Here is something the sales team at most B2B SaaS companies knows but rarely acts on systematically: your competitors' worst customers are publicly broadcasting their dissatisfaction on G2, Trustpilot, and Capterra. They are naming specific problems. They are rating those problems at one or two stars. In many cases they are explicitly naming the tool they are switching to.
That is not noise. That is a qualified lead list.
The challenge is extracting signal from the raw review data at scale. An individual reading fifty one-star reviews of a competitor can identify patterns — "this is the third person who mentioned the onboarding process was unusable" — but doing that for five competitors, across three review platforms, on a weekly basis, is the kind of analytical task that either gets done poorly or does not get done at all. The information exists. The bandwidth to process it does not.
I became interested in this problem while building competitive positioning for a client who sold project management software. Their main competitor had 847 reviews on G2. My client had looked at the most recent page and drawn some conclusions. What they had not done was read all 847 reviews, classify each complaint by the feature it related to, aggregate those classifications to understand which product areas drove the most dissatisfaction, and cross-reference that against the competitor's current feature roadmap announcements to understand where they were and were not investing.
That complete analysis, done manually, would have taken several days. It would have needed repeating every quarter as new reviews accumulated.
The Competitor Review Intelligence actor adds the intelligence layer that raw review scrapers lack.
It collects reviews from G2, Trustpilot, and Capterra for whatever products you specify. Then it passes each review through GPT-4o-mini with a carefully designed aspect classification prompt. The model identifies which specific product dimension each review is commenting on — pricing, UX, customer support, integrations, performance, onboarding, reliability, features — classifies the sentiment for that specific aspect, and extracts the exact quote from the review that supports the classification.
The output is not "this was a negative review." It is: "primary complaint: pricing. Aspect: pricing. Sentiment: negative. Quote: 'The price doubled overnight with no warning and support was completely unresponsive.'"
Two additional signals emerge from this classification. The first is churn intent detection. Reviews that contain phrases like "switching to", "moving to", "cancelled because", "replaced with" are flagged automatically, along with any competitor names mentioned. Someone who publicly says "we cancelled our subscription and moved to [your product]" is, empirically, the warmest possible lead for that competitor. They have already made the decision. They have already written the farewell note. A well-timed, personalised outreach citing the specific problem they named in their review has a response rate that is qualitatively different from cold outbound.
The second signal is feature gap intelligence. When the same product area — "integrations", "API", "reporting" — accumulates consistently negative reviews across a large sample, that represents a genuine product weakness. For a competing product that has invested in that area, the aggregated review intelligence is a ready-made sales objection library and positioning argument. "We have heard from a lot of companies coming from [Competitor] frustrated by the limitations of their API. Here is what our API actually does."
Sentinest and SentiSum charge £300 to £1,500 per month for this kind of review intelligence, marketed primarily to enterprise product and CX teams. The Competitor Review Intelligence actor delivers functionally equivalent output at £0.015 per classified review, with the scraping and classification combined in a single actor run.
A G2 analysis of 200 reviews across three competitors, with full aspect classification and churn signal detection, costs approximately £3.00.
The insight that prompted this build was simple: the data was always there. The cost of extracting it intelligently was too high for the companies that most needed it. This actor makes the economics work.