The 7 Figure Seller Summit just wrapped, and the survey we ran before it started told a story I did not expect.
Forbes covered why sellers are turning to AI, and we got a mention.
Amazon Ads expanded its AI controls, Amazon Business crossed $60 billion, and there's a new seller-fulfilled delivery requirement with a September deadline that most people have not seen yet. Let's get into it.
In today's issue
What Summit attendees told us about AI, and why it surprised me
The 3 step method for adopting AI, from an 8 figure seller
We got mentioned in Forbes
Amazon Ads expands Brand+ and Performance+
Amazon Business hits $60 billion in annualized sales
A new 90% delivery requirement for FBM sellers starts September 30
The AI Workflow Hackathon replay is up
7 Figure Seller Summit 12 Recap and Takeaways
The summit wrapped this week. 3 days, 15 sessions, all of it on AI.
Before it started, we surveyed the pass holders. 44 people answered.
1 number stopped me.
36 of 44 either don't use an AI agent harness, or don't know what one is.
These are not beginners. These are people who signed up to learn AI.
40 of 44 already have the model. Almost nobody has the setup.
Here's the rest of what the survey said.
21 use ChatGPT. 19 use Claude. So 40 of 44 already pay for a frontier model.
Then it falls off a cliff. 24 use no agent harness at all. Another 12 said they don't know what an agent harness is. Only 3 use OpenClaw. Only 3 use Hermes.
I thought that might just be our room. It isn't.
Chris Rawlings, founder of Sophie Society, spoke on Day 1. He told us he polled the audience live at an ecom AI event, where every person in the room had shown up specifically for AI:
"I asked them how much they're using Claude in their business. And 19 out of 20 of them were either not using it at all or were using it in a very limited capacity."
He also put a number on how small the real group is:
"There's something like between 5 and 8 million sellers on the planet. And it's literally only in the thousands, the number of sellers that are actually really good at applying AI."
2 different rooms full of AI-motivated sellers. Same answer both times.
1 of our own survey respondents named the gap better than I could:
"Moving from AI levels. Going from chat AI uses to cowork uses, to agents doing work and beyond."
TAKEAWAY: The model was never your bottleneck. You already pay for it. If you're copying and pasting between a chat window and Seller Central, you don't have AI in your business. You have a smart intern with no desk.
Most of us are automating the wrong 20%
I put the survey answers side by side and the gap jumped out.
Here's what sellers told us they use AI for right now:
Listings, images, and content: 13
PPC and advertising: 10
Coding and workflow automation: 4
Internal operations and admin: 4
Inventory, forecasting, and supply chain: 4
Now here's what they said they're still doing by hand and most want to automate. Of the 44 answers to that question, our team coded:
PPC and advertising terms in 15
Listings and content in 8
Inventory, forecasting, and supply chain in 7
Content is where AI is easy. PPC and inventory are where the money is. Almost nobody has crossed over.
In their own words:
"I have built Excel-based rules, used bulk files, AI tools and PPC software. These solutions save time, but the workflow is still fragmented and requires manual data exports, checks and implementation inside Amazon Advertising."
"I want to automate my workflow with listings as much as possible. I have more than 1000 of them. And every change is a waste of time. I need automation."
Ritu Java, CEO of PPC Ninja, gave the cleanest diagnosis of the whole summit on Day 2:
"Our strategy is not the bottleneck. Our busy work around the strategy is the bottleneck."
TAKEAWAY: Writing bullet points faster does not move profit. Catching a stockout does. Look at what you actually pointed AI at last month and ask yourself if it was the easy task or the expensive one.
Why it feels like slop
The number 1 hesitation in the survey was not cost. It was trust.
"security and the AI doing bad work that costs me money"
"Quality, frustration with getting prompts to produce the outputs we envision"
"PPC decisions affect real client budgets, so I need automation that follows clear rules, detects unusual situations and allows human review before important changes are implemented."
Read that last one again. That's not a complaint. That's a spec. He described the fix while he was describing the fear.
In reality, slop is a structure problem, not a model problem.
Moshe Scheiner showed this live on Day 3. He's an 8 figure seller, 10 years on Amazon, and he just got funded by Perplexity.
He opened a clean ChatGPT window in front of us. Dropped in his real Amazon bulk report. Asked it to act as Einstein and give him recommendations.
Then he checked the output against his own account:
"A lot of these are wrong for many different reasons because I'm focused on organic ranking. And if I'm doing an auto campaign, the last thing I want is to increase. It doesn't really have that context."
His picture of what's actually broken has stuck with me all week:
"Imagine you had Einstein on your account, but every 10 seconds he had to close his eyes. Every time he closes a session, he wakes up and he lost his memory. And he's got to manually upload those to Amazon and manually test those."
3 things missing, in his words: context, memory, and the execution layer.
