How To Gauge Customer Churn Risks Using Traditional vs. AI-Based Metrics

For CX and Operations leaders, understanding customer churn risk is essential to proactively address issues before customers leave. Your contact center is on the front lines of customer interactions, so it logs many leading indicators of churn and has levers to influence retention through service quality.

Knowing which metrics to track (and how to track them) allows you to get a better understanding of your customer churn risk and even identify which customers are about to leave. 

3 Key Metrics To Determine Your Churn Risk

Traditionally, contact center and operations managers had to rely on simple post-call surveys to assess customer happiness and deduce customer churn risk from there. But there are several problems with that. First, almost no one will take a post-call survey except those who are extremely angry or happy with their experience. So, you are getting a submission bias. But you are also only getting a score, not the context behind the score. This might give you an indication (which is better than nothing), but it certainly doesn't tell you the whole story.

However, thanks to advancements in Generative AI, businesses can now track voice-of-customer metrics across all calls. These metrics are generally much more accurate, as AI considers many facets, such as customer sentiment during the call, the context of the entire conversation, and what the customer said. This gives the AI a better ability to determine the metrics needed to manage your contact center's performance.

Here are the specific customer experience and behavior metrics you should track to determine your customer churn risk:

• Customer Satisfaction (CSAT): Indicator of immediate dissatisfaction or satisfaction.
• Net Promoter Score (NPS): Indicator of longer-term loyalty trends.

• Customer Effort Score (CES) or Net Effort Score (NES): Early warning signal of emotional disengagement.

Let's go through those metrics one by one to see how you can measure, track, and manage them in a traditional versus an AI-based manner before we discuss other important indicators you might want to consider monitoring as well.

1. Customer Satisfaction (CSAT)

Traditional CSAT Score

The CSAT score measures customers' satisfaction with a specific interaction or overall experience. The score is traditionally obtained by asking customers to rate how satisfied they were with their experience in a post-call survey. The survey scale can be a number scale (1–3, 1–5, or 1–10) or a percentage scale (0% = very bad, 100% = fantastic).

AI-Enhanced CSAT Score

However, with AI determining the score, you can get a much more nuanced picture. As you can see in the screenshot below, MiaRec allows you to analyze different factors:

• Issue Resolution: Was the customer's problem resolved?
• Agent Performance: Empathy, clarity, professionalism.
• Customer Engagement: Level and progression of frustration.
• Efficiency: Length of interaction, hold times, escalations.
• Extra Mile: Additional proactive measures taken by agent.

These factors can be fully customized and assigned different weights according to your needs. MiaRec will then score each customer interaction and provide a quick summary, calculation, justification, and recommendations to improve the score.

2. Net Promoter Score (NPS)

Traditional Net Promoter Score (NPS)

The Net Promoter Score (NPS) measures customer loyalty by asking customers how likely they are to recommend your business to others, typically scored from 0 (unlikely) to 10 (very likely). It is often hailed as a clear, simple metric indicating overall customer sentiment and potential churn risk. Customers are categorized as:

• Promoters (9-10): Highly loyal and enthusiastic.
• Passives (7-8): Satisfied but not enthusiastic.
• Detractors (0-6): Unhappy and potential churn risks.

Low NPS scores after interactions are a clear warning sign. For example, in healthcare, a patient repeatedly dissatisfied with call support (e.g. scheduling or billing inquiries) is more likely to switch providers. In banking, declining satisfaction with service correlates with eventual attrition. However, these scores are usually obtained from post-call surveys or feedback forms and suffer from submission bias and low response rates.

AI-Based Net Promoter Score (NPS)

Just as with the CSAT, MiaRec can provide you with an explanation about the key factors that were considered for the score, as well as a justification and suggestions for improvements.

3. Customer Effort Score (CES) or Net Effort Score (NES)

Traditional Customer Effort Score (CES)

Customer effort is one of the strongest predictors of loyalty. If customers have to put in a lot of effort to resolve their issues, they tend to become frustrated, disengage, and leave. Research shows that 96% of customers who experience high-effort service report being disloyal (likely to churn) compared to only 9% who have low-effort experiences.

Metrics that indicate effort include the number of times a customer must contact the company to resolve an issue, transfers between agents, or long hold times. A contact center can measure this via CES surveys or by proxy (e.g. count of contacts per issue). A spike in effort required for a particular customer is a red flag.

