Benchmark experiment · Synthetic data

Who's canceling their SaaS subscriptions right now?

Market Signal Miner scans social posts for users actively abandoning paid tools — surfacing the exact moment someone is switching, before competitors reach them.

The pipeline

4-stage funnel. Deterministic scoring.

Stage 0Free

Keyword gate

Regex pre-filter eliminates ~40% of posts before any API call.

Stage 1+2API

Batch extraction

5 posts per Claude call. Returns a structured signal object or null.

Stage 3Free

EIS scoring

Deterministic 4-weight score: budget · friction · specificity · urgency.

Stage 4Free

Persist leads

Signals above EIS 7.0 written to JSONL + wiki. Deduped via Jaccard.

~72% token savings vs. naive single-pass at 100 posts

The finding

EIS alone leaves 33% false positives

High-EIS posts include people who already switched — scoring identically to active leads. A timing-state classifier fixes this completely.

EIS only

Precision66.7%
Recall100%
F1 Score80.0

5 of 15 flagged leads are already-decided posts

EIS + Timing filter

Precision100%
Recall100%
F1 Score100

Past-tense signals routed to intel bucket, not leads

Example signals

What a high-signal post looks like

r/
9.0

Notion

$120/monthtoo expensivemid-migration

"Actively migrating to Outline — self-hosted, open source. Need to be done by end of quarter before our annual renewal hits."

gh
9.8

GitHub Actions

$800/monthcancelingevaluating now

"Evaluating self-hosted Gitea + Woodpecker CI right now. Would pay up to $200/month. Canceling at end of billing cycle."

hn
9.0

Intercom

$480/monthcancelingevaluating now

"Actively trialing Crisp and Chatwoot right now, free trials run out in 14 days so we need to pick."

Early access

Get early access

We're opening up to a small group of sales and GTM teams first. Drop your email and we'll reach out when your spot is ready.

Or explore the benchmark →