming less engaged?
- Which behaviors changed before the decline became visible?
- Which products or features appear underused?
- How do strongly engaged customers behave differently?
- Which accounts may benefit from education, outreach, or a targeted offer?
- Are engagement patterns improving or deteriorating across the customer base?
The score was therefore a starting point for analysis rather than an endpoint.
Analytical storytelling through Tableau
The scorecard experience was designed to make a complex customer-health model understandable to business users.
Tableau provided the visual layer for exploring engagement levels, underlying indicators, customer segments, and changes over time.
The design emphasized the relationship between the high-level signal and the detailed behavior beneath it: users could begin with the engagement picture and progressively move into the measures that explained it.
Valuable even as a prototype
The Homelife Engagement Scorecard was not ultimately adopted as a production product.
That outcome does not diminish what the work demonstrated.
The project showed how a large collection of behavioral and operational measures could be transformed into a coherent customer-health model, tracked longitudinally, and tied to concrete Product and Marketing decisions.
It also reinforced a principle that has shaped much of my analytical work:
The most useful metric is rarely the one that simply reports what happened. It is the one that helps people understand what the information means and what they might do next.