The short answer
AI outbound fails when it is used to manufacture demand that is not there, because what AI does well is find demand that already exists, gather context about it and get a relevant message in front of it quickly. Pointed at a market that is not asking for anything, the same tools only make the noise louder and faster.
Key points
- The split that matters is demand-driven against supply-driven, because AI multiplies whatever it is aimed at, including a bad idea.
- Volume has stopped paying for itself, with average cold email reply rates falling from 5.1 percent in 2024 to 3.43 percent in 2026 across one platform's own 20 million sends, while carefully targeted campaigns sit several times higher.
- Buyers want to serve themselves and still want a person, since Gartner found 67 percent prefer a rep-free experience1 and, separately, that 69 percent turn to a sales rep to validate what AI told them2.
- AI should not be how someone first hears of you, because being discovered is a different problem from being contacted, and the first real conversation is worth a person.
What is MotiveGTM’s perspective on supply-driven and demand-driven marketing?
Our position is that this distinction decides everything else, because marketing that begins with your own capacity to sell is supply-driven and AI has made it cheap enough to be dangerous, whereas marketing that begins with something the buyer did is demand-driven and is the only kind worth pointing AI at, which is the kind we build.
Supply-driven starts with you, in the sense that you have capacity to sell, a quota to hit and a list you can buy, so messages go out because you need them to.
Demand-driven starts with them, whether that is a company posting three roles for a system you implement, a new VP arriving who has bought this category twice before, or somebody reading your pricing page twice in a week. The trigger sits on their side, which means the message has a reason to exist before anyone writes a word.
The trap is that the same tools serve both, since a model that can research a company and draft a message in four seconds is equally happy doing that for the five accounts with a real reason and the five thousand without one, and it holds no opinion about which of the two it is doing.
That is not a moral position, it is an arithmetic one. Every competitor now has the same lists and the same models, so volume has stopped being an advantage and the scarce thing is a reason to reach out at all.
Why does AI make supply-driven outbound worse?
Because writing speed was never the constraint, and the scarce thing in outbound has always been a reason to talk. Supply-driven programmes were slow before, and that slowness was doing some quiet work, because a person had to spend twenty minutes on an account and somewhere in those twenty minutes they would sometimes notice there was nothing worth saying. Remove the twenty minutes and you remove the last check on whether the message should exist at all.
Three things then happen at once. Everyone in your category buys the same tools and builds the same lists, so the same buyer hears from four vendors in a week in four near-identical messages. Mailbox providers tighten: since February 2024, Google and Yahoo have required bulk senders3 to authenticate properly, offer one-click unsubscribe and keep spam complaints below 0.3 percent, and mail that fails is rejected rather than filed away. And buyers learn the shape of machine-written outreach, so the thing that was supposed to feel personal becomes the clearest signal that nobody was paying attention.
Two questions decide how much trouble you are in: whether anyone is in market for what you sell, and how many others are chasing the same people. Put them on two axes and the shape of the problem is easier to see.
Down the side: how many buyers are in market. Along the bottom: how many vendors are chasing them. Pick a corner.
Nobody looking yet, everyone chasing
The failure zone, and the one most AI outbound lands in. Sending faster raises the number of people who now associate your name with an unwanted message. Stop sending, and spend the quarter becoming findable instead.
When does AI outbound work?
It works when there is demand to find and the work is the finding, which is genuinely hard by hand and genuinely suited to a machine: watching thousands of companies for the handful of public events that suggest something changed, checking a record against several sources so the basics are right, assembling what is known about an account into one place, drafting a first version, and keeping track of what happened afterwards so the next batch is smarter than the last.
None of that is writing a clever email, it is reconnaissance, and done well it means that when a person does reach out they are reaching out to someone who has a reason to reply, at a moment when the reason is still fresh.
Can AI be the first way someone hears about you?
It can be, and it should not be, which becomes obvious as soon as you think about how you would want to meet a supplier you had never heard of. Almost nobody wants that first contact to be an automated message that arrived because a system noticed their job title, since people would rather have come across you already, in a search result, in an answer an assistant gave them, in something a peer mentioned, or in a piece of writing that was useful before anyone asked them for anything.
That is why we treat being found and being contacted as two different problems. If a company is invisible, outbound is being asked to do a job it was never good at, which is to introduce a stranger under time pressure. Fix the being-found part and outbound changes character: it stops being an interruption and becomes a follow-up to something the buyer half-remembers. Our Discoverability work exists for exactly that reason.
What is the human part for?
The human part does two things, one at each end of the process.
Before anything is sent, someone looks at what the system produced and asks the question no model asks well, which is whether this is worth a person's attention at all. That judgment kills most of what gets drafted, and the killing is the point of it.
After someone replies, you reach the part that cannot be handed over, because a buyer who replies is telling you they are willing to spend some of their attention on you, and what they want in return is to be heard, to be taken seriously and to deal with someone who remembers the last conversation.
It is worth saying plainly what happens when that part gets automated, because an AI voice agent that calls a prospect and performs being a person is not saving time so much as making a statement, and the first thing that buyer learns about your company is that you will fake a relationship to save yourself an hour. That is not a technology problem but a positioning one, and it is expensive in a way no dashboard will ever show you.
None of this is a B2B quirk either, since nobody buying anything enjoys being handled, and the only difference in B2B is that the deals are larger and the memories are longer.
Illustration
Two teams sell the same integration service in the same quarter, and they use the same tools to do it.
The first sends four thousand messages, with competent copy and a job-title filter for targeting, and lands a reply rate of about one percent, most of those replies negative, two of them angry, and the sending domain in need of repair afterwards.
The second points the same tools at one thing, companies posting a second job for the same system inside a quarter, which produces about twelve companies a month. A person reads all twelve, discards half, sends six messages that took ninety seconds each to finish, and gets three conversations out of them.
The difference is not the model or the copy, it is that the second team only wrote to people who had done something first.
Does this mean AI SDRs are a bad idea?
Not as a category, though very much as a replacement for the relationship. An AI SDR that researches accounts, monitors signals, builds the list, drafts openers and keeps the CRM honest is doing work that was never worth a person's day. An AI SDR that carries the conversation, handles the objection and sits in the middle of a new relationship is being asked to do the one part of the job that only exists because a human being is on the other side.
The published comparisons that favour one side or the other are mostly written by vendors selling that side, so the numbers deserve some caution, whereas the buyer research is harder to argue with: people are content to serve themselves for most of the process and then want a person at the moment the decision gets real.
How do you know which mode you are in?
Four questions, answered honestly:
- Can you name what happened at a target account in the last thirty days that makes this a good week to write to them, and if the honest answer is their job title, are you willing to call that supply-driven?
- If you paused all outbound for a month, would anybody find you, and if they would not, is the problem really send volume rather than discoverability?
- Would you be comfortable if the person on the other side saw exactly how their name got onto your list?
- When someone replies, how many minutes pass before a person who can help them is involved?
The uncomfortable one is usually the second, because outbound is the easiest thing to buy and the hardest thing to buy your way out of.
Sources
Figures last checked on 19 September 2026. Buyer research and platform benchmarks both move, so we revisit them quarterly.
- 1 Gartner, “67% of B2B Buyers Prefer a Rep-Free Experience”. March 2026.
- 2 Gartner, “69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights”. May 2026.
- 3 Google, “Email sender guidelines”. Authentication, one-click unsubscribe and the spam complaint threshold for bulk senders.
- 4 Instantly, “Cold Email Benchmark Report 2026”. Reply rate trend. Worth noting this is a sending platform reporting on its own users, not independent research.