A customer review is written by someone who experienced the result, not the process behind it. Yet that result often carries a trace of how it came about. A review mentioning that a quote arrived "within an hour" says something different about the organisation behind it than a review speaking of "finally, a response after two weeks". Both are feedback about speed. Only one of the two points to work that may no longer be done by a human alone.
When AI takes over part of the work behind a service, it is not the service itself that changes but the pattern in which customers experience it. Response time to a question was long a matter of who happened to be at the desk that day. Where a system handles the first triage and an employee only assesses and decides, that dependency on staffing disappears. This changes not only speed but also consistency: a customer who responds late in the evening gets the same treatment as a customer on Monday morning. Reviews that speak of even service at unusual hours point more to that shift than reviews about occasional speed.
That does not apply to every company in the same market, and that is exactly the point. One company has a system perform the first assessment with an employee approving or rejecting with reasons; another still does this entirely by hand, and a third has not touched the process at all because customer contact there deliberately remains personal. The difference lies not in ambition but in where the task sits: fully automatable, partly under supervision, or unchanged human work. Reviews show which side of that three-way split a competitor is on, provided you know what to look for.
Three signals are usable without assuming too much.
First: response time that does not vary with time of day or busyness. A human process has peaks and troughs; an automated first step usually does not. Complaints about slow responses on Monday but not on the weekend point to human work. Consistent speed, even outside office hours, points to a system taking over part of the work.
Second: the tone and specificity of answers. Reviews reporting that a response was "very detailed and straight to the point" may point to a system that has already compiled the data before an employee replies. Reviews complaining about generic, not-very-relevant answers point more to a process without that preprocessing, or to a system handling too much autonomously without supervision to correct it.
Third: how errors are handled. A complaint that is picked up quickly with a personal, reasoned response points to human oversight that approves or rejects with reasons — precisely the second category in the three-way split. A complaint that is handled neatly but generically, without anyone appearing to genuinely look into it, points more to a process that has been fully handed over without a correction layer.
The pitfall is reading quality as proof of automation, or the reverse. A slow answer can be a deliberate choice for personal contact, not a lack of technology. A fast answer can be a small, dedicated team without any AI use at all. Reviews give an indication of outcome, not of cause. They are also selective: satisfied customers generally write less than dissatisfied customers, which makes the picture structurally more negative than reality. And reviews say nothing about what a competitor itself claims to do; for that it is more useful to look at how a competitor's own sales people talk about their way of working or at what job listings give away about the role AI plays in a team.
Whether a company uses AI to lighten the workload is, moreover, a different question from what the company does with that freed-up capacity — the latter falls under its own statutory requirements for personnel policy and is not a conclusion that can be drawn from reviews.
Reviews are worth it as a supplementary trace, not as a primary source. They are useful when you set them alongside other signals: what lost quotes reveal about where a competitor is faster or slower than you are, and what a competitor's own website claims to do on speed and service. Reviews confirm or contradict those claims from the customer's perspective. Read in isolation, without that other trace, a review is mainly an opinion about a single experience, and too narrow to build a judgement on about the AI use of an entire organisation. Those wanting to search structurally without figures from the competitor itself will find a broader approach in the method for obtaining information about competitors who do not publish their own results.
All reviews combined tell you something about competitors, but the question that yields the most concerns your own organisation: which work here can genuinely be taken over by AI, which work still requires oversight, and which work remains human work. That is mapped per task with the work scan from FTE TO AI.
Would you like to know whether your own assumptions about speed and service hold up? Name the dimensions on which you believe you are winning and take the free dimension check: you will immediately see which of those claims can be defended with evidence and which remain, for now, an assumption. The full benchmark, with the evidence matrix and the comparison against your peer group, is under construction.