AI-Enhanced Customer Effort Score (NES/CES)

You can also determine a NES or CES score using advanced Generative AI. As you can see in the screenshot below, you can see that MiaRec determines the score by considering multiple factors:

• Initial Issue Clarity: Did the agent quickly understand and respond clearly?
• Repetition: How often did the customer repeat their issue?
• Time Taken: Hold times, delays in resolution.
• Transfers/Escalation: Customer-initiated requests to escalate.
• Agent Responsiveness: Clarity and effectiveness of agent responses.
• Resolution Status: Was the issue fully resolved?
• Customer Tone: Emotional journey, frustration, confusion.

This is more accurate than traditional methods such as post call surveys and it is done automatically for every single relevant call.

NEW: MiaRec's Customer Churn Risk Dashboard

Alternatively to tracking individual metrics to gauge customer churn risk, you could use MiaRec's brand-new AI-generated "Churn Risk" dashboard. MiaRec automatically assigns churn risk to every interaction based on the context of the conversation. This means that the "Churn Risk" is entirely independent of the other metrics mentioned above.

Image: Screenshot of MiaRec's CX KPIs, including the Churn Risk, on a single call.

Image: Screenshot of MiaRec's Churn Risk Dashboard.

In the MiaRec Dashboard, you can see how many calls have been evaluated for customer churn indications, how many of those that already churned are at high risk, etc. In addition to the numeric values, the dashboard gives you a visualization of the churn risk over time and a correlation of the call reasons (topics) and their associated churn risks.

Other Metrics to Consider When Analyzing Your Customer Churn Risk

In addition to the three metrics above, you can add more context and insights into your churn analysis to get an even more comprehensive picture. Below are other metrics you should consider:

First Contact Resolution (FCR)

The First Call Resolution (FCR) metric measures how many customer inquiries were resolved in the first call. This is usually expressed in a percentage. A customer record would show a low FCR if the customer had to call back multiple times for the same issue, signaling unresolved problems and frustration – a precursor to churn. For instance, if a policyholder has to call three times about a billing error, their churn risk rises with each repeat call.

This is usually tracked in call logs by issue ID. At the moment, this is difficult to track with Generative AI because the AI considers the context of that particular call. However, it is possible to generate an AI Insight that looks for indications that this is a first call or whether the customer had previous calls. The AI can then determine whether or not the call was resolved in the first call based on the context of the conversation. You can then run reports over all customers to determine the percentage of calls resolved.

Topic Analysis (Call Disposition Codes)

Another way to gain valuable insights is by looking at the number of times a customer has called in connection with what topics they called about. Generally speaking, customers who call frequently with problems are at higher churn risk. A churn risk model should factor in how often a customer contacts support and what they contact about. Certain contact reasons are especially predictive – e.g. a financial services study found that customers calling about a charge dispute had a much higher likelihood of churning soon after. Likewise, in healthcare, if a patient calls multiple times about insurance coverage issues or waits, they may be ready to leave.

Key metrics include: number of calls or chats from the customer in last 30/60 days; escalation rate (how often they ask for a supervisor); and whether they’ve explicitly mentioned cancellation. Call disposition codes (reason categories) are data points to monitor – for example, “requested account closure” or “complained about service” should trigger immediate retention actions.

Customer Sentiment

Modern contact centers gauge sentiment using speech analytics or conversation intelligence. Negative sentiment scores, angry tone, or language like “I’m very disappointed” or threats to leave are qualitative signals to quantify. If call transcripts show a customer using words that indicate frustration or intent to leave, that’s a leading indicator. For example, conversation analytics might flag that customers making strong “remedy demands” (like requesting refunds or guarantees) or who get an apology from the agent multiple times tend to churn​. These text analytics can be turned into churn risk predictors.

Account/Service Changes

While not strictly a contact center metric, agents often have visibility into account changes during interactions. Signs such as a customer downgrading their plan, reducing usage (which might be mentioned on the call), or asking about how to transfer their account can indicate churn intent. An agent noting, “the customer asked about the closing account process” is a red flag in financial services. The contact center should capture these signals in CRM notes, which feed the risk calculation.

Want to learn more about how to calculate your customer churn risk, join us for a live webinar on March 27th, 2025 where we will go through all the metrics and how to use them in detail Click below to save your spot.