A 7-person agency built an AI-assisted Reddit ops room that replaced 40% of cold outbound
Northwind Growth — 7-person B2B demand-gen agency
Northwind runs demand-gen for Series A–B B2B SaaS companies. Their cold email reply rates had fallen from 4.1% to 0.7% over 18 months. Rebuilding around Linkeddit + Claude MCP + HubSpot turned Reddit into their single largest net-new channel inside one quarter.
Background
Northwind's thesis is that B2B founders trust other founders over cold email. Reddit is where those conversations happen — but the operational problem was always the same: by the time a strategist saw a promising thread, it had 400 comments and the moment was gone. The team needed a system that surfaced threads in under an hour and gave them a ready-to-edit first draft.
The problem
Three structural problems with cold email had compounded: (1) deliverability collapsed after the Google/Yahoo sender changes; (2) the ICPs they were chasing were on 3+ cold email sequences a week; (3) every agency on LinkedIn was running the same plays. They needed a channel that was still asymmetric — where the right founder answering the right question couldn't be faked.
Pipeline configuration
Northwind Growth runs 2 Linkeddit pipelines. Each one is scoped to a narrow set of subreddits and keyword patterns so the lead queue never turns into noise.
Agency-looking founders
looking for marketing agencyfired my marketing agencycan't find a good agencyB2B agency recommendationfractional CMO- —Contactability score ≥ 65
- —OP account age ≥ 90 days
- —Thread must be a self-post, not a link
- —Exclude threads locked by moderators
Cold email failures (product fit)
cold email not workinglow reply ratesinstantly not deliveringapollo reply ratesoutbound dying- —Contactability score ≥ 70
- —Minimum 5 comments on thread
- —OP must have commented back at least once
AI Content Writer workflow
- 1.The Content Writer runs in 'agency-empathy' mode — a custom persona the team configured to match their senior strategist's voice.
- 2.Each draft includes 3 required structural elements: (a) acknowledge the OP's actual problem, (b) offer one specific tactical suggestion without a pitch, (c) leave the door open for a DM only if the OP asks.
- 3.Strategist reviews in an internal dashboard that diffs the AI draft against the final reply — the deltas feed back into the prompt for the next session.
- 4.Rejected drafts get tagged with a reason code (tone, inaccuracy, off-brand) so Priya can audit prompt quality weekly.
Linkeddit MCP + AI integration
Northwind's Slack workspace has a Claude-backed bot named 'Pipe' that runs the Linkeddit MCP and writes to HubSpot. Strategists never leave Slack.
search_leads— used by Pipe every 30 minutesget_lead_insights— on-demand for any lead a strategist flagsget_user_posts + get_user_comments— to assess whether the OP is a real founder or a lurkerfetch_post_comments— reads the full thread so the reply doesn't repeat existing suggestions
- —HubSpot MCP — creates a deal + contact + source thread URL + auto-assigns to the right strategist
- —Slack MCP — posts the lead in a dedicated channel with an 'Accept / Pass' button
- —Notion MCP — drops the thread + draft reply into the client's weekly insights doc
- —Gmail MCP — if the OP publishes their email on Reddit, drafts a warm follow-up (never auto-sends)
You're Pipe. Pull today's leads from the 'agency-looking' pipeline with score ≥ 75. For each: (1) check if the OP's last 20 Reddit comments show they run a company ≥ 10 people, (2) draft a reply in Priya's voice that opens with a specific tactical suggestion, (3) create a HubSpot deal at 'Reddit Inbound' stage with the thread URL, and (4) post in #reddit-leads with an Accept button.
Want to run this workflow yourself? Set up the Linkeddit MCP server or connect via the Claude connector.
Daily rhythm
- 8:30am — Pipe drops 6–10 overnight leads in #reddit-leads with draft replies.
- 9:00–9:45am — Strategist shift: review, edit, post replies from personal Reddit accounts.
- 12:00pm — Pipe's midday sweep, usually 3–5 fresh threads.
- 4:30pm — Priya runs a Claude prompt to audit the day's replies: reply rate, upvote median, any leads gone cold.
Thread breakdown
Subreddits monitored
Results
- —47 inbound discovery calls per month from Reddit alone, up from 0 three months prior.
- —$38k in new MRR signed in quarter one of the Reddit channel.
- —Cold outbound volume reduced 40% with zero pipeline drop.
- —Strategist time-per-reply dropped from 38 minutes (pre-Content-Writer) to 22 minutes average.
Lessons
- 1.Persona calibration is a multi-week investment. The first month's AI drafts were 'too helpful' — not punchy enough. Weekly delta audits fixed it.
- 2.Agency replies from the brand account tanked. Replying from individual strategists' accounts 3.7x'd response rate.
- 3.The Slack-button handoff matters more than it sounds. Review latency dropped from 4 hours to under 30 minutes once the accept flow was one click.
“Our cold email ops room runs 4 SDRs and still loses money. Our Reddit ops room is Pipe plus two part-time strategists and drives more pipeline. The difference is Linkeddit's MCP — Pipe doesn't just find leads, it stages them in HubSpot, drafts the reply, and tells us who to route them to.”
Run the agency ops-room playbook
Same stack, same playbook. Free to start, under 5 minutes to set up.