An agentic sales-intelligence engine that finds where a payments platform is a genuine strategic fit — and scores every prospect by ecosystem leverage, not in isolation.
Sales teams burn time on prospects that look right but don't actually need the product. Silas is an agentic sales-intelligence engine I designed and built for my team at Equinox Payments. It maps the payments supply chain, validates real product-fit before it scores anything, ranks prospects on a five-dimension ecosystem-aware model, and outputs an interactive dashboard that explains the why behind every score. Built on Claude with a prospecting/enrichment data layer and a Python scoring engine.
Payments distribution isn't a flat market — it's a supply chain. Decisions flow downhill: from the gateway, to the software vendor (ISV), to the sales organization (ISO), to the merchant. A merchant uses whatever their software supports; that software connects through whatever gateway it certified with. Leverage lives upstream.
A generic lead list ignores all of that. It treats a downstream reseller and the gateway that powers a thousand merchants as if they were the same size of opportunity. Silas was built to see the whole chain — and to prioritize where a single relationship unlocks an entire downstream base.
Two failure modes waste the most time in technical sales:
Silas runs as a multi-step agentic workflow with a hard quality gate in the middle. Nothing gets scored until it has proven it actually needs what we sell.
| Stage | What happens |
|---|---|
| 1 · Understand the request | Resolve target segment, vertical, geography, size, and volume — sensible defaults, then move. |
| 2 · Validate need (quality gate) | Website-keyword and technographic checks confirm a real product touchpoint. Non-fits are disqualified before any scoring effort is spent. |
| 3 · Strategic research | Firmographics, technology stack, supply-chain relationships, growth and buying signals, decision-maker intelligence. |
| 4 · Map the ecosystem | Who supplies them, who they connect to, how many merchants/locations sit downstream — and an estimate of true opportunity size. |
| 5 · Score & rank | Five-dimension model → a priority tier and a specific recommended product for every prospect. |
| 6 · Dashboard & outreach | Interactive dashboard with the reasoning behind each score; outreach framing on request. |
Claude is the reasoning engine. It interprets the request, runs the research and ecosystem mapping, applies the scoring logic, and writes the strategic rationale. A prospecting/enrichment data layer supplies firmographics, technographics, and verified contacts; a Python engine applies the scoring model deterministically and renders the dashboard.
The scoring model is the heart of it — 100 points across five dimensions:
Scores roll up into priority tiers, and each prospect gets a specific recommended product for its use case — so the output is a decision, not a spreadsheet.
The design decisions that made Silas strategic rather than just fast:
| Decision | Options weighed | Call & rationale |
|---|---|---|
| Where the quality gate goes | Score everything, then filter · vs · validate need first | Validate first. Disqualifying false positives before scoring is what keeps the whole pipeline honest. |
| How to score | Prospect in isolation · vs · ecosystem-aware | Ecosystem-aware. One upstream relationship can be worth a whole downstream base — the model has to reward that. |
| Handling indirect players | Score them directly · vs · score by their upstream partner | Score by the partner behind them, and surface that partner as the real opportunity when it's the bigger prize. |
| Export behavior | Auto-export results · vs · human approval | Never auto-export. Always show a sample and the cost first, then wait for a person to say go. |
An intelligence tool is only useful if you can trust it. The risks and the guardrails:
| Risk | Mitigation |
|---|---|
| False-positive prospects | Website-keyword + technographic validation gate before scoring; competitors and non-fits auto-excluded |
| Missing or thin data | Absent data never auto-disqualifies — it's flagged, not punished, to avoid throwing away real prospects |
| Over-trusting a number | Silas explains why a prospect scored the way it did; the salesperson makes the call |
| Drift over time | A structured feedback loop captures rep judgments to tune the model against real outcomes |
What Silas is built to move:
The outcomes I'd own at scale: less time on non-fits, a higher share of conversations with prospects who actually need the platform, and outreach that lands because it references something specific and true.