Outbound engine: Find what is blocking outbound, then act on the next exception.

The problem

Outbound operations become difficult to control when sourcing, enrichment, verification, sending and CRM updates run independently. A contact can enter two sequences, receive outreach after replying or appear qualified in HubSpot while its verification failed. The system is designed to expose those handoffs, hold unsafe records and make recovery specific to the failed step.

The dashboard

13 views: Today, Client summary, Replies, LinkedIn threads, Follow-ups, Exceptions, Deliverability, Sequences, Lists and enrichment, Meetings, Opportunities, Economics, Run log.

The send-day conveyor: one lane per inbox from 08:00 to 18:00, send blocks drawn to scale, held sends hatched with the rule that held them, and the exception queue underneath with the people affected.

State lives in the URL (view, filters, date range), records open in drawers with their history, and every action that changes something is written to an activity log and is safe to repeat.

Data model

• contacts(id, account_id, email, email_status, enriched_at, cost_cents)
• accounts(id, domain, owner_id, crm_id)
• verifications(id, contact_id, provider, result, checked_at)
• campaign_memberships(id, contact_id, campaign_id, state)
• messages(id, membership_id, provider_id, reply_class)
• exceptions(id, entity_id, reason, resolution)
• integration_events(id, source, external_id, received_at)
• outbox(id, idempotency_key, destination, state, attempts).

Rules in the backend

1. Enroll only verified, unsuppressed contacts with no active enrollment in the same campaign.
2. Proposed bounce rule: hold a sending domain at a hard-bounce rate of at least 2.0% over the trailing seven days, with at least 100 attempts; flag smaller samples separately.
3. A reply or Calendly booking stops subsequent outreach before classification completes.
4. HubSpot writes use tenant:contact:operation:version idempotency keys.
5. Ambiguous replies enter review rather than automatically becoming qualified opportunities.

Storage

Recommend Supabase Postgres with tenant RLS. Proposed retention: raw callbacks 30 days, operational records 12 months, audit history 24 months, subject to deletion policy. Mask personal emails in shared views; restrict full addresses to authorized operators.

Reliability

Use bounded exponential retries, provider rate-limit handling and a dead-letter queue. Show checkpoint receipts, last successful collection, a dated last-updated time, stale data and failed CRM writes separately.

Security and access

Tenant authentication separates administrators, operators, approvers and read-only clients. Provider credentials remain in server-side secrets storage. Enrollment requires permission and an explicit approval gate.

Cost

Apollo access, Clay credits, verification requests, sending subscriptions and mailboxes incur costs. Track actual provider charges per processed record; no fixed enrichment price or savings claim is established.

Technology stack

• Apollo
• Clay
• TREG
• MONID
• Instantly
• HeyReach
• HubSpot
• Calendly

Map a workflow like this

Start with the handoffs your team already runs and where they break.

Discuss a workflow →