Core Concepts

The fundamental concepts you need to understand how Sellestial works.

The Big Picture

Sellestial is a data operations platform that sits between your HubSpot CRM and the messy reality of maintaining clean, enriched data. It works by running pipelines — automated workflows that continuously process your contacts and companies using AI, external data, and custom logic.

Think of it as a self-healing data layer: pipelines detect problems, validate quality, enrich gaps, and enforce standards automatically, in the background, as your CRM evolves.

Core Architecture: Two Tiers

Sellestial has two complementary systems:

Special Pipelines

Pre-built pipelines for common problems (Deduplication, Enrichment, Sequences, Visitor ID). These have custom UIs designed for their specific use case and work immediately after connecting HubSpot.

When to use

Standard data operations that most companies need.

General Pipelines

Flexible workflows you configure or build from scratch. Install templates from the Marketplace or create custom automation using the five pipeline kinds.

When to use

Custom workflows, specific business logic, or combining operations into complex chains.

Pipelines: The Fundamental Unit

Every automation in Sellestial is a pipeline — a workflow that processes CRM records.

What Pipelines Do

Receive records → Process data → Take action

1. Input: Records enter from HubSpot lists, manual enrollment, or events
2. Processing: Fetch context from data sources, execute Python code, run AI classification/reasoning, or generate content
3. Output: Update HubSpot properties, create tasks, emit events, trigger webhooks, etc.

Entity Types

Each pipeline processes one specific entity type:

• CONTACT — HubSpot contacts (individual people)
• COMPANY — HubSpot companies (organizations)
• DUPLICATE_PAIR — Pairs of records identified as potential duplicates
• CONVERSATION_THREAD — HubSpot inbox conversation threads (for future use)

The Five Pipeline Kinds

Each kind uses a different engine optimized for specific tasks:

Kind	What It Does	Best For	Cost Magnitude
Code	Execute Python logic without AI	Email validation, data transformations, API calls, calculations	n * 0.01 credits
Classifier	AI assigns predefined categories with confidence scores	Data quality assessment, lead scoring, categorization	n * 0.1 credits
StructuredData	LLM extracts structured information with predefined schema	Data extraction, field population, structured parsing	n * 0.1 credits
Agent	Autonomous AI with tools for web research and reasoning	Duplicate detection, data verification, complex analysis	n * 1 credits
Sequence	Generate personalized outreach content (emails, LinkedIn, calls)	Sales sequences, marketing campaigns	n * 10 credits

Examples

• Code: DNS MX record lookup to validate email deliverability
• Classifier: Classify contacts as real people vs role-based emails vs fake data
• StructuredData: Extract company information from website with predefined output schema
• Agent: Visit both company websites to verify if records are true duplicates before merging
• Sequence: Generate personalized 3-email conference outreach campaign

How to Choose

Choosing the right pipeline kind

🎯 Need outreach content? → Use Sequence

🔍 Need autonomous research or web verification? → Use Agent

📊 Need to extract structured data with a predefined schema? → Use StructuredData

🏷️ Need to assign discrete categories? → Use Classifier

⚡ Need deterministic logic or calculations? → Use Code

Rule of thumb

Use the simplest kind that solves your problem. Code is fastest/cheapest, Agent is slowest/most powerful. StructuredData and Classifier are mid-tier options for AI-powered data extraction and categorization.

See Pipeline Kinds for detailed comparison and decision trees.

Entity Lifecycle & Statuses

Entities pass through well-defined statuses as they flow through pipelines:

Enrollment Phase:

• ENROLLED — Discovered and added to pipeline, not yet scheduled
• SCHEDULED — Queued for asynchronous processing

Processing Phase:

• PENDING — Ready for processing, waiting for worker
• GENERATING — Worker actively running generator logic
• GENERATED — Output produced, awaiting confirmation

Confirmation Phase:

• WAITING_FOR_CONFIRMATION — Paused for manual validation
• CONFIRMED — Output approved, ready for final processing

Terminal States:

• COMPLETED — Successfully processed and persisted (success)
• FAILED — Error encountered, requires review (failure)
• IGNORE — Explicitly excluded from further processing (skipped)

Key concept

Each status represents a specific point in the pipeline’s execution flow. You can track entities through these statuses in the Pipeline Runs tab.

Data Flow: Sources, Processing, Actions

Data Sources

Pipelines pull contextual data from providers during processing:

• HubSpot — Contact/company/deal properties, communication history, associations
• LinkedIn — Personal profiles, company pages, posts
• External — DNS records, company news, financing events, website content

Key concept

Pipelines don’t just process the raw record — they can fetch enrichment context from multiple sources to make better decisions.

See Data Sources for the complete provider catalog.

Processing with Hooks

Pipelines process records through stages. Optional hooks (Python functions) let you inject custom logic:

onPreGenerate — Validate inputs, skip records conditionally
onReviewOutput — Evaluate results, decide if confirmation needed
onResultConfirmed — Execute side effects (HubSpot updates, webhooks, events)

Key concept

Hooks let you customize behavior without changing the core pipeline logic.

Actions

After confirmation, pipelines can:

• Update HubSpot properties with configurable write modes (always overwrite, write if empty, respect user edits)
• Create tasks and notes for follow-up
• Emit events to trigger downstream pipelines (enabling pipeline chaining)
• Call webhooks to notify external systems

Event-Driven Automation

The events system enables real-time, reactive pipelines that trigger automatically.

Event Subscriptions

Pipelines subscribe to events matching their object type:

Example: Contact enrichment pipeline subscribes to contact_created events → automatically enriches new contacts as they’re added to HubSpot

Configuration

• Event type — What to listen for (record created, updated, property changed, etc.)
• Cutoff date — Ignore events before this timestamp (prevents processing historical backlog)
• Re-enrollment — Allow entities to re-enter pipeline when event occurs

Event Emissions

Pipelines emit events after successful processing, creating opportunities for pipeline chaining.

