Disclosure: ApiVault Labs publishes the Actor used here. This guide is AI-assisted. Configuration is illustrative; no live collection or measured customer findings are claimed.
A support team can learn from a small set of public complaints without pretending it has measured a company's customer satisfaction. The useful deliverable is a reproducible question for investigation: which step confused reviewers, what evidence supports that reading, and what internal data could confirm it?
Write a coding rule before reading the sample
Choose one company, one publication window and a small collection budget. For an initial complaint-theme pass, the current Trustpilot Reviews Actor accepts:
{ "companyUrls": ["wise.com"], "maxReviews": 10, "maxPages": 2, "sortBy": "recency", "filterStars": [1, 2], "filterDateRange": "last30days", "maxConcurrency": 1, "maxRetries": 0, "includeRaw": false, "extractReviewer": false, "extractReviewerLocation": false, "extractReviewerTotalReviews": false } This is an example about a public company page, not a finding about that company. maxReviews and maxPages apply per company. Narrow filters can leave fewer matches than the requested maximum. Public source access can fail; inspect SUMMARY and ERRORS before analysis.
Before reading, define a short coding sheet: review ID, publication date, stated problem, quoted evidence, theme, and ambiguity. A synthetic rule might assign “status explanation” only when text describes uncertainty about a process stage. “Service is bad” alone would remain unclassified. Allow multiple themes per review and retain an explicit unclassified group; theme totals can exceed the number of reviews.
Keep the sampling funnel beside the themes
Record requested companies, filters, collection time, delivered rows, per-company status and stop reasons. Deduplicate by reviewId when combining exports. Do not substitute company-wide review totals for the number actually coded.
Synthetic arithmetic: suppose eight unique delivered reviews include three about status explanations, two about fees, one with both themes and four with neither. The first two theme groups overlap. Report each theme against eight reviewed records, explain overlap, and keep the underlying IDs. This arithmetic describes a selected low-star sample; it says nothing about the share of all customers experiencing those issues.
The company profile score in COMPANIES describes a different population. Publication date, experience date and collection time also answer different questions. Preserve them separately rather than treating a recently published account as proof of a recent incident.
Use the finding to choose a follow-up
For each theme, write one operational check. “Compare the quoted status wording with the current help page” is testable. “Company support is ineffective” exceeds the evidence. Reviews contain source statements, not independently verified transactions or identity checks.
Keep reviewer information out of the working sheet when unnecessary. The field switches above reduce normalized reviewer fields; leaving includeRaw:false avoids separately adding the original review object. Review text can still contain personal details, so restrict access to the export.
Current pricing checked for this Actor: $0.0005 per delivered review, plus the $0.00005 start event at standard memory; Apify platform usage is additional. Ten delivered reviews mean $0.00505 in event fees before usage, not a guaranteed count or total run cost. Empty or failed collection can still incur start and resource costs.
Use the exact Trustpilot Reviews Scraper schema. A similarly named Actor can have a different input contract. Save the input, storage IDs and diagnostics with the coding sheet so another analyst can understand what the sample covers.

