How Sales Signal
is built.
For technical evaluators: what's real, what's automated, and exactly where AI makes a decision.
Two data layers, not one list
Sales Signal runs on two distinct data layers. The first is a live, monthly-refreshed universe of every SEC-registered advisory firm — sourced directly from the SEC’s own public Form ADV filings, not scraped or third-party licensed. The second is your own roster: the specific accounts you’ve chosen to track, cross-referenced against that universe. Your roster is small and yours; the universe underneath it is large, shared and kept current automatically.
How the universe stays current
The firm universe is rebuilt monthly by an automated data pipeline — download the SEC’s published dataset, parse it, update our records. No AI is involved in this step. It’s the same category of deterministic data infrastructure a bank’s data-operations team would run, not a judgment call an agent is making.
Where AI is actually engaged — and where it isn’t
Only three steps use a language model: confirming that a detected signal is genuinely relevant before it reaches you, writing the pre-call brief and drafting outreach for your review. Everything else is regular code, not AI: loading data, looking up accounts, scoring signals, delivery, and the approval and cooldown rules behind every send. That’s deliberate. AI should make judgment calls where judgment is needed, and nowhere else.
How the AI work is split
A lightweight, fast model handles classification and relevance-confirmation — deciding whether a detected event clears the bar to reach you — and also generates the brief itself, grounded in your account’s own regulatory data. A separate model is reserved specifically for the outreach-drafting step: the one piece of AI-generated content that goes out in your voice, under your name, to a real prospect. That split is deliberate — routine judgment gates and structured research run on one model; the step closest to an actual send gets a dedicated pass of its own.
How a brief — or a SignalPlus card — lands on a specific offering of yours
Every 8-section brief ends by naming the one offering from your catalog that fits the account best, and every SignalPlus card is built around that same fit. It’s never a generic pitch. The two get there in different ways. In the brief, the AI weighs the account’s regulatory data and signal history and picks the best-fit offering as it writes. In SignalPlus, the match is a fixed, rules-based step: the account and today’s signal are compared against your offering catalog (the same one AI Edge uses), and only matches above a set relevance score count. A small, fast model then writes one line on why it’s worth a look.
Built on
LangGraph-based multi-step orchestration · a tiered large-language-model architecture matched to each task · Supabase for account and signal data · Pinecone vector search for SignalPlus’s match against your own offering catalog · SEC Form ADV as the primary regulatory data source.
Vertical scope, stated plainly
Sales Signal is built today for the SEC-registered RIA and wealth-management market. Extending it to another industry or data source is real engineering work, not a settings change. It’s available as a Custom Build (from $10,000), not a claim that we already cover every industry.
Want to see it
on your accounts?
Book a 30-minute scoping call. We'll walk through the architecture with your team and confirm the fit.