Example flow:

Email Validator (Code)
  ↓ emits: email_validated
Contact Enrichment (Agent)
  ↓ emits: contact_enriched
Lead Scoring (Classifier)

Key concept

Events decouple pipelines. Upstream pipelines don’t need to know about downstream consumers. This enables building complex workflows from reusable components.

Control Mechanisms

Review Policies

Control when outputs get approved:

Auto-confirm — Immediate approval, no human intervention
Require review — Manual confirmation for every result
Auto-confirm after delay — Timed review window (e.g., 10 minutes)

Best practice

Use human review for high-impact operations (merges, deletions) until confident, then transition to auto-confirm.

Processing Limits

Control speed and volume:

Enrollment limits — How fast records enter (e.g., 1 contact per 5 seconds)
Generation limits — How fast processing happens (e.g., 50 contacts per day)
Weekend blocking — Pause processing on weekends

Why this matters

Limits control costs, prevent API rate limiting, and let you scale gradually.

Versioning

Pipelines use immutable versioning:

• Each deployment gets a unique hash (e.g., fa697484cb)
• Complete version history with audit trail
• Rollback to any previous version instantly
• Changes compile automatically but require explicit deployment

Key concept

You can modify and test changes without affecting production until you explicitly deploy.

Write Modes: Respecting Data Provenance

When pipelines update HubSpot properties, write modes control update behavior:

Always overwrite — Replace all values (good for standardization)
Write if empty — Only populate blank fields (safest)
Write if not modified by user — Respect manual edits via property history
Skip — Never write this field

Why this matters

Different data has different trustworthiness. User-entered data is often more reliable than automated enrichment. Write modes let you enforce this hierarchy.

Marketplace: Pre-Built Templates

The Marketplace offers production-tested pipelines ready to install:

Template kinds:

• Code — Deterministic logic for validation, transformations, and API integrations
• Classifier — AI categorization for data quality assessment and segmentation
• StructuredData — Structured data extraction and normalization
• Agent — Research, verification, and complex analysis with web tools

Installation flow:

1. Browse marketplace — Explore pre-built pipeline templates
2. Install template — Click “Use Template” (RevOps Assistant guides you through setup)
3. Configure pipeline — Set sources, limits, and field mappings in Settings tab
4. Activate pipeline — Toggle activation in Settings
5. Run first test — Test with sample records in Test tab
6. Deploy — Deploy the pipeline to enable mass processing
7. Add input sources — Connect HubSpot lists or other sources
8. Monitor & confirm — Review results in Pipeline Runs tab and confirm to push to CRM

Key concept

Templates are starting points. Every aspect is customizable before deployment. The RevOps Assistant will guide you through initial setup.

Cost Model: Credits

Sellestial uses credit-based pricing — you pay for actual processing, not seats or subscriptions.

Cost by Pipeline Kind

Processing n records costs approximately:

• Code: n × 0.01 credits (nearly free for most operations)
• Classifier: n × 0.1 credits (low-cost AI categorization)
• Agent: n × 1 credit (AI with research and tools)
• Sequence: n × 10 credits (multi-step content generation)

Additional Services

• Contact Enrichment: LinkedIn profile data and professional information
• Company Enrichment: Firmographic data and business intelligence
• Visitor ID: Anonymous website visitor identification

Cost Optimization

• Start with free validation — Use Code pipelines for basic validation first
• Filter before enriching — Use Classifiers to identify quality records before expensive operations
• Set daily limits — Control spending with processing caps in pipeline settings
• Target strategically — Enrich high-value segments, not entire database

Monitor consumption in real-time via the Usage dashboard.

HubSpot Integration

Sellestial connects directly to HubSpot via OAuth:

Inbound (from HubSpot):

• Lists as pipeline input sources
• Property values for context
• Communication history
• Associations between objects

Outbound (to HubSpot):

• Property updates (respecting write modes)
• Task/note creation
• Custom properties (typically prefixed sellestial_)
• Association management

Key concept

Sellestial reads and writes directly to HubSpot — no middleware, no sync delays, no data silos.

Strategic Workflow: The Right Sequence

1. Hygiene first — Clean existing data

• Enable deduplication rules
• Run validation pipelines
• Clean and normalize records
• Merge duplicates

2. Enrichment second — Fill gaps in clean data

• Configure enrichment with conservative write modes
• Set daily limits
• Monitor success rates

3. Automation third — Custom workflows

• Install marketplace templates
• Build custom pipelines
• Create event-driven chains

Why this order matters

Enriching dirty data wastes money. Clean first, then enrich clean records.

What Makes Sellestial Different

Declarative, not imperative: Configure what should be true (e.g., “contacts should have valid email domains”), and Sellestial enforces it continuously. You’re not writing scripts that run once.

AI-native: Modern LLMs handle tasks that previously required custom code: classification, extraction, normalization, research, and nuanced decision-making.

Event-driven: Build sophisticated multi-stage workflows by chaining pipelines through events, without tight coupling.

Data-centric: Core mission is data operations (hygiene and enrichment), not a general automation platform. This focus drives architecture and feature decisions.

Next Steps

New to Sellestial?
Start with the Quick Start Guide to connect HubSpot and install your first pipeline in 5 minutes.

Ready to build?

• Pipeline Kinds — Deep dive into each kind
• Template Marketplace — Browse pre-built pipelines
• Data Sources — Available providers and field mapping

Specific use cases:

• Data Deduplication — Configure duplicate detection
• Data Enrichment — Automated contact/company enrichment
• Sequences — AI-powered outreach (legacy)