My own experiments suggest that this is the wrong place to apply it. The logic sounds intuitive. If AI can understand my ICP, what my business offers, and whether a prospect may need it, why not let AI talk to the prospect? Three months ago, I redesigned my outbound system for Fit To Scale. I iterated in steps and observed what actually moved response rates.
I started with the classic Apollo setup: filter for your ICP, create customised email templates, and run a one-touch campaign. Three years ago, this type of campaign produced roughly a 3% positive response rate for me. This time, it produced effectively zero. Next, I added AI.
AI verified the Apollo leads, collected GTM evidence, scored and selected leads, and wrote a custom message pointing directly to that evidence. I sent the emails manually at first to keep control over what the AI produced. The positive response rate increased to around 1%. But a second test was more interesting.
The improvement was not an attribute of better AI writing. It came from a higher-quality lead list and a better understanding of my target group through the GTM evidence. I achieved the same result with templated emails while keeping AI for research, analysis, and routing. Then I went multi-channel and multi-touchpoint, adding value with each new communication attempt. LinkedIn became the second channel.
Positive responses increased to roughly 3%. Email added almost nothing to the overall result. In my campaigns, it has become a dead horse for communicating with CEOs. Then came the biggest improvement.
The last iteration increased positive responses from 3% to 10-14% across my two best-performing campaigns. And both campaigns are entirely pre-scripted. The difference is not the message.
It is who enters the campaign.
Each campaign is optimised around a specific market signal:
1. Hiring for GTM while the current sales team consists of the founder plus up to four entry-level sales positions
2. A Seed or Series A funding round within the previous four weeks
To provide the leads, I have two agentic workflows running periodically and feeding matching companies into the campaigns. This changed the economics of outbound for me. I do not have to send hundreds of messages every week. A handful of precisely targeted ones can be enough to keep the pipeline full.
Sender reputation stays intact. Credit and tool costs stay low. Messaging stays under control. And there is no AI slop, no logical incoherence, and no vague personalisation pretending to be relevance. The useful role of AI in outbound, at least in my tests, is not to have more conversations for you.
It is to decide who should enter a conversation, when, and why. Which signals do produce the best results for your company, what do you do to source them, and what does it cost you?