How IndustryLens Collects, Verifies, and Scores Competitive Intelligence

Win/loss signals, feature sentiment, switching intent, named customer evidence.

Headcount direction, new product bets, engineering priorities — 4–6 weeks before any press release.

Funding rounds, partnerships, executive changes, and product announcements — with timestamps and verbatim quotes.

Organic positioning shifts, community sentiment, launch traction, and feature complaints straight from users.

Live campaign themes, ICP targeting, messaging pivots, and budget signals — all from public ad registries.

Direct-page scrapes with week-over-week diffing so every pricing change and new tier gets flagged, not inferred.

Category narrative, analyst positioning, and funding context that shapes how buyers perceive competing vendors.

Primary-source URLs (official pricing page, official press release, ad library record) score highest. Aggregator mentions, second-hand blog coverage, and community posts score lower. The score is pre-assigned per source type and cannot be negotiated up by volume.

The single most important signal. A claim confirmed by two or more independent source types (e.g. a job posting plus a changelog entry) scores near-maximum. A claim from a single source caps at 0.72 on this dimension regardless of other factors.

Does the insight link to a stored, timestamped source URL? Does it include the verbatim string it was derived from? Derived figures must also carry a recorded calculation. Missing any element drops this sub-score to zero, which floors the overall confidence below the 85% dispatch gate.

Measured by the scope inflation check (Layer C, check 3). Does the claim stay within what the evidence actually shows? A specific tier launch from a pricing page update is well-scoped. A broad market claim from a single LinkedIn post is over-scoped and loses points here.

The full 350+ source pipeline runs once a week. Each competitor is re-scraped end-to-end.

Verified insights from the Sunday scrape land in customer inboxes by Monday morning.

A report read on Tuesday cites observations at most nine days old. No stale signals.

Any contested claim receives one of three responses (update, retract, or stand by) within five business days.

Source overlap is a feature, not a bug. When G2, Glassdoor, and a company blog all independently describe the same feature launch, the confidence score rises. When they contradict each other, a flag is raised instead of an insight being shipped. Redundancy is how hallucination detection works at scale.

The 94% figure is the median confidence score across all insights shipped in the past 90 days that have received user feedback. Confidence is calculated as a weighted composite of: source authority, multi-source agreement, claim scope fit, and citation completeness. Only insights above the 85% gate enter customer briefings — so the 94% median reflects that most pass with margin to spare.

Accuracy is the percentage of shipped claims that have not been contested or corrected by customers, competitors, or our own internal QA cycle. As of Q2 2026, fewer than 2% of shipped claims have resulted in a correction request. We disclose every correction publicly on the relevant report page.

Yes — an AI synthesis pass converts raw observations into insight drafts, but that pass is followed by an adversarial verification pass that checks every claim against its source. Insights that fail verification are dropped, not edited. The pipeline also runs 30 automated checks across four layers before anything reaches a customer briefing.

Klue and Crayon are strong enterprise tools that require a dedicated CI analyst to get value. IndustryLens is built for teams without that resource — one person can run CI for the entire org. We also differ on transparency: every claim ships with a confidence score and a clickable source URL, so your team can verify what we deliver.

Bot-block detection (Layer A, check 4) and format validation (Layer A, check 5) catch this automatically. The source is flagged as degraded, removed from the active pool for that run, and a pipeline alert is triggered. Insights that relied on that source in previous runs are not re-surfaced until the source recovers.