Creator Campaign Score & Audience Signal API

Turn supplied creator or influencer aggregate metrics into transparent 0–100 campaign-shortlist scores, component scores, cohort ranks, percentiles, and anomaly screening signals. This Apify Actor is designed for influencer marketing, creator discovery, campaign planning, UGC shortlists, and agency workflows that already have authorized metrics.

It performs no scraping, makes no social-platform API request, uses no proxy or browser, and needs no login. It accepts no bio, caption, comment text, audience-member data, or arbitrary source object.

What it answers

• How do supplied creators compare within this campaign cohort?
• Which rows have stronger reach, engagement, view, consistency, or data-quality evidence?
• Which metric combinations cross transparent anomaly thresholds and deserve human review?
• Which records are incomplete or based on very small samples?

The score describes only the supplied aggregate metrics. It is not a judgment of a person's worth, proof of audience authenticity, a fake-follower detector, a brand-safety decision, or a prediction of reach, conversion, revenue, or ROI.

Input

Use inline creators, or select an Apify Dataset with creator_dataset_id (READ only). Each row uses this closed contract:

{
  "creator_id": "campaign-row-17",
  "platform": "instagram",
  "followers": 42000,
  "average_likes": 1680,
  "average_comments": 84,
  "average_views": 31000,
  "posts_sampled": 24,
  "interaction_cv": 0.42,
  "view_cv": 0.55
}

interaction_cv and view_cv mean standard deviation divided by mean across the represented posts. Use one consistent collection scope and measurement method per cohort. The Actor does not authenticate caller-declared metrics or provenance.

creator_id is converted to a stable 24-character SHA-256 pseudonym using your private id_namespace. Raw creator IDs are hidden unless include_creator_ids is explicitly enabled. Use a campaign- or workspace-specific namespace to prevent easy cross-project joins.

Output

Each creator_score row contains:

• campaign_suitability_score from 0 to 100;
• reach, engagement, view, consistency, and data-quality component scores;
• the observed engagement rate and view-to-follower ratio;
• deterministic cohort rank and percentile;
• a closed list of anomaly signals and data warnings;
• a stable privacy-minimized row ID.

A final free cohort_summary records analyzed, planned, delivered, omitted, and charged row counts; spending-limit status; mean score; source truncation; signal count; fixed weights; and mandatory limitations. The default Dataset starts with the exact billable creator-row prefix, followed by the summary.

Exact scoring model

The Actor uses no ML model or hidden training data.

• Reach (15%): log-scaled supplied followers; one million reaches the component ceiling.
• Engagement (30%): saturating function of (average_likes + average_comments) / followers.
• Views (20%): saturating function of average_views / followers.
• Consistency (15%): inverse bounded mean of supplied interaction and view coefficients of variation.
• Data quality (20%): documented-field completeness plus a sample-size component that reaches its ceiling at 20 posts.

When an optional component cannot be computed, its weight is redistributed across the available components. The output always exposes every component used. Cohort order is score-descending with the pseudonymous row ID as the deterministic tie-breaker; percentile is relative only to rows in the current run.

Anomaly signals are not proof

The Actor emits a screening signal when supplied aggregates show:

• engagement rate at or above 20%;
• view-to-follower ratio at or above 5;
• at least 20 average comments and a comment-to-like ratio at or above 1;
• zero supplied views alongside positive supplied interactions.

These thresholds identify combinations worth checking. They do not establish manipulation, fake followers, fraud, or authenticity. small_sample_warning (fewer than five represented posts) and incomplete_metrics_warning are not counted as anomaly signals.

Pricing

Pay per event: $0.003 per creator-row-scored plus Apify's $0.00005 Actor start event at the default 512 MB memory. The live Monetization panel says User pays platform usage costs: No, so normal run platform usage is included in these event prices; post-run Dataset access can still incur normal storage-operation charges under Apify's Store billing rules. Apify may scale start-event count when a caller increases memory. One successfully delivered creator_score Dataset row equals one paid unit. The free cohort summary is never a creator-row event. A spending limit produces an exact paid leading prefix and reports planned, delivered, omitted, and charged counts.

At the default 512 MB: 25 creators cost $0.07505; 100 cost $0.30005; 1,000 cost $3.00005, before any taxes or plan-specific charges shown by Apify.

Good agent requests

• “Rank these supplied creator metrics for a campaign and explain each component.”
• “Score the creator rows in this Dataset without exposing raw creator IDs.”
• “Find extreme engagement or view-ratio signals in my authorized influencer shortlist.”
• “Compare this cohort using deterministic creator campaign scores.”

Do not choose this Actor when you need Instagram, TikTok, or YouTube scraping; comments or profile text; audience demographics; contact enrichment; authenticity certification; or automated decisions about people.

Privacy and security

The Actor has limited permissions, reads only an explicitly selected Dataset, writes its run Dataset and KVS, and has no network collector. Source rows are bounded by row, per-row byte, and total-byte limits. Unknown row and input fields fail closed. No supplied metric or identifier is logged by application code.

Use only data you are authorized to process. Keep raw IDs hidden unless an operational join requires them. Share only synthetic, redacted reproductions in support requests.

Benchmark and reproducibility

The repository includes an original CC0 synthetic 40-case benchmark with injected normal and anomaly conditions. Release checks require deterministic output, exact schemas, exact event accounting, anomaly-signal recall of at least 0.90, and no raw identifier leakage. Synthetic recall measures only the documented injected thresholds—not real-world follower authenticity.