What AI SDRs Actually Replace in an 8 Person Team
AI handles invites, follow-up timing and first-draft replies. Calls and discovery still need people. Real LinkedIn accept, reply and meeting rates inside.
Are AI SDRs replacing human SDRs, or is this hype?
Both, depending on which part of the job you mean. Nobody has taken an eight person SDR team, switched it off, installed software and kept the same meeting count. What software does reliably now is list building, connection requests, follow-up timing and first-draft replies. Calls, discovery and anything with four stakeholders still need a person.
The loudest version of the claim comes from vendors. Alex Olley has quoted an AI SDR CEO telling him "AI SDRs are going to replace your entire company", which is the pitch, not the product. What is actually gone is the 90 minutes a rep spends each morning sending invites and chasing the day 7 follow-up.
Our published LinkedIn numbers are the part of the job software can hold on its own: 20 to 30% of invites accepted, 12 to 18% of first messages answered, 3 to 5% of invites ending as a booked meeting. Those rates survive 4,000 invites a month with nobody touching the send button. The call that follows the booking is still a person on a headset.
What is $30,000 a month for 8 SDRs actually buying?
That is $3,750 per rep per month. The question is how many meetings come out of it. Eight reps sending 25 LinkedIn invites a day across 20 working days is 4,000 invites a month. At our published meeting rate of 3 to 5% of invites, that is 120 to 200 meetings, or $150 to $250 per meeting.
Run your own number before you buy anything. If you are getting 60 meetings out of 4,000 invites, that is 1.5%, and the gap is targeting or copy or profiles, not headcount. Software will send bad messages faster.
The reply load is the part most teams underestimate. 4,000 invites at a 20 to 30% accept rate is 800 to 1,200 new connections. At a 12 to 18% first-message reply rate, that is 96 to 216 replies, plus another 5 to 8% on follow-ups. Call it 150 to 300 conversations a month that somebody has to read and answer.
Which parts of the SDR job can software actually run today?
The repeatable top of the funnel. Building and enriching the list, sending connection requests at a safe volume, running the follow-up cadence on time, drafting replies in your voice, answering the same six objections from your playbook, and putting a meeting on the calendar. We go through the category in more detail here.
That last loop is the one teams of eight almost never run manually, because nobody wants to open 40 comment profiles by hand. We cover the alternatives for doing it on LinkedIn.
Which parts still need a person?
Calls, for a start. Alex Olley's point is that an AI SDR is not legally allowed to make cold calls in almost every market, which rules out the phone entirely if that is half your pipeline. Then anything where the prospect says something your playbook has never seen.
A prospect who replies "we got burned by a vendor in your space last year and my CFO vetoed the category" is where software stalls. It can book a meeting around that sentence. A rep can ask who the CFO is and call them on Tuesday.
The other place is anything with more than one person on the other side. A security questionnaire, a procurement form and three stakeholders who each want a different answer is four weeks of email that no objection playbook covers. Software gets the first call on the calendar. The next eight touches belong to whoever owns the account.
So what should we automate and what should we keep?
Split by whether the task is the same every time. LinkedIn invites, follow-up timing and the first reply to a known objection are identical across 4,000 prospects. Discovery calls, pricing negotiation and anything involving a procurement team are different every time, and that is where the $3,750 a month earns out.
| Part of the job | Software today | Keep a human |
|---|---|---|
| List building and enrichment | Yes, by filter and signal | Only for named account lists |
| Connection requests at 20-25/day | Yes | No |
| Follow-up cadence | Yes, never misses day 7 | No |
| First reply and known objections | Yes, from your playbook | Spot-check the first 50 |
| Cold calls | No, not legally | Yes, all of it |
| Discovery and qualification | No | Yes |
| Multi-stakeholder enterprise deals | No | Yes |
The "spot-check the first 50" row is the one that decides whether this works. A rep reading the first 50 AI-drafted replies and correcting them is a week of work. Skipping it is how you find out in month three that the software has been pitching the wrong product to the wrong title.
Why are AI companies still hiring SDRs then?
Because reply volume goes up and someone has to work it. Close's read of the data is that sales teams are using AI to raise what each rep gets through rather than to cut heads. If one rep can hold 300 conversations a month instead of 80, the constraint moves from sending to answering, and answering is still a job.
Work it on one rep. 300 conversations a month, at 10 minutes each to read the thread, check the account and write something a human would send, is 50 hours. That is more than a full week of a rep's month before they dial anyone.
TitanX puts it as AI performing parts of the SDR role for some companies and some use cases, which matches what we see. The teams that cut headcount to zero and bought software are the ones who come back in six months.
If we cut from 8 reps to 4 plus software, what happens?
Payroll drops by roughly $15,000 a month at $3,750 a rep. The four remaining reps stop sending invites and spend their day on calls and replies. Your invite volume stays at 4,000 or goes up, because software does not take holiday. Your meeting quality depends entirely on whether the four you kept are the four who can close.
The risk is the handover. If the software books a meeting and the rep opens the thread cold, the prospect notices in the first 30 seconds. Whoever takes the call needs the full message history in front of them before they dial.
Before you change anything, look at the profiles of the people doing the sending. A 20% accept rate and a 30% accept rate on 4,000 invites is a difference of 400 connections, and the headline is a large part of it. The headline rewriter is free and takes two minutes per rep.
Who should not buy an AI SDR?
Anyone whose pipeline comes off the phone. If six of your eight reps are dialling, software covers the two on LinkedIn and leaves $22,500 of payroll untouched. Also anyone with a deal size under $2,000, where 150 extra conversations a month costs more to service than it returns.
Gelee is the wrong tool for a few of these specifically. It runs LinkedIn only, so if your buyers live in email it will not help. There is no self-serve signup, so if you want to swipe a card tonight and send 50 invites tomorrow, buy something else. Setup takes 15 minutes, but it happens on a call.
What would we do with the $30,000?
Keep the reps who can run a discovery call, move the LinkedIn sending and first-reply drafting to software, and measure one number for 60 days: meetings per 1,000 invites. If you were at 30 and you get to 40, the software paid for itself. If it stays at 30, the problem was never the sending.
On 4,000 invites a month, 30 per 1,000 is 120 meetings and 40 per 1,000 is 160. Judge the software on those 40 meetings, not on how many invites it managed to send. Invite count goes up the moment you add accounts, and it tells you nothing about whether the copy worked.
Monday's framing of a hybrid team is roughly where most eight person teams end up, with software on the volume and people on the conversations that need judgement. Pricing for Gelee is shared on a demo call.