AI SDR
AI reply handling: the 12 reply types and the right response to each
By Viktorio Shopov, co-founder of Sellinger. Reviewed by Asen Levov. Published . Last updated . 9 minute read.
Most replies fall into twelve types: ready to book, asks what it is, asks the price, wants information, not now, wrong person, already has a solution, sceptical, refusal, opt out, machine mail and needs a person. Each has a right response, a wrong one and an owner. The agent handles the routine ones, and a person takes price, complaints and anything the agent has no fact for.
Deciding in advance how each type is handled is the difference between an agent that feels competent and one that stumbles on the third message. The sample replies below are illustrations of the pattern, written for this post. They are not quotes from a campaign.
1. Ready to book
"Sure, send me a link."
Right: a link or two times, one line, no extra pitch. Wrong: a paragraph of features. Owner: the agent. Nudge once if the link is not used. On our account 56% of bookings confirmed in the chat needed one more nudge.
2. Asks what it is
"What do you actually do?"
Right: one or two sentences with the answer first, tied to their words, then an offer of a short look. Wrong: "Great question!" and a brochure. Owner: the agent. The booking link on our account typically went out after three prospect messages, so answering well earns the offer.
3. Asks the price
"How much is it?"
Right: state a price only if you have given the agent one. Otherwise say it depends on the setup and offer a short call, or hand over. Wrong: inventing a number. Owner: the agent if it has a public price, otherwise a person. See should an AI agent give a price quote or stall.
4. Wants information
"Send me some info."
Right: one short, relevant link or paragraph plus one question that moves things forward. Wrong: an attachment dump. Owner: the agent.
5. Not now
"Not a priority right now, maybe next quarter."
Right: accept it, ask when to come back or say you will check in at that time, and stop. Wrong: pushing. Owner: the agent. Set a reminder in the CRM rather than relying on memory. In one of our threads a prospect rebooked himself 20 days after the agent let go gracefully.
6. Wrong person
"I do not handle this, try our head of operations."
Right: thank them, ask for the right name if they did not give it, and write to that person. Wrong: ignoring the referral. Owner: the agent. In the Lamothermic study 12 of the 42 buyer side outcomes were referrals to the right buyer, so this type is worth a good answer.
7. Already has a solution
"We already use another tool."
Right: acknowledge, ask what they would change about it, and leave it there if nothing comes. Wrong: a feature comparison. Owner: the agent. See the comparison pages for how tools differ, but do not argue in a thread.
8. Sceptical
"Is this just another automated sales pitch?"
Right: a short, honest answer about what it is and why you are writing, with a specific reason. Wrong: defensiveness or a long defence. Owner: the agent for the first reply, a person if the doubt continues. One more fact helps more than reassurance.
9. Refusal
"Not interested."
Right: a one line thanks and a clean end. Wrong: a follow up. Owner: the agent. Refusals rise with each touch: on our own account the share of first replies that were negative was 0.9% at the intro and 14.7% from the third follow up. That is one reason to cap follow ups (how many follow ups).
10. Opt out
"Stop contacting me."
Right: stop, record it, confirm in one line if the channel allows. Wrong: one more message. Owner: the system, then a person for anything unclear. In Sellinger a reply such as STOP or unsubscribe, in English, Bulgarian or German among other forms, puts the contact on the team's suppression list, the agent stops and the lead is closed out. A weak or unclear phrase never writes the permanent record: the agent stops and a person decides. See opt outs and suppression.
11. Machine mail
An out of office message, a helpdesk ticket acknowledgement, a satisfaction survey, a no reply sender.
Right: do not count it as a human reply and do not answer it. Wrong: treating "we received your message" as interest, or answering a robot. Owner: the system. In Sellinger these are marked as machine mail, never counted as a human reply, and the lead goes to a person with a named reason. See keeping AI agents out of machine mail.
12. Needs a person
"This is wrong, I want to speak to someone." Or a complaint. Or a question the agent has no fact for.
Right: stop, flag, and give the person the history and the reason. Wrong: a confident guess. Owner: a person. Takeover is one click, and the agent resumes from the full history on release (take over a conversation).
What does the mix look like in a real study?
The Lamothermic case study counted what 293 replies turned into. Of 293 replies, 42 were buyer side outcomes: 23 asked for a quote, information or an NDA, 12 referred the agent to the right buyer, 6 agreed to a call and 1 booked a call. Map them to the list above. Quote, information and NDA requests are types 3 and 4. The 12 referrals are type 6. The 6 who agreed to a call are type 1. The rest sit mostly in the polite types: not now, refusal, courtesy.
The practical reading is that the types you handle best are rarely the ones that arrive most often. Type 6 produced almost as many outcomes as type 1 there, and many setups treat a referral as a dead end. Check your own sample for the same pattern.
A made up example: sorting ten replies
The ten replies below are invented for this post. Sort them as the agent should.
| Reply | Type | Owner |
|---|---|---|
| "Sure, send me a link." | 1 Ready to book | Agent |
| "What does your tool do?" | 2 Asks what it is | Agent |
| "How much per seat?" (no price given to the agent) | 3 Asks the price | Person |
| "Not a priority until Q2." | 5 Not now | Agent |
| "Talk to Maria, she runs sales ops." | 6 Wrong person | Agent |
| "We use another vendor." | 7 Already has a solution | Agent |
| "Is this a bot?" | 8 Sceptical | Agent, then a person if it continues |
| "Not interested." | 9 Refusal | Agent, one line |
| "Out of office until the 12th." | 11 Machine mail | System |
| "This is wrong, call me." | 12 Needs a person | Person |
Three of ten go to a person or to the system. The point of the table is that every reply has an owner before it arrives.
Which types get confused?
- Not now and refusal. "Not now" invites a later check in; "not interested" does not. Treat an unclear one as not now only if the wording allows it.
- Sceptical and refusal. "Is this just a pitch?" is a question. Answer it. "Please stop pitching me" is an opt out.
- Courtesy and interest. "Thanks for connecting" is neither. It needs one question to find out.
- Out of office and a real reply. An out of office message holds a return date but no interest. Do not count it, and do not answer it.
When two types fit, pick the one with the more careful response: a person before the agent, a stop before a follow up.
How do you use the list?
- Copy the twelve types into your campaign rules.
- For each, write the response you want and who owns it.
- Mark the types that always need a person: 3 when you have no price, 8 if the doubt continues, 10 when unclear, 12.
- Sample twenty threads a week and check each reply against its type.
- Add types for your market. A hotel group will see "is this room available" and a foundry will see "attached is our drawing". They are your thirteenth and fourteenth.
The rule packs in the template library give you a starting point for the written rules, such as the inbound lead handling pack and the B2B SDR pack. For what happens after the first reply in more detail, read what an AI SDR does after the first reply.
Common questions
How many reply types are there?
Which replies need a human?
What counts as a real human reply?
Should an AI answer every reply?
How Sellinger helps
Read how it works on See the reply rules in a template, or see a real campaign in Read the Lamothermic case study.
