Switch from agency outbound to in-house in 14 days

Cancelling a lead-gen agency without dropping pipeline is a sequencing problem, not a tooling problem. The teams that get burned are the ones who cancel before they have proven their own machine works. The teams that get this right are the ones who run both for two weeks, then have a confident conversation.

This is the day-by-day plan we have watched founders use to move off €2,000 to €3,000 a month agencies and replace the output with their own senders, their own messages, and their own data. The fourteen days assume one focused person; double the timeline if you are squeezing this between other work.

Day 1. Inventory what the agency actually delivers

Before you can replace the output, you need to know what it is. Most agency relationships drift over time, and the actual deliverable looks different from the original contract.

Pull the last three monthly reports and write down four numbers: invites sent, accepts, replies, demos booked. Then ask the agency for a list of every account they currently send from. This last one is the conversation that surfaces whether the senders are yours, theirs, or a mix. Most agencies use accounts you do not own; that is a problem you will need to solve in week two.

Days 2 to 3. Connect your own senders

Spin up a Funkel workspace (the seven-day free trial is the actual product, not a feature gate). Connect your own LinkedIn account first. If you have a co-founder or an employee willing to lend their account for outbound, connect theirs as the second sender. The Growth plan includes two senders by default.

New senders sit in a fourteen-day warmup window where the daily cap ramps from 20% to 100%. That ramp is intentional: a brand-new account at full volume looks like automation. The agency is still running for now, so the warmup window does not slow you down.

Day 4. Pick exactly one signal to start with

The mistake here is turning on five signals because you can. Pick one. The one with the cleanest signal-to-message relationship for your category:

• Selling into a category with active competitors: start with competitor_pain.
• Selling to people in transition: start with recent_job_changes.
• Have an audience that engages with your or adjacent content: start with company_followers or competitor_urls.

Running one signal for two weeks gives you a clean baseline. Running three at once means every reply is unattributable and you cannot tell which signal is paying for itself.

Days 5 to 6. Write your ICP and one opener

The agency’s ICP is probably outdated and almost certainly too broad. Use the URL-based ICP draft on app creation as a starting point, then narrow it. Be specific about titles, company size, and regions. Vague ICPs fill the funnel with people who never reply.

For the opener, pick AI mode if you are running a signal that produces variety (competitor_pain, engagement signals). Pick manual mode if you are running a tight cohort (recent_job_changes for one specific role). Either way, write one opener and stop. Do not try to A/B test in week one; you will not have enough volume for the result to mean anything.

Day 7. Soft launch on lower caps

Turn on the agent. Lower the per-sender weekly invite cap to forty for week one. Across two senders, that is eighty invites a week, which gives you enough to read a real signal without putting your account at risk during the warmup.

Leave review mode on for the first three days. Funkel queues every draft and waits for your approval before sending. After you have approved twenty messages and read like the AI is in your voice, switch review mode off.

Days 8 to 10. Watch the numbers, fix the obvious

Three things to watch in the analytics panel:

• Acceptance rate. Below 25% means the opener or the targeting is off. Edit one or the other, not both.
• Lead rejections. If the “Why leads aren’t reaching you” panel shows a high reject rate at the enrichment or score-floor step, your ICP is too narrow or your score floor is set too high.
• Reply quality. Two demos of the wrong fit beats six demos that ghost. If your replies are curious-but-wrong, tighten the ICP. If they are curious-and-right, you are on track.

Days 11 to 14. Scale to your real cap

If the numbers held up through week one, raise the weekly invite cap back to Funkel’s default of eighty per sender. That is one hundred and sixty invites a week across two senders, comparable to what most agencies deliver. If the numbers did not hold up, do not raise volume; fix the opener or the targeting first.

Daily digest email lands in your inbox each morning at 09:00 UTC summarizing the previous day’s accepts, replies, and demos. Skim it; do not let the analytics panel pull you in every hour.

Day 14. Have the cancellation conversation

By day fourteen you have your own data. You know what your senders produce, what your reply rate looks like, and what the cost per demo actually is. Compare it to what the agency delivered in the last three months on the same four numbers from day one.

The conversation:

We are taking outbound in-house starting next month. We appreciate the work, here is the wind-down: please pause any new sends today, finish responding to leads in flight through end of next week, and send us the full contact list and any in-progress conversations by Friday. We will settle the final invoice when that arrives.

Notice what is not in there: an apology, a request for permission, or an explanation. You are not asking. You are sequencing the handover.

What to skip

• Do not cancel the agency on day one. You will lose two weeks of pipeline you cannot afford to lose.
• Do not run more than one signal in week one. Attribution becomes impossible and you will not learn what works.
• Do not connect five senders just because you can. The two included with Growth produce comparable volume to a mid-tier agency at a fraction of the cost.
• Do not skip the warmup. New senders ramped too fast are the leading cause of LinkedIn restrictions.

For the broader case on why the in-house path is worth the two weeks of overlap, see why Funkel.