Outbound SDR.
Builds the list, researches every prospect, runs the multi-channel cadence, and books meetings on your AE's calendar — without torching your domains.
What it does from day one
(no ramp-up, no aspiration)ICP list building + waterfall enrichment via Apollo/Clay with pre-send email verification (<3% bounce target)
Per-prospect research dossiers: funding, exec changes, hiring signals — 5-10 reasons to reach out per account
Personalized 3-5 step email sequences written, loaded into Smartlead, throttled across warmed mailboxes
LinkedIn touches within safe caps (~25 invites/day) via Unipile hosted sessions
Reply triage: auto-handle the easy 70% (OOO, send-info, referrals, opt-outs), draft the rest for approval
Meeting booking against the AE's live calendar with confirmations, reminders, and no-show rebooking
How the work flows
(brief to logged output)Onboard & calibrate — client submits ICP, offer, tone, suppression lists; approves sequence copy and targeting once before anything sends
Build & enrich — pull prospects, waterfall-enrich, verify emails, research and score each account into the cadence queue
Execute cadence — email + LinkedIn (+ opted-in SMS/WhatsApp follow-ups only) fire on schedule, every send logged
Handle replies — webhooks wake the agent per reply: classify, auto-respond or escalate, negotiate times, book, confirm
Log & report — real-time CRM sync plus weekly digest (sends, replies, meetings, deliverability health); learnings feed back into targeting
The stack it plugs into
(real tools, official APIs)Apollo.io API — B2B contact search and base enrichment
Clay (or FullEnrich waterfall) — multi-provider enrichment and signal research
Smartlead API — cold-email infra: mailbox rotation, warmup, sending, reply webhooks
Unipile API — hosted LinkedIn/WhatsApp sessions so the ban risk lives on their infra, not our box
ZeroBounce/NeverBounce — pre-send email verification
Calendly Scheduling API / Cal.com — programmatic booking on the client AE's calendar
HubSpot / Pipedrive API — CRM of record for every touch, reply, and meeting
Its machine
(8GB RAM · 4 vCores, dedicated)A Python/Node orchestrator with a job queue, a webhook listener for reply events, and one headless Chromium for research scraping. Sending lives on Smartlead's infra and the LinkedIn session lives on Unipile's — the machine only makes API calls, which is deliberate: self-hosting a LinkedIn browser session is the highest-ban-risk architecture in 2026. 8GB is comfortable.
Seat economics
medium token load~150 prospects/day at ~10K tokens each ≈ 30-45M tokens/month — tens of dollars with caching, well inside the $499 seat; the fair-use lever is capping pages ingested per prospect (3-5) and caching company research across contacts.
Flat seat from $499/agent/mo — one agent, one dedicated machine. See pricing.
Judged by
(the numbers that matter)Meetings booked per month (human SDR benchmark: 8-15 qualified; ~80% show rate)
Reply rate / positive reply rate (B2B cold email averages 3-5%; ~30% LinkedIn connection acceptance)
Deliverability health — bounce <3%, complaints <0.1%, no LinkedIn restrictions (roughly half of AI SDR deployments hit a domain-reputation wall in 90 days; staying healthy is the differentiator)
Tasks in this role
(each documented in the task library)Never alone
(guardrails — how trust works here)Consent hard-block in code: no SMS/WhatsApp without recorded opt-in (TCPA is $500-$1,500 per message) — channel-gated, not judgment-gated
New sequences, pricing claims, and any targeting change require client sign-off before the first send; the agent never invents claims or discounts
Angry/legal replies, GDPR requests, and low-confidence classifications escalate to a human within minutes; opt-outs honored instantly and permanently
Deliverability circuit breakers: ≤50 emails/mailbox/day, ≤30 LinkedIn invites/day, auto-pause the whole campaign on bounce >3% or complaints >0.1% — agent never provisions new domains or resumes a paused channel alone
Full policy: our AI disclosure.
Try it
(alpha)"Watch an SDR work your lead — live": visitor enters a company URL and persona; left pane shows the agent researching and narrating its reasoning, right pane is a mock inbox where the personalized sequence materializes in ~30 seconds. Then the visitor plays the prospect — types a brush-off or objection — and the agent handles it live and books a slot in a sandboxed calendar. Zero real sends, zero deliverability risk. Cold calling ships first — try that demo today.