Lead Intel Studio

Lead Intel Studio is a starter app for a custom lead-research and outreach system. It now includes a working CSV upload flow that persists imported contacts, enriches each record, groups account intelligence, and lets you generate outreach drafts for selected contacts or the full workspace.

Why this approach

This should be a normal application, not an OpenClaw-first automation stack.

• CRM data, message approvals, and outbound controls benefit from explicit roles, logs, and database records.
• The AI layer should be swappable so you can use OpenAI, Anthropic, or both depending on cost and quality.
• Dripify works well as an execution channel, but the core system of record should stay in your app.

Current MVP contents

• A workspace dashboard in app/page.tsx
• A client-side import and selection UI in app/workspace-client.tsx
• A sample JSON blueprint route in app/api/blueprint/route.ts
• A CSV import route in app/api/contacts/import/route.ts
• A draft-generation route in app/api/drafts/generate/route.ts
• CSV parsing logic based on your export shape in lib/csv.ts
• A local persisted data store in lib/database.ts
• Workspace persistence and query logic in lib/workspace-store.ts
• Domain types for contacts, research, and drafts in lib/types.ts
• Heuristic grouping and classification logic in lib/pipeline.ts
• AI and fallback workflow logic in lib/workflows.ts

Recommended production architecture

1. Ingestion
   • CSV upload for lead exports and account exports.
   • Salesforce OAuth app plus REST API for smaller syncs and Bulk API 2.0 for larger pulls.
2. Data model
   • accounts
   • contacts
   • research_runs
   • source_evidence
   • buying_groups
   • message_drafts
   • campaigns
   • engagement_events
3. Research pipeline
   • Normalize records.
   • Resolve account domain and LinkedIn identity.
   • Gather company and prospect facts with evidence links.
   • Classify organization type and buying group.
   • Generate outreach messaging from approved playbooks.
4. Execution
   • Gmail draft or send support with approval states.
   • Dripify CSV export so the sequencing engine stays consistent with your LinkedIn workflow.
   • Salesforce task and note writeback.
5. Feedback loop
   • Capture opens, replies, booked meetings, and opportunities.
   • Score which combinations of title, organization type, opener, and CTA are converting.

Suggested next implementation steps

1. Add database storage and auth.
2. Replace heuristic enrichment with a queued worker that calls the model provider and web research provider.
3. Add Salesforce OAuth and contact sync.
4. Add Gmail draft creation and send approval.
5. Add a Dripify CSV formatter for campaign uploads.

Setup

1. Install dependencies:

pnpm install

2. Start the app:

pnpm dev

3. Open http://localhost:3000.

CSV workflow

1. Open the app.
2. Upload a lead export CSV from the workspace.
3. The contacts are parsed, enriched, and stored in the local app database under .data/lead-intel-studio.
4. Select one or more contacts.
5. Generate outreach drafts for the selected contacts or for the full imported set.

AI configuration

• If OPENAI_API_KEY is set, outreach generation uses OpenAI.
• If OPENAI_ENABLE_WEB_RESEARCH=true, OpenAI web search is also enabled for the request.
• If no API key is set, the app falls back to built-in heuristic enrichment and drafting.

Current persistence note

• Local development persists to .data/lead-intel-studio.
• On Vercel this MVP uses /tmp/lead-intel-studio, which is useful for demo flows but not durable storage.
• For durable production persistence, the next step is swapping the local PGlite store for Neon or another hosted Postgres database.

Current recommendation on tooling

• Best core build choice: custom app with provider-swappable AI orchestration.
• Best AI setup: keep OpenAI and Anthropic both available behind a common interface.
• Best outreach execution mix: Gmail for direct email drafts and Dripify for LinkedIn-centric sequencing.
• Best immediate input path: CSV first, then Salesforce sync once schema and approval flow are settled.

Important guardrails

• Store evidence links for every enrichment claim.
• Separate confirmed facts from model inference.
• Keep a human approval step before first-touch sends.
• Rate-limit outreach and monitor bounce or spam signals.
• Keep compliance review in scope for CAN-SPAM, privacy, and platform usage policies.