Building Executive Collateral That Actually Converts

The Problem

A conversation with the company's highest-performing Account Executive revealed a gap that open rate metrics had been masking. Cold outbound emails were getting opened but not getting responses. The subject lines were strong. The content wasn't. When pressed on what reps were sending, the answer was clear: there was nothing in the email that gave a prospect a reason to respond. No data, no context, no business case. Just another vendor claiming their product was transformative.

The underlying problem was structural. Sales reps had no credible, prospect-specific data to anchor cold outreach in a business conversation. Without that, even the best subject line leads to a dead end.

The Brief That Proved the Concept

Rather than improving email copy (a surface-level fix), the initial goal was to give reps something worth sending. For airline prospects specifically, the relevant business question was operational: what does Average Handle Time actually cost at scale, and what happens to that cost when it moves even slightly?

Public data sources were used to model AHT benchmarks across U.S. airline contact center operations, then validated against internal customer ROI figures from actual deployments. The result was a dataset grounded in both industry reality and proven customer outcomes, not vendor claims.

V1 was built in approximately an hour using AI-assisted development tools. It presented the data clearly but read too much like a marketing piece. V2 was rebuilt from scratch with a different design philosophy: research-grade over polished, datasheet over brochure. The response at Sales Kickoff was immediate: reps wanted to use it and asked to collaborate on further development.

But the sales kickoff also revealed the limitation of the one-off approach. Reps covering different verticals needed their own versions. Personas within the same account had different priorities. The airline brief worked for a VP of Contact Center; it wasn't what a CFO or a CIO needed to see. And even when the right brief reached the right person, there was no way to know if they'd opened it, how long they'd spent on it, or whether they'd shared it internally.

The brief was the proof of concept. The problem it exposed required a platform.

The Build

The platform, SIGNAL, went from concept to production across a single engineering cycle, built by one person using AI-assisted development tools without a dedicated engineering team. It spans 10 PostgreSQL schemas, a React/TypeScript frontend, and an Express.js backend, with Salesforce as the data source of record throughout.

Research Integrity Pipeline

The foundation is a 14-stage ingestion pipeline that governs every finding that can appear in a piece of collateral. No stat reaches a prospect unless it has passed through the full pipeline.

Stages include exact duplicate detection via MD5 hash, near-duplicate screening via pgvector cosine similarity (0.82 threshold), source tier classification that rejects Tier 4 publishers and non-whitelisted sources, LLM-based claim normalization, confidence scoring across four levels (verified, single-source, preliminary, anecdotal), URL content verification, a hallucination guard using LLM cross-reference checking, full semantic deduplication with embeddings, auto-tagging, and human review gates.

Pipeline outcomes are explicit: committed, held for review, merged with existing, rejected, duplicate, or error. Nothing passes through on a best-effort basis. Multiple LLM providers power research sessions, each implementing a standard interface; findings produced by any provider pass through the same ingestion pipeline.

Collateral Builder

Collateral-ready findings (those that have cleared the full pipeline) are surfaced to reps filtered by vertical, asset tier, and persona relevance. Reps assemble these into collateral pieces; the platform tracks which specific findings appear in which piece via a finding-usage audit table, maintaining a traceable chain from prospect-facing asset back to the primary source document and confidence rating. Pieces move through a draft-to-published status lifecycle with timestamp tracking.

Playbook Engine

A sales playbook engine covers 17 personas across the full enterprise buying committee: VP and Director-level roles across Contact Center, CX, Operations, Quality Assurance, Digital Transformation, Technology, Finance, Procurement, Compliance, and Insights/VoC functions.

Each persona has a corresponding 5-step email cadence tuned to their psychology, a branching talk track with discovery, gap statement, and pivot paths, and a set of matched pain points, objections, and success metrics. Six messaging themes can be applied cross-persona. Buying committee structure maps every role to a function: Key Buyers, Decision Influencers, and Ratifiers.

Outreach Orchestration

The outreach layer connects the playbook engine to live Salesforce data. Given an account, it fetches contacts via the SFDC Live Query client, matches each contact's title to a persona using keyword and seniority matching, computes a priority score adjusted for engagement recency and persona confidence, and returns a ranked prospect list with cadence recommendations.

From that list, the compose endpoint generates personalized email drafts per prospect using the matched cadence template, creates collateral share links, and logs the outreach session.

