Understanding how a call went shouldn't depend on whether the customer filled in a survey. Predicted CSAT gives you a satisfaction signal on every eligible call, automatically, using the transcript and call data. No surveys. No manual review. No blind spots.
This article explains how Predicted CSAT works, how to set it up, and how to get the most out of it.
Before you start
| Requirement | Details |
|---|---|
| License | AI Assist Pro (required per agent for score generation) |
| Role | Admin to enable and configure; Admins and Managers to view scores and analytics |
| Location | Call page, Conversation Center, Agent Workspace, General Performance widget |
What Predicted CSAT tells you
After each call ends, Aircall's AI reads the transcript and generates a satisfaction score from 0 to 100. The score reflects how the call likely felt from the customer's perspective: not just emotional tone, but whether the issue was resolved, how much effort the customer had to make, and how the agent handled the interaction.
Every score comes with two driver explanations grounded in specific transcript moments, for example: "Issue resolved and customer expressed thanks," or "Customer had to repeat their issue multiple times." These drivers tell you not just what the score was, but why.
Note: Predicted CSAT is not the same as Mood (sentiment analysis). Mood captures emotional tone during the call. Predicted CSAT is a broader judgment about the likely outcome of the experience. A customer can sound frustrated throughout a call but leave satisfied because the agent solved their problem. Both signals are shown together where available.
What this means in practice:
- Spot low-satisfaction calls without waiting for survey responses, so you can follow up or coach faster.
- Understand why a call scored the way it did, not just that it did.
- Filter calls by score range in Conversation Center to focus your team's attention where it matters most.
How the score is calculated
The AI evaluates six signals from the transcript and call metadata. It does not listen to or analyse the audio recording. Tone of voice is not a factor — everything is derived from what was said.
| Signal | What it looks at | Why it matters for your team |
|---|---|---|
| Resolution | Was the issue solved? Did the agent confirm everything was addressed? | The single strongest indicator of whether a customer leaves satisfied. |
| Customer sentiment | How did the customer feel at the end of the call, and did that improve or worsen during the conversation? | A declining sentiment arc often signals an interaction that went off track. |
| Customer effort | Did the customer have to repeat themselves or get transferred multiple times? | High effort is one of the most reliable predictors of churn. |
| Empathy | Did the agent show active listening and understanding? | Customers who feel heard are more forgiving of imperfect outcomes. |
| Product knowledge | Was the agent able to answer questions accurately? | Uncertainty or incorrect information erodes confidence quickly. |
| Closing etiquette | Did the agent wrap up professionally and check for remaining questions? | A strong close can lift overall satisfaction even when earlier parts of the call were difficult. |
Resolution and customer sentiment carry the most weight, together accounting for 70% of the score. The final score is an integer from 0 to 100.
How to enable Predicted CSAT
Predicted CSAT is designed for support and success calls. The signals it measures — resolution, customer effort, closing satisfaction — are meaningful in those contexts. They produce less reliable scores on sales calls, where the conversation dynamic is fundamentally different. We recommend enabling it selectively by number rather than switching it on account-wide.
Step 1: enable at company level
Steps
- Go to AI Assist Pro > Settings.
- Find the Predicted CSAT section.
- Toggle it on.
This enables score generation for all calls where the agent has an AI Assist Pro license. The number-level settings in Step 2 let you refine exactly where scores are generated.
Step 2: configure per phone number
Steps
- Go to Numbers and select a number.
- Open the number's Settings.
- Find Predicted CSAT and set it to Enabled or Disabled.
Important: The number-level setting always overrides the company-level default. Set sales numbers to Disabled to keep your analytics focused on the calls where satisfaction measurement is most meaningful.
Where to find scores and what to do with them
Reviewing a specific call
On the call page, the score appears alongside the Mood indicator, with the two driver explanations and linked transcript snippets showing exactly which moments influenced the result. Use this view when reviewing a call flagged for follow-up, or when coaching an agent on a specific interaction.
Finding low-satisfaction calls at scale
In Conversation Center, filter calls by CSAT score range. For example, set a filter for scores below 40 to surface calls that may need recovery contact or coaching prioritisation. You can combine this filter with agent, team, or date filters to narrow your focus further.
Giving agents visibility into their own performance
In Agent Workspace, agents can see the Predicted CSAT score for their own calls in the call details panel, alongside the call summary. This gives agents a self-serve signal on how their calls are landing without needing a manager to flag it.
When a call does not receive a score
Some calls are not eligible for scoring. If a score cannot be generated, a placeholder message on the call page explains why. Calls are not scored when:
- The call lasted fewer than 30 seconds.
- The call had fewer than 3 back-and-forth exchanges, such as a wrong number or a very quick transfer.
- The call was a voicemail.
- The transcript was unavailable.
- The agent does not have an AI Assist Pro license.
FAQ
Does Predicted CSAT replace Mood (sentiment analysis)?
No, and the two signals are most useful together. Mood shows how the emotional tone shifted during the call. Predicted CSAT tells you how the interaction likely landed overall. A call can show negative Mood but a high CSAT score if the agent resolved the issue well despite a difficult conversation, and vice versa.
Does it replace Call Scoring?
No. Call Scoring measures whether the agent followed your team's defined process. Predicted CSAT measures how the customer likely experienced the outcome. They answer different questions: one is about process adherence, the other is about the customer experience. Both are worth tracking.
Can I adjust the scoring formula or weights?
Not in the current version. The algorithm uses fixed inputs and weights, which keeps scores consistent and comparable across agents, numbers, and time periods. Configurability is not available in v1.
Will historical calls receive a score?
No. Scores are generated for new calls only, from the moment Predicted CSAT is enabled on a number. Historical calls are not backfilled.
What happens with transferred calls?
The score covers the full call. The driver explanations highlight the moments most relevant to the overall customer experience, regardless of which agent handled that part of the conversation.
Do agents see their own scores?
Yes. Agents can see scores for their own calls in Agent Workspace. Managers and Admins can view scores across all calls in Conversation Center and on the call page.
How do I know if a score is accurate?
You can rate any score using the thumbs up or thumbs down option on the call page. This feedback is used to improve accuracy over time. Treat individual scores as directional signals rather than precise measurements, and use trends and patterns across multiple calls to draw conclusions.
Aircall Status
Help logging in