Volume spends the market

A study published this spring paired 100,000 cold emails, half written by AI sales agents and half by people, and followed them for fourteen days. The AI emails got a 4.1 percent reply rate against 5.2, booked a meeting in 0.7 percent of cases against 1.1, and were flagged as spam 8 percent of the time against 3. None of that stops AI outreach from winning on a volume slide, because an agent can send many times as much. Read by someone whose buyers are a few thousand executives in one region, the same numbers describe how to use up a market.

What one meeting uses up

Per booked meeting, the study's rates translate into a count of people. At 0.7 percent, an AI agent needs about 14,300 emails for 100 meetings. At 1.1 percent, a person needs about 9,100. At 8 and 3 percent spam flags, that is roughly 1,140 spam reports against 270 for the same 100 meetings. And about 13,700 recipients who got something and did not answer, against 8,600.

Agents are sold on activity: ten to fifty times what a person sends, according to one industry overview. That makes the monthly meeting count go up. It does not change what each meeting cost in people.

In a market of millions of small businesses, those people are noise. In a market you can count, they are most of the market. As a worked example, assume 2,000 relevant decision makers in a region. Getting 100 meetings the agent way means sending each of them about seven emails, and the spam reports along the way add up to more than half the size of the market. The volume model quietly assumes the list is bigger than the sending.

Reputation is shared

Every spam report is a vote on the sender, and it does not stay with the campaign that earned it. It sits on the domain and the sender name, and those also carry the invoice reminder, the event invitation and the reply to a warm referral. This week a sales agency reported the same on the phone side: its numbers now get flagged as spam almost daily, where it used to take a few thousand dials. That is a single report from the US.

The study adds a detail that matters for anyone running sequences: emails sent a day apart reached the inbox 71 percent of the time, three days apart 93 percent. Speed is paid in placement. And the starting point is not generous. Validity's benchmark for 2025 puts average inbox placement at 87.2 percent, Microsoft at 77.4, and outside the US Microsoft is the second-largest mailbox provider in most markets.

What to measure instead

When the buyers are a few thousand senior people, I would put three numbers next to meetings per month:

• Market size. How many people could ever buy, and how many are already on a list. If nobody can answer that, the campaign has no denominator.
• Contacts spent. Emailed without reply, unsubscribed, marked as spam. Counted per quarter and compared with the market size, it tells you how long the market lasts at the current pace.
• Touches per person. A cap per contact and quarter, set before the campaign starts, so the sequence tool cannot decide it for you.

Where the agent belongs

AI has a good place in outreach to a small market, and it sits before the send button. It pays on the research side: noticing that a company changed its CEO, restructured or announced a succession, and preparing a message that refers to it. That moves the reply rate and leaves the volume where it was.

The sending should stay slow, personal where it can be, and in the channels that already carry trust: a podcast, a webinar, a referral, a newsletter someone chose to receive. They produce fewer meetings per month. They do not consume the people they did not convert.