Churn Prediction Intervention
Cumulative listings over time, split by what each listing IS — agents vs MCP servers (the other type is too thin to chart yet).
Per buyer tier: the headline % is how that tier's own prices moved over the selected window (D/W/M above); the line below is its contribution to the niche's move(those contributions — not the headline moves — sum to the niche figure above). A thin tier (1–2 comparable agents) is shown and flagged, not hidden.
When a listing starts or stops showing a public price, the niche average moves even though nobody actually repriced — that's why the headline figure above only counts real price changes.
Too few priced agents in this niche for a reliable ladder (n=2). Individual prices are on each agent's card below.
3 agents tracked in this niche — most upvoted first.
Methodology. Prices are observed daily from vendor pricing pages (headless render + LLM extraction), normalised to monthly USD, and tagged with a confidence level. Figures are conservative — a price is never invented; agents whose pricing can't be verified are counted as unobserved. Agents can pull this same per-niche report programmatically via our MCP server's niche_report tool — see the docs. Last price scan: 2026-07-16 06:14 UTC · 4,921 captures today · liveness probe: 2026-07-16 14:30 UTC.