Most case studies are written by the marketing department six months after the fact. This one is written from the campaign logs. In May we ran an AI Sales campaign for a recruitment firm on the founder's own LinkedIn account. Twelve days later it had started 521 conversations, surfaced 52 interested prospects, and put 6 meetings in the diary. Here is exactly what happened, including the parts that did not go to plan.
The setup, and why it took six days
The unglamorous truth about AI outbound is that the campaign is decided before the first message sends. We spent the first six days on three things.
First, the list. We pulled a Sales Navigator search of hiring decision-makers in the firm's exact niche and ran every prospect through AI scoring before they entered the campaign: does this title own the hiring decision, is the company in the size band that actually buys contingent recruitment, is there a timing signal like live vacancies. Roughly a third of the raw search got cut. That cut is the campaign. Every deleted prospect is a connection request that did not get wasted on someone who could never buy.
Second, the voice. The AI drafts every message, but it drafts in the founder's voice, trained on how he actually writes: short, direct, no exclamation marks, no "I hope this finds you well". If the messages read like a bot, the acceptance rate tells you within 48 hours.
Third, the approval loop. Nothing sends without the founder tapping approve. That is not a compliance checkbox, it is the quality mechanism. In the first two days he rejected about one draft in ten, the AI learned from what he changed, and by day five the rejection rate was near zero.
Days 7 to 12: what the numbers actually looked like
Campaign live on day 7. The rhythm from there:
| Metric | Result over 12 days |
|---|---|
| Connection requests sent | ~1,100 |
| Connections accepted | 33% |
| Conversations started | 521 |
| Replies to first message | 56% |
| Interested prospects tagged | 52 |
| Meetings booked | 6 |
Two numbers matter more than the rest. The 33% acceptance rate told us the list and the profile positioning were right, because cold acceptance on a generic list sits nearer 15 to 20%. And the 56% response rate told us the opening message was doing its job, because it was written around the prospect's situation, not the firm's pitch.
What went wrong, because something always does
Two things, and they are instructive.
One prospect turned hostile. Not lukewarm, hostile: they objected to automated outreach on principle and said so at length. The system's hard-stop rule fired, the conversation was paused permanently, and no follow-up of any kind went out. This is the part of AI outbound nobody talks about: the machine has to know when to stop more reliably than a human SDR does, because at this volume one badly handled objection is a public reputation problem.
And LinkedIn throttled the pace twice. The platform moves its internal limits around, and a campaign that pushes daily volume too hard gets its sending slowed. The system respects the throttle, backs off automatically, and the campaign loses a day. If a vendor tells you their tool never hits rate limits, they are either sending too little to matter or lying.
The maths that makes it a business case
Six meetings from 12 days is the headline, but the pipeline behind it is the asset. The 52 interested prospects who did not book immediately went into structured email follow-up, and interested-but-busy prospects convert for months afterwards. For a recruitment firm where one placed role is worth £8,000 to £15,000, one client from this campaign pays for the system for a year. That is the whole ROI conversation.
What the AI actually did, listed plainly
- Scored and cut the prospect list before a single request went out
- Drafted every connection note and every reply in the founder's voice
- Held every message for human approval, and learned from every edit
- Tagged intent on every reply: interested, not interested, question, objection
- Moved interested prospects into email follow-up with enriched addresses
- Stopped dead on the one conversation that demanded it
The founder's job was approvals over coffee and taking the calls. That is the honest division of labour, and it is why this works at £499 a month while an SDR costs £3,000.
If you want the same engine on your own account: AI Sales, live within 7 days, and we owe you 1 client in 90 days or it's free until you get one.