A customer effort score (CES) is a survey measure of how easy or hard a customer found it to get something done with a firm, usually asked straight after an interaction on a numbered scale. It captures friction. With AI agents in the path, friction is often where things go wrong.
Why does customer effort matter once agents act?
AI agents can remove effort: no queue, no hold music, an answer at midnight. They can also add it: repeating information the agent should have kept, being asked to rephrase, going round in a loop, or being told to call a number to do the thing the agent could not. Customers notice effort sharply, and high effort predicts that they will call back, complain or leave.
CES is a direct way to hear about that friction. It is especially useful when an agent appears to perform well on internal measures, such as speed or containment, while customers find the experience laborious.
How does CES work in practice?
The firm asks one question after a contact, sometimes with a free-text follow-up. Common forms ask the customer to agree or disagree that "the company made it easy to handle my issue", on a scale from strongly disagree to strongly agree. Results are reported as an average or as the share of customers at the easy end.
To make it useful for agents, tag every response with what happened in the conversation: the intent, whether the agent handled it alone, whether it handed over, and how many turns it took. Then a low score points to a specific cause. Raw averages hide that, as with any score of an AI agent's answers.
An example from an insurer's claims line
Take a motor insurer whose claims agent takes first notice of loss. Satisfaction scores are reasonable, but CES for claims with third parties involved is poor. Reading the free text, the team finds customers are asked for the other driver's details by the agent, then asked again by the claims handler after transfer.
Nothing was wrong with any single answer. The effort came from the handover: the agent's notes were not passed with the transfer. Fixing the handover so the handler sees what the agent collected improves CES for those claims without changing what the agent says.
How is CES different from CSAT?
Customer satisfaction score asks how pleased the customer was. CES asks how hard they had to work. The two often agree, but not always. A customer can be satisfied with a polite, friendly agent that took four attempts to understand them, then rate effort as high. Equally, a customer can find a refusal easy to receive, because it was clear and fast, while being dissatisfied with the outcome.
For AI agents, effort is often the more diagnostic of the two, because it points at specific steps: repetition, loops and handovers.
Which terms sit next to CES?
- Net promoter score: a survey of whether customers would recommend the firm, a relationship measure rather than an interaction one.
- Repeat contact rate: a behavioural signal of effort, measured without asking.
- Survey scores are not certification evidence; an eval score is not a bound on what an agent may do, and neither is CES.
- Our guide to AI support agents that know when to say no covers the handover design that affects effort most.
Questions people ask
Does a clear refusal raise customer effort?
Usually not. A clear refusal with a reason and a next step is low effort. A vague answer that leaves the customer to work out what to do next is high effort, even if it sounds friendlier.
Who should see CES results for an agent?
The process owner and the team that maintains the agent's brief. Low scores by intent are a signal to look at the brief, the sources or the handover, and changes there go through recertification.