His 80/20:
"People are always trying to take shortcuts with AI. I think you're thinking about things the wrong way. AI is not about saving time necessarily, it's about doing things better to make better decisions."
TAKEAWAY: Stop asking AI what to do. Give it your real data, your written rules, and a place to stop for your approval. Most of the hallucinations go away when the structure shows up.
The 3 step method for adopting AI, from an 8 figure seller
Giorgio Piccoli runs an 8 Figure Brand, American Flat.
27 people on the team. 10,000 picture frames shipped every single day.
He did not hire an AI consultant. He changed the culture. And his OpEx dropped:
"Our OpEx is basically a third of what it was this time last year."
He was also honest about how hard the adoption part is:
"Anyone who says that they've snapped their fingers and everyone is adopting AI is lying."
Here's his method. He borrowed it from lean manufacturing and it's 3 steps.
1. Set the standard.
Enthusiasm has to come from the top or nothing moves. Giorgio runs a company wide morning meeting every day, and a different person leads it each time. They read a passage from a book, they train on the AI tools, and everyone learns to lead.
Then he protects the build time:
"Now we have 1 hour where I do not want you to focus on work. I want you to focus on building and I want you to experiment."
He also built the plumbing so people had something to build on. All their business data flows into BigQuery through APIs: Amazon, Slack, Notion, the accounting system, the warehouse system. Every Claude skill anyone writes gets shared company wide in GitHub.
2. Observe the standard.
Watch what people actually ship. Giorgio tracks how many skills get uploaded to GitHub as his adoption metric. He also invites outsiders into the morning meeting on purpose, because being watched changes behavior.
3. Celebrate the standard.
Whenever someone builds a workflow that works, he highlights that person publicly.
"It's really about highlighting that person and celebrating exactly what's happening."
His marketing director, in his words, "became a 3x employee overnight."
Here's why he says bringing in an outside expert won't fix this:
"You may have one person who's coming in and fixing all those things, but the culture of the company really hasn't changed. At the pace that AI is evolving right now, a lot of it is gonna be stale in 6 months."
TAKEAWAY: If you have a team, even 2 VAs, this is the highest leverage thing on this list. Nobody adopts AI because you sent them a link. They adopt it when it's on the calendar, protected, and rewarded.
What it looks like when it's built right
3 more sellers from the stage this week. Every number below is what they said on the recording.
Andrew Erickson, Inventory Hero. 3 seven figure brands, 2 exits.
"$418,000. That's how much I lost, and that's why I quit my day job."
That was opportunity cost from stockouts in his third year. Lost sales, plus lost rank, plus lost reviews, plus getting pulled out of Subscribe and Save.
That same year, he also spent $87,000 in storage fees. $65,000 of that landed in Q4, because Q4 storage rates triple.
"One extra week optimized would save me $4,000. Two or three weeks, we're saving $15,000."
What he runs now is an agent inside Slack. It reads his inventory, flags what's low, suggests the order quantity, builds the purchase order, renders a branded PDF, and drafts the email to the supplier.
It does not send.
"I would never expect an AI bot to place a $65,000 purchase order on my behalf. This takes a 3 minute task in 3 minutes instead of 3 hours."
His averages, measured across 12,000 FBA shipments: 45 days production, 25 days ocean, and 6 to 7 days for Amazon to receive it, sometimes over 30. That's 12 to 14 weeks minimum.
Now put that against Amazon's published arrival deadlines. Prime Big Deal Days inventory has to arrive by September 2 for AWD, September 9 for FBA with minimal splits, and September 16 for Amazon-optimized splits. Black Friday Week and Cyber Monday are October 14, October 21, and October 28.
Work backward 12 to 14 weeks from those dates and you land on right now.
Mina Elias, Trivium Group. 42 agents running right now.
Mina runs the agency side of this. He has 80 to 90 human employees, and he added 50 to 60 AI ones.
The cost surprised me:
"For all 42 of these agents running autonomously, it costs about a thousand a month."
His total spend on the platform is closer to $4,000 a month, but $3,000 of that goes to building new agents. The ongoing run cost is the $1,000.
What they actually do: monitor listings and account health, watch competitors, flag wasted ad spend, catch keyword opportunities, and turn his YouTube videos into LinkedIn posts and emails. On the agency side, 1 agent reads every client Slack channel and call transcript daily and flags the questions a manager forgot to answer.
Here's the part I'd underline. He was direct about why most people's version doesn't work:
"Mine is trained on me. An AI will have general knowledge, so it can do general things. But that is the worst thing that you want. You can't bring a general person off the street and say, hey, come here, run my Amazon."
He fed his agents hundreds of hours of his own SOPs, internal training, client calls, and YouTube videos. Then he set hard guardrails in a master behavior guide: propose changes, don't implement them. Anything global needs his approval.
His 80/20:
"Take a team member, write down the list of things that they do. Then implement 1 agent. 1 agent a week is 52 a year. You will in 12 months have transformed your business. You just have to be consistent."
