AI · Lead Enrichment
AI Enrichment Worker — 3 Options
After a lead is captured, the system can enrich it (firmographics, contacts, verified emails)
before manager approval. This step is optional and pluggable — the Client Company picks one option.
Whichever is chosen writes the same enrichment_data field; the CRM does not change.
Lead captured ──▶ [ AI Enrichment Worker ] ──▶ Manager Approval ──▶ Assigned
│
choose one ─── (1) Paid AI · (2) Local model · (3) Skip for now
│
choose one ─── (1) Paid AI · (2) Local model · (3) Skip for now
Option 1 · Paid AI / SaaS enrichment
A hosted provider (Apollo.io, People Data Labs, Clearbit/Breeze, Hunter.io) via API/webhook.
Pros
- Fastest to integrate — API key + webhook
- Largest coverage, incl. net-new contacts & emails
- No infrastructure; provider keeps data fresh
- Built-in email verification & phone data
Cons
- Recurring cost per credit/lookup; scales with volume
- Buyer data leaves your environment (privacy/GDPR)
- Vendor lock-in, rate limits, small free tiers
- Uneven coverage for some export regions
Cost owner: the subscription is a third-party service —
provided & paid by the Client Company.
Paid / freemium provider alternatives to Apollo.io
| Provider | Strength | Free tier | Notes |
|---|---|---|---|
| People Data Labs | Person + company enrichment API | ~100–1000 records | Closest match to Apollo's enrich API |
| Hunter.io | Email finder + verifier | ~25–50/mo | Best-in-class email finding |
| Snov.io | Email finder + verify + drip | Monthly credits | Also does drip outreach |
| Clearbit → HubSpot Breeze | Firmographics, logo | Limited | Free Logo API effectively unlimited |
| Prospeo / Findymail / Anymail Finder | Email finding | Small | Cheap fallbacks |
| Abstract API | Firmographics from domain | Free tier | Simple REST |
| RocketReach / Lusha | Contact lookups | Limited | Per-lookup credits |
| ZoomInfo / Cognism | Large B2B database | None (enterprise) | Highest coverage & cost |
Free / open public data sources (great for export B2B)
| Source | Data | Cost |
|---|---|---|
| GLEIF (LEI database) | Global legal-entity identifiers — verified counterparties | Free |
| OpenCorporates | Company-registry data worldwide | Free API tier |
| EU VIES | VAT-number validation for EU buyers | Free |
| Clearbit Logo API / favicon / Gravatar | Company logos & contact avatars | Free |
| WHOIS / DNS / MX lookups | Domain signals | Free |
Option 2 · Local model (self-hosted) PLANNED · DEFERRED
An in-house Enrichment Worker: scrape public pages + a local LLM, augmented by free registries.
Lead (domain / name)
→ fetch site (Crawl4AI / Playwright) → clean text (trafilatura)
→ local LLM (Ollama: Qwen2.5 / Llama 3.1 / Mistral) → structured JSON
→ merge registries (GLEIF LEI · OpenCorporates · EU VIES)
→ email pattern + SMTP/MX verify → write enrichment_data
→ fetch site (Crawl4AI / Playwright) → clean text (trafilatura)
→ local LLM (Ollama: Qwen2.5 / Llama 3.1 / Mistral) → structured JSON
→ merge registries (GLEIF LEI · OpenCorporates · EU VIES)
→ email pattern + SMTP/MX verify → write enrichment_data
Local stack (all free / open-source)
| Layer | Tool options | Role |
|---|---|---|
| LLM runtime | Ollama (or LM Studio / vLLM) | Runs the model locally |
| Model | Qwen2.5 7B/14B-Instruct · Llama 3.1 8B · Mistral | Extract & normalize structured data |
| Scraping | Crawl4AI · Firecrawl (self-host) · Playwright | Fetch company site pages |
| Content clean-up | trafilatura | Strip HTML to clean text |
| Search (optional) | SearXNG (self-hosted metasearch) | Find the company site & mentions |
| pattern generation + SMTP / MX verification | Free contact discovery | |
| Registries | GLEIF · OpenCorporates · EU VIES (free) | Verified counterparty data |
Pros
- No per-lead cost; unlimited volume once running
- Private — data never leaves the environment
- No vendor lock-in; pipeline fully owned
- Strong firmographics + verified registry data
Cons
- Needs a server (ideally GPU) + maintenance
- Can't invent net-new leads (enriches known domains)
- Email/phone discovery weaker than paid DBs
- Scrapers need occasional upkeep
Server configuration
| Tier | Model | GPU / VRAM | CPU / RAM | Storage |
|---|---|---|---|---|
| Minimum (CPU-only) | 7B–8B, 4-bit | none | 8+ cores / 32 GB | 50 GB SSD |
| Recommended | 7B–14B instruct | 12–16 GB (RTX 4060 Ti 16GB / A4000 / 3090) | 8 cores / 32 GB | 100 GB SSD |
| High quality / volume | 14B–32B | 24 GB+ (RTX 4090 / A5000 / A6000) | 12+ cores / 64 GB | 200 GB SSD |
Stack: Ubuntu 22.04 · Ollama + Docker (Crawl4AI / SearXNG) + Python worker. Cloud GPU can be rented on-demand instead of buying hardware.
Infrastructure owner: the server / GPU is provided by the Client Company.
Option 3 · Ignore for now (manual entry)
Skip automated enrichment; Sales completes leads manually. The worker slot stays in the flow so Option 1 or 2 can be switched on later without redesign.
Pros
- Zero cost, zero infrastructure, fastest to ship
- No privacy/vendor concerns
Cons
- Manual effort per lead; slower, less consistent
- No auto-fill of firmographics or emails
Decision — AI enrichment is deferred. The CRM ships with Option 3 (manual entry);
the pluggable worker slot stays in the flow. When enabled later, Option 2 (Local model) is the intended approach
— zero per-lead cost, private data, GPU server provided by the Client Company at that point. Paid providers / free
sources remain optional add-ons. No GPU or subscription is needed until enrichment is switched on.