The 8 Questions That Separate Real AI SDR Vendors From Hype
Ask AI SDR vendors for connection accept, reply and meeting rates with denominators, plus 90-day total cost. Real benchmarks and disqualifiers you can use today.
What is the first question to ask an AI SDR vendor?
Ask this: "What is your connection accept rate, your reply rate and your meeting rate, and out of how many invites?" Every number needs a denominator. Vendor marketing likes claims like 10x outreach and 50% lower cost per meeting, which tell you nothing about how many invites produced one meeting.
Here are ours, so you have something to hold theirs against. Connection accept 20 to 30%. First-message reply 12 to 18%. Follow-up reply 5 to 8%. Overall meeting rate 3 to 5% of invites sent.
Work it through on one warm account at 25 invites a day. That is about 500 invites a month, which gives 100 to 150 accepted connections, and 15 to 25 booked meetings. If a vendor tells you 60 meetings a month from a single LinkedIn account, they are claiming a 12% meeting rate on invites, and the next question is what volume that took.
Disqualify: any vendor who answers with "it depends on your ICP" and never comes back with a figure.
What should I ask about price so nothing shows up later?
Ask for the total cost of the first 90 days in one number: subscription, onboarding fee, seats, LinkedIn accounts, and contact data if it is not included. Then ask what triggers an overage. Then ask whether there is an onboarding fee, what it covers, and how many of your own hours it replaces.
The arithmetic matters more than the sticker. A $1,000 a month tool with a $1,500 onboarding fee is $4,500 across 90 days. If month one is ramping and you get 40 meetings across months two and three, that is $112 a meeting. A $600 a month tool that needs 15 hours of your month writing copy and clearing an approval queue costs $600 plus 15 hours, and at $100 an hour that is $2,100.
Gelee has no self-serve signup and pricing is shared on a demo call.
Disqualify: a vendor who quotes a monthly figure and will not put the setup fee, the seat count and the overage trigger in the same email.
Who writes the replies, and who answers them at 11pm?
Ask whether the AI sends replies or drafts them for approval, and get the answer as a workflow rather than a feature name. Both are legitimate products. Only one of them frees up your evening. If it drafts, ask how many drafts a week you should expect to review at 500 invites a month.
Three follow-ups worth asking in the same breath:
If the answer to all three is "you", you are buying a sending tool with an AI label, which may still be the right purchase. The four categories of LinkedIn outreach tool split roughly along this line, and the price difference between them is mostly the reply handling.
What should I ask about whose voice it writes in?
Ask what they ingest before the first message goes out. SaaStr's framing is the useful one: the goal is to clone the best person on your team, and if it is just you, clone you. So ask which of your artefacts they take: sent DMs, call recordings, your objection responses, your posts.
Gelee's setup runs about 15 minutes and takes your writing plus your objection playbook. That is short because the inputs are narrow, and a vendor who wants three weeks of onboarding is not automatically worse, they are just ingesting more.
Disqualify: "our model is trained on 100 million cold emails" with no answer about your own messages. Trained on everybody's outbound means it writes like everybody's outbound.
What should I ask about my data and my LinkedIn account?
Ask three things. Does the vendor use your prospect data and your message history to improve the model other customers use. Where does your LinkedIn session live, and is it a dedicated IP. What happens to your conversations if you cancel on day 40.
The first one is the one people forget. Vendors do train on customer data, and it can produce leakage, where your information surfaces elsewhere. Get the answer in writing.
The third one decides whether you can leave. If your reply threads live inside their inbox and export is a CSV of message text with no thread structure, you are rebuilding your pipeline by hand.
Disqualify: a vendor who cannot say yes or no to "do my messages train your shared model".
How do I check the volume they promise is actually safe?
Ask for their daily invite cap and their ramp schedule for a brand new account. Safe on a warm, aged account is 20 to 25 invites a day. A new account should sit at 5 to 10 a day for the first two weeks. LinkedIn does not publish a hard limit, and the real ceiling moves with account age and how people respond to you.
Anyone offering 100 invites a day from one account is offering a restriction. If you need 2,000 invites a month, that is four or five accounts, and the question becomes what they charge per account.
While you are asking, note that accept rate is mostly your profile, not their copy. A 12% accept rate on a headline that says "Founder | Building cool things" is not a software problem, and the headline rewriter fixes that in five minutes.
What should I ask before signing a 12-month contract?
Ask for the exit. Specifically: is there a 30 or 60 day pilot, what happens at month four if the meeting rate is 1% instead of 3 to 5%, and who owns the sequences you wrote. Contracts get cancelled when the decision was made on budget rather than fit.
A pilot with a number attached is better than a pilot. Write down the figure you will judge it on before you start. Something like: 500 invites in 30 days, at least 100 accepted, at least 12 replies, at least 5 meetings booked. If the vendor will not agree that those are reasonable, that is information.
Which answers should disqualify a vendor on the spot?
Six answers end the call. "It depends on your ICP" with no number after it. "100 invites a day, no problem." A monthly price with no setup fee attached. No yes or no on whether your messages train their shared model. "The AI handles everything." And "contracts are annual" when you ask about month two.
| Question to ask | Answer that ends it | What a good answer sounds like |
|---|---|---|
| What is your meeting rate per invite? | "Depends on your market" | "3 to 5% of invites, here are the campaigns" |
| Daily invite volume per account? | "100 a day, no problem" | "20 to 25 warm, 5 to 10 for the first two weeks" |
| Total cost for 90 days? | Monthly price only | Subscription plus setup plus seats, in writing |
| Do my messages train your model? | Cannot answer | A yes or no, and where it is documented |
| Who answers the replies? | "The AI handles everything" | Which replies it sends and which it queues |
| What if I cancel in month two? | "Contracts are annual" | Notice period and an export you can read |
The middle column is not a hypothetical set. Every one of those has been said on a sales call by somebody selling outbound software.
Who should skip this checklist and not buy an AI SDR?
If your average deal is under $2,000, or your buyers are not on LinkedIn, or you have not yet closed anything at your current price, more conversations will produce more no. Buy none of these. If you already have two SDRs who are booking meetings, buy them a better sending tool and keep the humans on replies.
Two more cases where Gelee is the wrong purchase. If you want to enter a card tonight and be sending by lunchtime, Gelee has no self-serve signup, and the tools sorted by the job you need done covers the ones that do. And if you are a RevOps lead buying for twelve reps with Salesforce as the system of record, the first question is CRM-native agent or standalone tool, which is covered properly here and is not a question we are the right answer to.
Where we do fit: one to five LinkedIn accounts, a founder or small team, buyers who reply on LinkedIn, and nobody with time to answer at 11pm. Ask us the six questions in the table. The Post Commenters preset, which DMs everyone who comments on a watched post and then replies publicly once the DM has sent, is the one thing on the list most vendors will not have an answer for.