> For the complete documentation index, see [llms.txt](https://docs.churned.io/churned-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.churned.io/churned-docs/metrics-and-definitions/metrics/health-risk-and-engagement-scores.md).
# Health, risk and engagement scores
The AI-driven scores Churned computes for every active customer, every day
Every active customer is scored daily by models trained on your own historical data. This page defines the scores, their bands and the explanations that come with them. The scores are shown per customer on the customer page and CSM cockpit, and aggregated on the home and technical dashboards.
## AI health score
*An all-in-one measure of the health of your customers.*\
*How healthy your customer base is.*
The Churned AI Health Score is a single number from 1 to 10 that captures how healthy a customer is, today and in the near future. It combines the churn prediction with transactional, profile, support and engagement data.
**Not just rules, real intelligence**
Unlike traditional customer-success tools where a health score is a checklist of rules ("logins < X = unhealthy"), the Churned score is model-driven. Churned analyses the historical behaviour of all your customers, including those who churned, to learn which patterns precede churn: declining usage, missed or failed payments, downgrades, declining response to communication, support contacts. For every organisation multiple model families (gradient boosting, random forests, regularised regression) are trained and evaluated on your data and the best performing one is deployed. The model is retrained periodically and scores every active customer daily.
**Health score and risk score**
The health score is the mirror image of the [risk score](#risk-score): both run from 1 to 10 and always add up to 11. A customer with a risk score of 8.5 has a health score of 2.5. Dashboards use the health score so that a higher number is always better.
**How to interpret it**
The score maps directly onto the three [health levels](#risk-levels) used throughout the dashboard:
* 1.0–4.0: 🩸 *Sick* → **Unhealthy** (high risk). Strong churn signals; immediate action recommended.
* 4.1–7.0: ⚠️ *Concerning* → **Neutral** (medium risk). Warning signs are present; monitor closely and engage proactively.
* 7.1–8.9: 💪 *Healthy* → **Healthy** (low risk). Stable, active customers; maintain the relationship.
* 9.0–10: 🌟 *Excellent* → **Healthy** (low risk). Loyal, high-value customers with strong long-term potential.
***
## Risk score
*The ultimate measure of churn risk for each one of your customers.*\
*Advanced indicator of churn risk leveraging the full power of the Churned AI engine.*
The risk score ranges from 1 to 10. A high risk score indicates a high chance that the customer will churn soon; a low risk score indicates that the customer is unlikely to churn in the near future. It is the mirror image of the health score (they add up to 11). The technical dashboard plots the average risk score of a segment over time.
***
## Risk levels
*A segmentation of your customers into three levels.*\
*How many customers and corresponding value are in each group.*
Each active customer is placed in one of three levels based on their health score. The home dashboard tile is called **Health Levels** and labels them **Healthy**, **Neutral** and **Unhealthy**; older texts and some dashboards use the risk wording **Low Risk**, **Medium Risk** and **High Risk**. Both name the same groups:
| Health level | Risk level | Health score | Risk score |
| ------------ | ----------- | ------------ | ---------- |
| Healthy | Low Risk | 7.1 to 10 | 1.0 to 3.9 |
| Neutral | Medium Risk | 4.1 to 7.0 | 4.0 to 6.9 |
| Unhealthy | High Risk | 1.0 to 4.0 | 7.0 to 10 |
The tile shows, per level, the total ARR and the number of customers together with their change compared to 30 days ago.
{% hint style="info" %}
Since we want as many customers in the Healthy group, an increase there compared to the previous period turns green, while an increase in the Unhealthy group turns red.
{% endhint %}
***
## Risk change
*Keep track of recent changes in the risk score of your customers.*\
*Quick view on how a customer's risk is evolving.*
For every customer Churned stores yesterday's score, the previous level and the average and maximum score of the previous week. The risk change trends upwards when the risk of churn is increasing and downwards when it is decreasing, so you can spot customers who are deteriorating before they become Unhealthy.
***
## Top 3 risk drivers
*Explanation for the churn risk by our AI engine.*\
*The top three drivers of your customer's churn risk.*
Churned goes beyond predicting churn risk: for every customer the model also uncovers the drivers behind the prediction, so you can see why the risk is high or low. The top 3 risk drivers reveal:
* the three main drivers of risk for that customer,
* whether each driver increases or decreases the risk,
* whether the driver's value is low or high for that customer,
* whether the driver is something you can influence (communication, payment method, product usage) or a fixed characteristic (age, region, acquisition channel).
{% hint style="info" %}
Drivers that help reduce churn risk are always shown in **green**. These are "good" drivers.\
Drivers that increase churn risk are always shown in **red**. These are "negative" drivers.
{% endhint %}
{% hint style="info" %}
**Example**
Consider a customer with *high churn risk* whose top 3 risk drivers are:
1. **Low number of logins**
2. **High number of products**
3. **Low NPS score**
The main driver behind the high-risk prediction is the *Low number of logins*, coloured red because it contributes to a higher churn risk, just like the *Low NPS score*. The *High number of products* appears in green since it attenuates the churn risk.
{% endhint %}
***
## Risk change drivers
*Explanation for recent changes in the risk score.*\
*Which recent changes in behaviour moved the customer's risk.*
The predicted risk fluctuates over time. The risk change drivers list the features whose recent change explains that movement. A customer whose risk was kept low by **High number of logins** will show **Decreasing number of logins** as a risk change driver as soon as usage drops.
***
## Health breakdown
*How the health score is composed.*\
*Which part of the relationship is weak or strong.*
Every health score is split into a small number of health breakdown groups, each with its own sub-score (1 to 10) and its change since the previous day. Which groups exist depends on the data you connect; typical groups are **Product engagement**, **CSM engagement**, **Support**, **Payments** and **Customer profile**. A customer can be healthy on Payments but weak on Product engagement, which points directly to the type of action that will help. The same grouping is used on the Model Insights page to group feature importances.
***
## Engagement score
*An all-in-one indicator of the engagement level of your customer.*\
*How engaged is your customer: a unique indicator produced by our AI engine.*
While the health score covers every dimension, the engagement score isolates one: behavioural engagement. Customers rarely churn suddenly; they disengage first. The engagement score is produced by a dedicated model that uses only engagement-related features (logins, product usage, email opens and clicks, campaign responses, support contacts) to estimate churn likelihood. This makes it a pure measure of engagement-driven risk.
{% hint style="warning" %}
The engagement score requires an engagement data source (product usage, email statistics, campaign responses or web behaviour) to be connected. Without one the score is not available and the engaged / non-engaged metrics are hidden.
{% endhint %}
The score ranges from 1 to 10:
* 1–4: 💤 *Disengaged* → high churn risk due to low or declining engagement.
* 5–6: ⚠️ *At risk* → engagement is inconsistent or trending downward.
* 7–8: 💬 *Engaged* → healthy interaction levels across touchpoints.
* 9–10: 🔥 *Highly engaged* → consistently active, responsive and loyal.
***
## Next best action
*The best way of preventing churn and retaining your customer.*\
*See which action you should take next.*
When the Co-pilot is enabled, Churned measures the impact of each retention action on comparable customers and ranks the actions most likely to prevent churn for each individual customer. The list takes into account which actions were taken recently, so you will not be advised to repeat the same action for a while. See [Co-pilot](/churned-docs/co-pilot.md) for setup and use cases.