Identifying Friction Points

Once the journey is mapped the real work begins: which step on that map is genuinely a problem? Mapping showed you what happens. Friction point analysis shows you where it hurts.

Return to the card unblocking example. The identity verification step appeared on the map as 90 seconds. But that was one call. Look at the records of hundreds of calls and a different picture can emerge: in some the step finishes in 30 seconds, in others it reaches three minutes. That variance is the signature of a friction point. The width of the distribution, not the average duration, is the real signal.

The most reliable source is data: call records, step level durations, first contact resolution rate, handover rate. Where data is missing, structured interviews with experienced customer service agents are the second best source. They know which problem keeps recurring and which step irritates customers, and they can usually predict it before the data is collected.

An important distinction: finding a friction point is not the same as finding something to automate. In the card unblocking example the note against identity verification read "the same information is asked a second time". The answer to that is not a Voice AI Agent. The answer is to find out why the same information is being asked twice and remove the repetition. Some friction points come from a process design error, and all they need is for that step to be simplified rather than accelerated. Making this distinction early keeps the project from producing a perfect solution to the wrong problem.