Buyer Intent AI

Rank supplied discussion posts against the product or service you sell. Start with reproducible rules, or enable optional AI to assess semantic product fit. Every result keeps locally copied source evidence and an explicit explanation of how it was assessed.

What it does

This is a scoring layer for posts you already collected, including Reddit exports. It detects declared recommendation requests, switching, comparisons, problem descriptions and explicit budget mentions. It flags suspected vendor promotion, resolved needs and known stale posts before prioritizing possible business inquiries. It does not find contact details, crawl sites, send messages or promise a sale.

Rules use literal keyword matches and transparent signal weights. AI can assess semantic fit for the strongest eligible questions, while deterministic caps still apply. A post with no declared buying question cannot exceed20; promotion, a resolved need or a known stale post cannot exceed10. Scores are rankings, not purchase probabilities or proof of budget or decision-making authority.

Input

The Console starts with six clearly synthetic demo posts from example.com and Rules mode, without a model request. These rows demonstrate the report; they are not real prospects. Replace the sample with your own posts. To use datasetId instead, clear the prefilled posts field first. Direct API calls do not receive hidden demo data: provide your own posts or a dataset.

Provide productProfile, 1-12 keywords, and either posts or datasetId. Source rows may have id or postId, title, text, body or selftext, and url, sourceUrl or a Reddit permalink. A supplied createdAt date or numeric created_utc enables the age filter; unknown dates stay unknown.

Only the first maxPosts source rows are sampled, before filtering; default and maximum30. Larger datasets are marked truncated. A selector requests READ permission only for the chosen dataset. Invalid rows are skipped with a count; conflicting duplicates fail rather than silently changing the score.

The positive screening excerpt and AI text are limited to800UTF-8 bytes per post. Negative disclosures are checked against the full provided source before clipping, up to a hard20000-byte source limit. Longer sources fail rather than hiding a qualifying suffix. Email and phone-like patterns are redacted; this is not a complete anonymization service. The model receives pseudonymous IDs and prepared excerpts, not source URLs or author/contact fields. Product profiles are limited to1500UTF-8 bytes.

analysisMode defaults to rules. To use AI, select ai and enable allowPaidAI after reviewing credit usage. Default6 and maximum8 eligible candidates are assessed in one model call. Remaining posts retain rules assessments. language supports English and Polish explanations. minScore defaults to30, and maxAgeDays to30; set the age limit to0 to disable it.

Output

One dataset item contains the complete report. leads contains records above the chosen threshold; assessments contains all processed rankings for audit. Each record has the original source ID and URL, priority score, fit score, actual analysis mode, signals, keyword matches, freshness, flags and an evidence object copied from prepared source text.

The priority formula is base signal score multiplied by 0.25 + 0.75 * fitScore /100, rounded and capped by the negative filters. Base weights are35 for recommendation requests,25 for switching,20 for comparisons,15 for problems and15 for budget mentions, capped at100. A rules keyword match gives80 fit points and two matches100; AI supplies an integer semantic fit score. These are product rules, not a statistical conversion model.

modeRequested, modeUsed and aiStatus expose what actually ran. AI errors fail without a completed report by default. If allowRulesFallback was explicitly enabled, the completed rules report is labelled failed_rules_fallback and is billable at the same report price. If no candidates are eligible, no unnecessary model request is made. The key-value store contains RANKING, REPORT, OUTPUT, and provider usage or response records when available.

Pricing and cost control

The intended Store fee is0.10USD per completed report under intent-report, regardless of how many posts or leads it contains. A report with no qualifying leads is still a completed analysis. Failed input validation emits no completed report. The fee is charged once after the dataset write succeeds.

AI mode uses one fixed openai/gpt-4o-mini request through Apify's AI proxy, with a12000-byte prompt ceiling,1200 output tokens,45-second request timeout and no automatic retry. Available Apify credits and current proxy/model pricing apply; no additional model API key is requested. Credit use can occur even if an AI response later fails validation. A maximum Actor charge of0.10USD caps the report event, not a substitute for checking AI credit usage. Rules mode makes no inference request. Development runs stay inside the verified free-credit limit; no paid plan, proxy or campaign is required by this Actor.

Limitations and fair use

Use source material you may process. Quotes come from normalized, redacted and possibly clipped source text; flags are signals to inspect rather than independently verified facts. Rules are conservative, can miss phrasing and can exclude genuine inquiries. AI interpretations may be wrong and require human review. Late positive statements outside the excerpt can be missed even though late negative disclosures remain protected. Missing source dates cannot establish that someone still needs a solution.

The Actor does not collect personal profiles, infer personal attributes, verify purchasing authority, alter a CRM or send outreach. It returns potential business questions and product-fit interpretations only. Its output is not evidence that a person will buy or that any revenue has been earned.

Example

{
  "productProfile": "Example service implementing n8n workflows for small teams",
  "keywords": ["n8n", "automation"],
  "posts": [{"id": "demo-1", "text": "Looking for n8n automation. Manual work is too expensive and we have a budget.", "url": "https://example.com/thread"}],
  "analysisMode": "rules"
}

For optional AI, change analysisMode to ai and explicitly set allowPaidAI to true. For dataset input, omit posts and supply the selected datasetId. Sample input and output are available in the source package.