The AI Edge

The Briefing

Five things from the past week that actually matter if you sell stuff.

1. OpenAI quietly shipped its Ads Manager. A pilot of advertisers can now place ads inside ChatGPT through a Google Ads-style dashboard, with live impression and click data and no agency in the middle. If your buyers research with AI before they buy, the front door just opened.

2. Anthropic dropped 10 finance-specific Claude agents this week. Pitch builder, earnings reviewer, KYC screener, and seven more, with Goldman and Blackstone as integration partners. None are marketing-specific yet. But the pattern is the signal: vertical agents shipped straight into a workflow stack. Marketing is next.

3. Claude Memory for Managed Agents went into public beta. File system-based memories that persist across sessions, with API control and audit logs. If you run long-relationship work — coaching, consulting, anything where context compounds — this is the piece that was missing. An agent that actually remembers your client.

4. Meta says Advantage+ AI is now lifting ROAS up to 22%. Their new GEM model is boosting conversions 5% on Instagram and 3% on Facebook. The tradeoff: more brands feeding the same machine means sameness. The edge is no longer access. The edge is voice and timing.

5. Google Marketing Live is May 20. Mark the date. Every signal points to AI-led ad formats and AI Mode in Search going from beta to default. Do not lock Q3 budgets before that keynote.

The Build

Most marketers are using AI to remove themselves from the work. This week I spent doing the opposite.

Here is what I built and why it matters.

I've been deep inside my high-end clients subscriber lists — using agents to do something humans physically cannot. Not generate copy. Not write hooks. Not "scale content." Something far more useful: find the people who are stuck.

If you run a funnel of any kind, you know the truth. People do not move through your sequences in straight lines. They opt in, then sit. They click email three of seven, then ghost. They book a strategy call, then cancel the night before. They watch 73% of a VSL and bounce on the order form. Each of those moments is a sticking point. Each one is a silent question.

For 15 years I have known this was where the gold was. For 15 years I have not had the bandwidth to act on it in real time. Nobody does. You cannot watch a list of 40,000 people and personally notice when one of them hesitates on the booking page at 11pm on a Tuesday.

An agent can.

This week I built a sticking-point detection agent. It runs against subscriber activity data, segments by funnel stage, and flags every prospect who shows the behavioural signature of almost. Almost booked. Almost bought. Almost replied. And here is the part most people get wrong: it does not auto-message anyone. It surfaces them to me, with a one-line read of what they got stuck on and a draft opener I can send by hand.

Here is the prompt running 24/7:

You are my subscriber intelligence agent. Loaded:
- Audience Intelligence (psychographics, pain points)
- Voice DNA (so any draft sounds like me)
- Funnel map (every step of every active funnel)
- Last 3 days of subscriber activity data

Your job is to find the "almost" people.
For each funnel stage, identify subscribers who:

1. Showed high intent in the last 14 days
   (multiple opens, deep scroll, repeat visits)
2. Stopped one step short of the conversion
   (booked but didn't show, watched but didn't buy,
   replied but didn't follow through)
3. Have behavioural data pointing to a specific
   sticking point (price, timing, fit, doubt)

Output a daily list, max 12 people.
Each entry:
  Name | Funnel stage | Inferred sticking point |
  One opening line for me to send, written in my
  voice, that opens the conversation without selling.

Bias toward intimacy, not closing. The goal is to
re-open the door, not push them through it.

What worked: The bookings climbed almost immediately. Not because the agent was selling — it never sells. Because the messages I sent landed at the exact moment those people were already thinking about the question. Someone bounces on the order form at 11pm Tuesday. By Wednesday morning a message sits in their inbox in my voice: "I noticed you got close to the order page last night. I am not chasing the sale. I am just wondering what you got stuck on — was it the price, the timing, or something I have not made clear yet?"

People reply. People reply fast. Some book the call right there. Others give me information I would have paid a researcher to surface. Either way, the connection is real, because the timing is real.

What didn't work: The first version tried to fully automate the send. I pulled that back inside a week. The whole reason this works is that a human is at the other end. The moment a subscriber sensed they were being pinged by a bot, the magic vanished.

AI does the watching, the inferring, the drafting. The human does the sending and the conversation. Cross that line and you are back to the same automated noise everyone else is making — which is the exact thing this whole approach is meant to escape.

Intimacy - Not Immediacy.

The Play

If you are feeling like AI is making your business less personal, not more, this one is for you. The fix is not to use less AI. The fix is to point AI at the part of the work that has always been impossible for a human — 24/7 attention to the right moments — and reserve the human moment for what humans have always been best at: actually showing up.

Here is the play. Pick one funnel. Just one. Map every step. Then list every "almost" behaviour at every step — booked but ghosted, opened but did not click, scrolled but did not buy, replied once and disappeared. Each is a sticking point with a specific question behind it.

Now run this prompt in Claude using whatever subscriber data you can pull out of your CRM or ESP:

Act as my subscriber intelligence analyst.

Below is a list of subscribers and their last 14 days
of activity in my [funnel name] funnel.

For each person, infer:
1. The funnel stage they reached.
2. The specific sticking point that caused them
   to stop.
3. The most likely emotional driver behind that stop
   (price doubt / timing doubt / fit doubt / trust
   doubt).
4. One opening line — written in my voice — that
   opens a real conversation about that sticking
   point without trying to close them.

Limit output to my top 10 highest-intent "almost"
people. Rank by recency and depth of engagement.

[subscriber data follows]

Send the messages by hand. By hand. That is the part that matters.

Try this today. Pull a list of 50 subscribers who hit your booking page or order form in the last 14 days but did not convert. Run the prompt above against their behaviour data.

Send the top 5 messages personally — your name, your voice, your inbox. You will book at least one call this week from someone who never would have booked otherwise. And the conversation you have on that call will tell you more about your funnel than any analytics dashboard ever has.

Download accelerating-prospect-outreach.md