Skip to main content

How it works

Honest detail on where Keendai gets data, how we score it, and how it lands in your workflow. Written for people who've built outbound systems before.

Sources

Two source-of-truth feeds. Each lead links back to the source URL and the timestamp it was scraped, so you can always verify the signal yourself.

Enrichment

After mining we enrich each candidate to maximize the chance you can actually reach someone:

Scoring methodology

Rule-based pre-filter

Hard ICP rules (vertical, geography, contact-channel requirement, operational status) run before AI scoring. Candidates that fail hard rules never make it to the model — saving spend and keeping the queue clean.

Gemini structured scoring

Each surviving candidate is passed to Gemini with your ICP, the mined signals, and the extracted facts. The model returns a structured JSON with sub-scores (ICP fit, signal strength, contact quality, recency) plus a plain-English reason that cites a specific number or date. No banned vague language.

Final blend

Rule pre-filter score + Gemini-weighted score (scaled by model confidence) + signal bonus, with an ICP-specific penalty for weak facets. The output is a 0–10 score that stays stable across reruns of identical inputs.

Transparent sourcing

Every signal on every lead links to its source URL and the exact timestamp it was scraped. The detail pane includes a signal timeline so you can audit the path from raw evidence to score. No mystery databases.

Export

Available at launch:

Fast-follow (weeks 1–2 post-launch): HubSpot and Pipedrive native push.

Data freshness

Mining is on-demand per ICP run. Each lead carries a canonical place_id so the same business never appears twice across exports for a workspace. Maps data decays roughly 3–5% per month; we plan periodic re-mines of older pool entries. Email is re-verified by Hunter at delivery time, not at mine time.