A Bronx Truffle Maker Sent 40% More Email. Its Unsubscribe Rate Went Down.

• Blog
• Ai Send More Email Less Unsubs

October 30, 2026

In email marketing, it’s a given: The more you send, the more people leave.

The Maccs, a Bronx bakery that makes toasted coconut and Belgian chocolate truffles, sent about 40 percent more email from March through September than in the five months before. Its unsubscribe and spam rate also fell, and each email earned about 25 percent more than before.

The biggest change in March was who wrote the first draft. Clickable Impact, the Orlando agency that runs The Maccs’ email, asked Claude, Anthropic’s AI model, to write every campaign’s subject lines and copy from a strategist’s short brief, while people still choose, edit, and approve every send.

The agency doesn’t credit Claude with all of the gains, but the usual cost of sending more email, a climbing unsubscribe rate, hasn’t shown up.

If you run email for a brand whose list has outgrown its team, the workflow below is worth copying. None of it needs a developer, and the briefs that run it are a few sentences long.

The business behind the inbox

The Maccs founder and CEO Michelle Goldberg, a triathlete, self-described “sweet freak,” and mother of four picky eaters, couldn’t find a dessert that didn’t leave her feeling heavy and sluggish.

“I’m all about the art of allowing,” she said. “Maccs are an allowable indulgence that I created from coconut and chocolate, my two favorite flavors.”

The Maccs bakes to order in the Bronx and sells packs of four truffles for $12, with a $24 minimum. There are more than 20 flavors, and some only appear seasonally, which matters for email: A pumpkin flavor that launched October 1 has to sell before winter.

The list holds about 33,000 email subscribers. Store revenue rose from $18,194 in September 2025 to $138,939 in September 2026, with most of that growth coming from paid social and creator programs. Revenue credited to email grew from $6,835 to $43,662.

In fall 2025, around the time she hired the agency to run her email, Goldberg rebranded the company. The timing made for a rough start.

“It was very challenging when we first went to Clickable because we had just rebranded,” she said. “Customers had known the brand by one name and signed up for email under that different name, so a lot went to spam.”

The rebrand also set the tone for everything since. “Since the rebrand, we’ve focused on building our community and increasing the amount of real content, not AI content,” Goldberg said.

That’s a striking thing to hear from a founder whose emails all start as AI drafts. The workflow below shows how the two fit together.

What changed, and what didn’t

The AI came in stages. From October through February, it only generated product images. In March, Claude began drafting every email. On September 29, the team built an automated sequence that times emails to each customer’s predicted next order.

Because those stages came months apart, The Maccs can show a before-and-after:

The program was already improving before Claude: Revenue per email climbed from 5.4 cents in October to 15.4 cents in February, a Valentine’s Day month that beat the March–September average. And because the team sent more, more people unsubscribed in raw numbers, even as the rate fell. The rate climbed again in September, to 0.39 percent, and the October calendar called for 28 campaigns.

The rebrand shaped the early numbers, too. Subscribers who didn’t recognize the new name sent a lot of mail to spam, and that fallout likely lingered through the winter. Some of the later drop in spam complaints probably reflects customers learning the new name.

The Maccs sent the most emails in June 2026 (356,453 emails). And it posted the year’s lowest unsubscribe and spam rates, according to John Surabian III, the agency’s brand growth strategist.

The workflow, in three stages

Stage 1: Campaign pictures from one real photo

The team started using Nano Banana in October 2025, asking the image generator to use a real product photo instead of booking a shoot. Here’s the prompt behind it, with the photo attached:

Ultra-detailed dynamic closeup shot. Two Products (reference) on top of each other, natural shadows. Smooth (Orange) gradient background. Fresh raspberries and Orange slices near the product. premium food photography inspired by luxury chocolate brands, Keep the product visuals exactly the same. High-quality. 1:1

Left: product photo. Right: Nano Banana’s image.

How to build it

• Start with a clean, well-lit photo of the product.
• Attach it as the reference and describe the scene: the background, props that tell the flavor story, and the shape you need (square fits email and social).
• Add a line that locks the product, such as “Keep the product visuals exactly the same,” so the product itself stays true to life.
• Check every image for things that couldn’t happen in a real kitchen before it ships.

Step 4 is there for a reason. Customers like to eat The Maccs frozen, so the team asked for a box in a freezer. The model drew icicles growing on the box, or parked it alone in an otherwise empty freezer.

“There definitely was a learning curve in the prompts,” said Surabian.

Goldberg added she’s learned you need to review every AI-generated photo and refine the prompt to get exactly what you want to represent your brand.

