AI changed the cost of sending a personalised message to nearly zero. It did not change what makes someone agree to a meeting. Here is where the line falls.
Short answer: AI genuinely replaces the mechanical half of an SDR's job — research at volume, drafting, sequencing, follow-up discipline. It does not replace the judgement half: choosing the target, deciding what the offer should say, reading ambiguity in a reply, and knowing when the market is telling you no. Teams that treat AI as a productivity multiplier for a person do well. Teams that treat it as a replacement for the person end up with a very efficient way to annoy their market.
Research at volume. A human researching a prospect properly takes several minutes. A model does a usable version in seconds across thousands of records. This is a genuine order-of-magnitude change and it is the reason the category exists.
Consistency. Follow-up four is where most human sequences quietly stop, and follow-up four is where a meaningful share of replies come from. Software does not get bored or discouraged.
Drafting. A competent model writes a better first-draft cold email than most junior reps, immediately, in any tone you specify.
Never having a bad week. No ramp, no dip after a rough quarter, no notice period.
Choosing who to contact. This is a strategy decision informed by where you win, which deals were painful, and which segment renews. That lives in your head and your CRM history, not in a signal feed.
Deciding what the offer should be. When a market does not respond, the answer is usually the proposition, not the subject line. Recognising that requires knowing what else the business could credibly promise.
Reading ambiguity. "Send me something in Q1" means very different things depending on tone, seniority and what else they said. Humans read that well. Models read it literally.
Knowing when to stop. A model asked to book meetings will keep trying. A good rep says "this segment is wrong, and here is what I would try instead."
Making bad outreach cheaper does not make it better; it makes there be more of it.
The failure pattern looks like this. A team buys a tool, points it at a broad list, and sends at volume. Reply rates are poor but volume compensates, so it looks like it is working. Meanwhile the domain reputation degrades, the addressable market gets burned, and a year later the same company cannot reach those accounts even with a good message, because everyone there has already deleted three of theirs.
The market has a memory. Efficiency spends it faster.
The version that works in practice is neither pure AI nor pure human. It is:
Our own campaigns run exactly this way. The AI-personalised outreach is what makes small, precise lists economic; the judgement about who goes on the list and what it should say is not automated, and I would not trust it if it were. Across our client base that combination averages 33+ qualified sales calls every 30 days.
The most instructive case study we have is PR Noir: 27 leads, 10 qualified meetings, one $90,000 deal in the first 30 days. The number people react to is the $90,000. The number that explains it is the 27. At twenty-seven contacts there is no room for a poor hit rate to be rescued by volume — which forces all the work upstream, into the human half.
"What happens when the market does not respond?" You want a diagnosis, not more volume.
"Who reads the replies, and how fast?" If the answer is a model with no human in the loop, ask what happens when someone replies with something the model has not seen before.
"How do you protect my sending reputation?" If they cannot describe warm-up, volume ramping and list verification without prompting, the efficiency they are selling comes out of your domain's future.
Cost is a separate question. A tool and a person and an agency have different cost shapes; efficiency alone does not tell you which is cheapest for your volume. That comparison is here.
Your market may be an outlier. PR services buyers are identifiable and used to being approached. A nine-month enterprise cycle into a market of 200 accounts behaves differently, and in that world the human half matters even more.
Models keep improving. The line I have drawn between mechanical and judgement work is where it sits today. I would expect it to move. I would not expect it to disappear, because choosing what to say to a market is a business decision, not a text-generation problem.
If you want to talk through where that line falls for your market, book 20 minutes or email moe@roimaxi.com.
Moe Alhosni is the founder of ROI Maximizer (M Ventures LTD, 4 Beau Street, Bath BA1 1QY). ROI Maximizer works on a pay-per-result basis — clients pay for booked, qualified meetings, not retainers.