Outbound as a system: funnel, SLAs, unit economics

The outbound funnel as a state machine

1OUTBOUND STATE MACHINE (contact-centric)23  discovered → scored → eligible | rejected4  eligible → enriching → enriched | enrich_failed5  enriched → validating → valid_email | invalid6  valid_email → sequenced → in_flight7  in_flight → replied | completed | bounced | unsub | paused8  replied → classified → routed → meeting_booked | nurture | closed_lost910  Rules of the machine:11  - one active sequence per contact per product motion12  - invalid / bounced / unsub are terminal for cold email13  - replied always pauses steps (HITL or auto-stop)14  - every transition emits an event for attribution + ops1516  Account-level twin: accounts score, contacts attach, meetings roll up.

1. 01Discovered — entered via intent, scrape, import, or CRM export; identity may still be weak.
2. 02Scored / eligible — passed ICP threshold; non-goals rejected with a reason code.
3. 03Enriched — required fields present at declared confidence; waterfall cost recorded.
4. 04Valid — email/phone passed validation; catch-all policy explicit.
5. 05In-flight — sequence active; stop conditions wired.
6. 06Replied / routed — human or classifier owns next action under SLA.
7. 07Attributed — touch logged so experiments do not lie later.

SLAs between stages

1STAGE SLAs (example B2B SaaS mid-market outbound)23  Stage hop                      SLA              Owner4  ---------------------------    --------------   ----------------5  List load → scored             < 4 hours        scoring service6  Eligible → enriched            < 24 hours       enrichment worker7  Enriched → validated           < 2 hours        validation hop8  Valid → first send             < 12 hours       sequence engine9  Positive reply → human touch   < 15 minutes     reply ops10  Meeting set → CRM opp          < 1 hour         CRM write-back11  Bounce spike detect            < 30 minutes     deliverability1213  Error budget example: <2% of eligible contacts stuck >48h in enriching.14  Breach → page the GTM eng on-call, not “check Clay tomorrow.”

Unit economics — the numbers that kill bad volume plans

1OUTBOUND UNIT ECONOMICS (back-of-envelope)23  Meetings needed / month ............... M4  Show rate ............................. S      (e.g. 0.70)5  Booked meetings needed ................ M / S6  Positive reply → meeting rate ......... R_m    (e.g. 0.25)7  Positive replies needed ............... (M/S) / R_m8  Positive reply rate ................... R_p    (e.g. 0.012)9  Delivered contacts needed ............. positives / R_p10  Validation pass rate .................. V      (e.g. 0.78)11  Enriched contacts needed .............. delivered / V1213  Cost stack per contact:14    list + score + waterfall enrich + validate + send infra + tool seats15  Cost per meeting = total monthly cost / meetings held16  Kill criterion: cost/meeting > max CAC share for outbound channel1718  Interview move: write this table before naming Clay or Instantly.

1. 01Cost per enriched contact — waterfall spend / successful enrichments (not attempts).
2. 02Cost per delivered first-touch — includes validation + send infra + seat amortization.
3. 03Positive reply rate — positives / delivered (define positive: interested, not OOO).
4. 04Meeting rate — meetings held / positives (or / delivered — pick one and stick).
5. 05Pipeline $ per $ outbound cost — only honest with multi-touch caveats (L7).
6. 06Domain health budget — bounce/complaint caps are hard constraints, not KPIs to game.

Observability for outbound

1. 01Funnel census — counts per state daily; alert on unexpected pile-ups.
2. 02Hop latency histograms — p50/p95 time in enriching, validating, awaiting first send.
3. 03Reason codes — rejected_icp, enrich_timeout, catch_all_policy, hard_bounce, unsub.
4. 04Provider health — waterfall step success rate and $/success per vendor.
5. 05Domain health — bounce rate, spam complaint rate, inbox placement samples.
6. 06Reply SLA breach rate — positives without human touch inside the budget.

Tool layer vs control plane

Interview framing — design outbound for X

1. 01Q: How is outbound different from marketing nurture? Outbound is interruptive, identity-resolved, and sequence-driven with hard stop rules and domain risk. Nurture is permissioned or product-triggered with softer reputation constraints. Mixing lists without policy is how you spam your own customers.
2. 02Q: What would you instrument first on day one? State census + bounce rate + positive-reply SLA. Those three catch zombie inventories, domain death, and the revenue-critical human hop. Opens are a lagging vanity metric under privacy noise.
3. 03Q: Where do you put human-in-the-loop? High-value account personalization, positive-reply handling, and exception queues (legal, VIP, angry replies). Not on every enrichment cell — that does not scale and creates silent backlog.

Anti-patterns that interviewers ding in five minutes

1. 01Tool-first answer — naming Clay/Instantly before states, SLAs, or unit economics.
2. 02Infinite volume — no bounce budget, no AE capacity model, no kill criteria.
3. 03Sheet as SoR — operational truth in mutable tabs without events or locks.
4. 04Open-rate north star — optimizing pixels while positives and meetings rot.
5. 05Silent retries — enrichment/send loops without idempotency or DLQs.

Outbound is a factory with reputation as inventory. You can speed the line until you melt the brand — good GTM engineering is knowing which dial is heat.