The Portal and Engagement Tracking

Collateral is delivered via a tokenized, email-gated portal. No login required on the prospect side; they enter an email address and receive JWT-gated access to the document, an AI-generated insights panel, and a rep call-to-action.

Every interaction is tracked in real time: scroll depth, time on page, CTA clicks, downloads, and internal share events. Events flow through a rate-limited ingestion layer into the signal engagement store and are aggregated by a background sync job into the collateral analytics table. Reps see a live engagement timeline per account in the SIGNAL dashboard.

The Sandbox

For prospects who need to model the numbers themselves, a separate interactive OpEx sandbox delivers a company-specific cost model, also email-gated, with its own engagement event tracking. Sandbox interactions (field modifications, competitor comparisons, snapshot saves) are weighted by intent and fed into the account-level intent score alongside portal events. Sandboxes expire after 30 days and are archived with full activity stats, email session history, and working config before purge.

Intent Scoring

Account intent scores combine engagement signals from portal and sandbox activity, Gong call signals, ZoomInfo propensity data, and deal stage progression. Positive signals carry explicit weights; negative signals adjust the score in the opposite direction. All signals apply time decay: 100% within 7 days, 60% within 30, 25% within 90, zero after. Scores are recomputed on demand and on a scheduled basis, with hot-account alerts and rep-level performance metrics surfaced in the SIGNAL overview dashboard.

Intelligence Layer

A 10-layer LLM security pipeline powers natural language queries against live Salesforce data. The pipeline covers input sanitization, prompt injection detection, prompt hardening, PII redaction, field access validation, bulk extraction prevention, row limit enforcement, anomaly scoring, and full audit logging.

A dual-path query architecture separates structured CRM queries from transcript analysis, where transcript data is handled in an isolated clean room with no database access, a direct architectural response to the indirect prompt injection risk that free-text call data introduces. Fifteen-plus pre-built query templates cover pipeline analysis, prospecting, deal briefs, analytics, and transcript review. All operations are read-only.

The Architecture Decision That Mattered Most

The research pipeline could have been simpler. An LLM could ingest a finding, decide if it looks credible, and pass it through. That architecture produces plausible results, and plausible is not good enough when sales reps are presenting data to CFO-level prospects in active deals.

The decision to build a 14-stage deterministic pipeline with explicit confidence ratings, source tier enforcement, and a human review gate was a product decision, not an engineering one. Sales collateral is only valuable if it's more credible than what a rep could produce on their own. The moment a prospect finds one stat they can't verify, the entire brief loses its authority.

The same principle governed the intelligence layer. The architecture makes a deliberate split between what the model handles and what deterministic code handles, and never crosses the line in the wrong direction.

The LLM handles language. Deterministic code handles everything that touches data accuracy, access control, or security. The system feels conversational to the user and behaves like a data system underneath.

Platform Evolution

SIGNAL shipped its first release on February 5, 2026. The current release is v0.11.0: 11 versions across 25 days, built and maintained by one person.

Six of those eleven versions shipped on the same day. The platform that started as a vertical data brief on February 5 had a full research pipeline, live Salesforce integration, a 10-layer LLM security architecture, an adaptive automation layer, and a prospect-facing portal with engagement tracking, all running in production, within 25 days.

The Result

SIGNAL is in active deployment, covering four industry verticals with a research library governed by a pipeline that enforces source integrity at every stage.

What previously required a rep to manually research a prospect, find credible data, write a personalized email, and send it blind (with no visibility into whether it was ever read) now takes a single API call. A rep enters an account, receives a ranked and persona-matched prospect list drawn from live Salesforce data, and gets personalized cadence-tuned email drafts with collateral share links ready to send. When the prospect opens the brief, engagement data flows back in real time. The rep knows who read it, how long they spent, whether they scrolled to the business case, and whether they shared it internally, before the follow-up call.

The problem that started the project (reps sending emails with nothing worth reading) now has a structural answer. The data creates the conversation. The platform tracks what happens after it starts.

The Insight

The original brief worked because it gave a CFO-level prospect a business reason to respond before anyone mentioned the product. The platform works for the same reason, at scale, across every vertical and every persona in the buying committee.

What changed between V2 of the airline brief and the full platform wasn't the insight; it was the infrastructure required to make that insight repeatable. Verified research, persona-matched messaging, engagement tracking, and intent scoring aren't four separate features. They're four stages in a single system designed to produce one outcome: a sales rep who walks into a conversation already knowing what the prospect cares about and whether they've engaged.

The brief proved the concept. The platform is the product.