Here's the trap. You decide to start outbound, so you go looking for a stack, and every list you find was written by somebody who sells one of the tools on it. The recommended setup comes to $400 a month before you have sent a single message.
You can run a complete outbound motion for $17 a month. Free LinkedIn search, Claude Pro, a free Notion database, and your own hands on the send button.
It works to about fifty messages a week. Then it stops, for reasons worth understanding before you start rather than after.
What it costs, actually
| Line | Cost | Verified |
|---|---|---|
| LinkedIn, free tier | $0 | n/a |
| Claude Pro, annual | $17/mo ($200 up front) | 21 Aug 2026 |
| Claude Pro, monthly | $20/mo | 21 Aug 2026 |
| Notion, free tier | $0 | 12 Aug 2026 |
| Total | $17 to $20/mo |
Note what is not on that list. Sales Navigator is $119.99, which breaks a hundred-dollar budget on its own. That exclusion is not a compromise, it is the design: free LinkedIn search is more limited, and those limits are what keep the volume at a level a human can personalise anyway.
The seven steps
Nothing here automates LinkedIn. That is the point, and it is the difference between this and the tools that got people restricted.
- Boolean search on free LinkedIn. Build one query for your ICP using AND, OR and NOT, filtered to 2nd and 3rd-degree connections. Save the search. This is the step worth spending an hour on, because everything downstream inherits its quality.
- Copy the visible results into Claude. Manually, from the page. Name, title, company, and whatever the profile shows. Twenty-five at a time is comfortable.
- Have Claude structure them into Notion. Connect the Notion database and ask for one row per prospect with your fields. You are using the model as a parser, which is the thing it is quietly best at.
- Filter in the prompt, not by hand. Give Claude your disqualifying rules directly: company size, title seniority, industries you do not sell to. Let it mark rows rather than deleting them, so you can audit what it dropped.
- Feed it your value proposition and your actual tone. Paste two emails you have written that worked. Tone transfers far better from examples than from adjectives.
- Generate one message per prospect, then read every one. This is not optional. The failure mode of AI outreach is a message that is fluent, specific and wrong, and it looks fine until the recipient reads it.
- Send by hand. Copy, paste, send. Slow, and the slowness is a feature: it is the last checkpoint before something with your name on it reaches a stranger.
Why not just automate it?
Because the automation layer is what LinkedIn enforces against, and the enforcement lands on your account rather than on the vendor.
In March 2026 LinkedIn removed HeyReach's company page and restricted its CEO, CTO, CRO and CMO profiles. Reporting attributes the action to the tool's cloud-proxy architecture, which LinkedIn treats as policy-violating regardless of whether an individual user stays inside daily limits. The product carried on working. The profiles did not.
A widely quoted figure puts around 40% of accounts on non-compliant automation as restricted in the first quarter of 2026. Treat that as an estimate. LinkedIn publishes no enforcement statistics at all, so anybody quoting a precise ban rate is guessing, including whoever produced that number.
What is not in dispute is the asymmetry. A restricted founder profile takes the pipeline down with it, and recovery runs through identity verification that takes days. Against a $17 stack, that is a poor trade for saving an hour a week.
Where it breaks
Three ceilings, and you hit them in this order.
Roughly 100 connection requests a week is the working limit practitioners have converged on, and it has been stable since about 2022. That is consensus rather than a published figure, so treat it as a soft ceiling rather than a rule.
Around fifty personalised messages a week is where most people actually stop, and it arrives before the platform limit. Reading and sending fifty messages properly is most of a working day spread across a week.
Then the motion itself. Once you know the message works, wanting more of it is reasonable, and more of it is a volume problem rather than a quality one. That is the moment to spend money, and not before.
What does the next step actually cost?
More than people expect, and it is worth seeing before you decide the manual path is beneath you.
Moving to email at volume means a sequencer, sending domains and mailboxes. We itemised that separately: the floor to send 6,000 emails a month is about $73.67, of which the software is barely half. That is four times the manual stack for a motion you have not yet proven.
Which is the argument for doing it by hand first. Fifty messages a week for a month tells you whether the offer lands, and it costs $17 to find out. Nothing you buy afterwards fixes an offer that does not.
So can you run outbound for under $100?
Comfortably, at $17 to $20, and the constraint that makes it work is the one that looks like a limitation. No Sales Navigator means narrower search. Narrower search means fewer prospects. Fewer prospects means you can actually read what you are about to send.
The stack is not a cheaper version of the expensive one. It is a different motion, and at fifty messages a week it is usually the better one, because at that volume personalisation is real rather than a merge field.
Buy the sending capacity when you have something worth sending at scale. Until then the expensive stack is solving a problem you do not have yet.