Ritu Java, PPC Ninja. More than 50 Claude skills running in production.
She showed the actual anatomy. Every PPC skill she builds needs exactly 3 things.
A real data source. Streaming and time series, not downloaded snapshots. In her words, "If you're relying on just snapshots of downloaded Excel files, that's very limiting data and a skill might fail."
Decision rules written as code. "You need to be able to convert parts of the skills into programs like Python scripts that'll run reliably. Otherwise it'll hallucinate."
A defined output. Bulk file, report, or email. Decide before you build.
She also draws the human line hard. Her negative keyword skill deliberately has no auto upload:
"This one definitely needs a human in the loop because of the severity of what could happen if the negation is done without careful thought."
Her 80/20:
"If there's any task that you find yourself repeating over and over again, monthly or weekly or even quarterly, just turn that into a skill. Stop the procrastination, turn it into a skill today."
TAKEAWAY: None of this came out of a chat window. The distance between "AI writes my bullets" and "$418,000 I will never lose again" is not a better prompt. It's data, rules, and an approval step.
What to do this weekend
Pick the task you repeat the most. Not your hardest one. Your most repeated one. The weekly PPC pull. The monthly inventory count.
Write the rules down the way you'd hand them to a new VA. If you can't write them, AI can't follow them.
Point it at real data. Export the actual report. AI guessing is AI hallucinating.
Name the approver. Anything that touches money, bids, listings, inventory, suppliers, or customers stops for a human.
Give it memory. Save what you decided and why, so next month it isn't starting from zero.
Run it 4 weeks before you automate the approval step. Not before.
You can still catch the end of Day 3 right now if you act fast.
Elaine Pofeldt wrote a piece for Forbes on July 29 about ecom sellers turning to AI to protect profitability, and the 7 Figure Seller Summit got a mention.
I'm grateful for it, but the reason I'm sharing it isn't the mention. It's that the framing is right. Sellers are not chasing AI because it's fun. They're chasing it because margins got squeezed by tariffs, fees, and ad costs, and this is the first lever in years that doesn't require more headcount.
TAKEAWAY: If you've been telling yourself AI is a distraction from the real work, the profitability math is the argument to revisit. Read the piece and see if your situation is in there.
On July 22, Amazon Ads announced 10 new capabilities across Brand+ and Performance+.
The one worth your attention is Audience Signals, in Open Beta as of July 20. It lets you feed your own first-party audiences in through Ads Data Manager.
Read that carefully though. Those audiences are optimization inputs, not hard targeting rules. You're not telling Amazon who to show ads to. You're giving its model a hint about who converts.
Availability varies by feature, region, and account, so check your own account before you plan around it.
TAKEAWAY: Before you hand an optimizer more signals, make sure the signal means what you think it means. Audit your consent and list provenance, your audience definitions, and your conversion events first. A model trained on a sloppy conversion definition will spend your money confidently in the wrong direction.
On July 21, Amazon reported that Amazon Business reached $60 billion in annualized gross sales in Q2, serving more than 11 million organizations worldwide.
Business buyers can get business-only pricing and quantity discounts on millions of products, with quantity breaks starting at just 2 units.
Now, that $60 billion is Amazon's number for the whole channel, not third party seller revenue. Don't read it as your opportunity. Read it as evidence the buyer is there.
TAKEAWAY: Don't chase B2B across your whole catalog. Find the 20% of your ASINs a business buyer would reorder in quantity, then check business pricing, quantity discounts, and your unit economics on those specific SKUs before you change a single offer.
This one is flying under the radar and it has teeth.
Starting September 30, US seller-fulfilled shipments to Amazon Business customers have to maintain a 90% Business Hour Delivery Rate over a rolling 14 day period.
If you fall below it, Amazon says it will notify you. If it hasn't improved by October 30, your seller-fulfilled offers can be deactivated for Amazon Business customers.
FBA is not affected. Your retail offers are not affected. But if a meaningful chunk of your FBM revenue comes from business buyers, that revenue is exposed while your retail listings look completely healthy.
TAKEAWAY: Pull the Business Hour Delivery Rate view in Account Health this week and see where you actually stand. If you're near the line, the fix is usually in carrier selection, handling time, and transit settings, and you want to make those changes before the rolling 14 day window starts counting against you.
If you missed the live hackathon, the replay is on YouTube now.
Andrew Erickson runs these about every 2 weeks.
Get smart people in a room building in real time and let everyone else watch how they actually think through a problem.
In his words:
"As long as you have 3, 5, 10 smart people in the room, and then sometimes there's 40, 50 people who are just kind of fly on the wall listening to that, everyone learns so much by doing that."
TAKEAWAY: Watching someone build a workflow beats reading about one. If you're stuck at the chat window stage, this is a low effort way to see what the next step actually looks like.
In case you missed it:
See you next week,
Gary