Stage 2: Claude writes the first draft of every campaign

Every email starts as a record in the team’s Airtable. A strategist creates a short brief, names the segment (e.g., everyone who opened an email in the last 180 days or everyone who bought dark chocolate), and sets the date. A copywriter pastes the brief into a Claude Project. The Maccs’ project holds the brand’s rule book, a word bank, flavor profiles, and past emails.

This is the brief for the October 1 launch, exactly as it sits in Airtable:

Position 1 of 28 (first, per client feedback). Launch email announcing Pumpkin Pie (white chocolate), Pumpkin Spice (dark chocolate), and the Fall Variety Pack (Pumpkin Pie, Pumpkin Spice, Apple Cinnamon). Limited time, fall only. No discount included. Launch date October 1. All three products are limited time, available through fall. The Fall Variety Pack is NOT live on the website yet (product description supplied from Shopify); confirm it is live before send. No product page URLs confirmed; CTA links to general collection page pending.

The brief goes beyond the product list and tells the human reader what isn’t ready, what to check before sending, and what’s off the table.

Since October, Claude has used the same format every time. There’s a line on format and voice, three subject lines, three preview texts, three headlines, the call to action, the body, the sign-off, and a P.S. Then a copywriter picks and edits, Goldberg approves, and the email goes to design.

For the fall launch, the team took the first option in each set. The subject line read “Pumpkin Pie, Pumpkin Spice, and a Fall Variety Pack,” and the headline read “Fall Has Arrived in Maccland.” The body shipped close to how Claude wrote it. The people chose among the options, built the sections, designed the email, and got it approved. It earned 30 cents per recipient, on 45 orders worth $4,735.

What Claude returned for the Oct. 1 launch (left) and the email as sent (right), with the team’s picks marked.

Surabian credits that fixed package with moving the team from eight to 13 sends a month in fall 2025 to 12 to 17 in 2026.

How to build it

• Identify your brand rules and banned words, a word bank, notes on each product, and a handful of your best past emails.
• Load them into a Claude Project, or whatever your AI tool calls a workspace with saved files.
• Write a brief for each campaign, including the segment, the goal, the offer rules, the date, and anything that isn’t ready yet.
• Ask for three options for every subject line, preview text, and headline. Picking is faster than fixing.
• Keep humans in the loop. A strategist approves the plan, a copywriter edits, and the brand owner approves the words.

The team added step 4 after a mistake shipped. Through September, Claude returned single drafts, and in the Sept. 4 self-care email a line meant for the team slipped into the copy customers read: “This is the heart of the Purpose pillar, and honestly, the heart of the whole brand.” Nobody caught it. “Every October document came back with options,” Surabian said, and each now gets a second review, marked V2, before it goes to design.

Stage 3: Timing each email to the customer’s next order

Stage 3 is the least tested.

On Sept. 29, the company gave one sentence to their AI assistant:

“Generate 3 emails predicting next order flow with founder’s tone, scrape the text based emails we sent in the past to master the tone.”

The system generated three emails timed around each customer’s predicted order date, including one with the subject line, “Your freezer stash, checking in.” Anyone who orders in the meantime drops out.

Without a prediction tool, you can approximate it. Export your orders, find the typical gap between a customer’s first and second order (use the median, so a few outliers don’t skew it), and time the same emails a few days ahead of it. The Maccs has run a plainer, calendar-based version of that reminder since August 2025.

What to steal

Divide the labor between AI and your team. At The Maccs, AI did the production: the pictures, the first drafts, the automated sequence. People kept the decisions about who hears what, and when.

Pull your last five campaigns and write, for each one, the brief you wish you’d written: the segment, the goal, the offer rules, and what wasn’t ready. If a campaign can’t be briefed in a few sentences, no model will rescue it.

Then add AI one stage at a time, and measure each stage against the months before it. If you switch everything on in the same week, you’ll never know which part worked.

Put revenue per email and unsubscribe rate on the same line of your report. If both move the right way as volume climbs, the extra sends are paying for themselves. If either slips, they’re costing you.

Early results and learnings

The Maccs didn’t do an A/B test comparing Claude’s copy against a person’s. And the store grew fast over the same months through paid social, creator programs, new flavors, and new offers, including the $24 minimum and bundles that lifted the average campaign order from $60 to $88.

The numbers show something narrower and more useful: A small team sent about 40 percent more email, lost a smaller share of its list to unsubscribes, and earned more per message while AI wrote the first drafts.

Goldberg sees email as one part of a larger system.

“All of it (the emails, the paid ads, and our own social media and website) gives you feedback on what customers want and why they’re buying or not buying,” she said. “They get the emails and see the social proof. One system supports another system.”

The new predictive flow’s first emails go out in November. By the holidays, The Maccs will know whether the AI’s sense of timing beats the calendar.

“You create a community for your brand and stand out by showing there are people who stand behind the product and care about every package that goes out,” Goldberg said.