Cold outreach with per-recipient AI-generated brand imagery — collapses the Apollo-tier value-add

B2B outbound flow for a product with a visual hook (in source: branded corporate gifting; the pattern generalizes to any product where a per-prospect mockup carries the message). Step 1: define 3 target personas + minimal subject/body variants in ChatGPT. Step 2: buy Apollo for one month ($59 → 2,800 contacts), apply ChatGPT-suggested filters, export the contact list to Google Sheets (Apollo blocks API export — you live with the manual download). Step 3: build (or reuse) a per-domain image generator that takes a recipient's work email, pulls brand assets via the BrandFetch API, and renders a personalized mockup of your offer with their logo applied. Step 4: have Claude write a script that walks the Google Sheet, calls the image endpoint per row, and creates a Gmail draft per recipient with minimal text + one big clickable image of the offer as the CTA. Step 5: review drafts; after a few hundred, the prompts stabilize and you stop reviewing. Reported economics: ~6-7¢/email all-in, ~20¢/click, ~30% open/click conversion.

The strategic insight is not the workflow — it's what it kills. The operator's takeaway, verbatim: "if someone offers me a cheaper contact base than Apollo I'll cancel that subscription instantly." Apollo's pricing assumed you needed sequencing + AI-personalization on top. With Claude writing the script in an afternoon and gpt-image producing the per-recipient asset at 2-3¢, the only thing you still buy Apollo for is the contact dump — and at $59 for 2,800 leads that's a one-month spend, not a recurring SaaS bill. Any B2B outbound team paying for the value-add tier of an Apollo / Outreach / Lemlist should run this math against their own funnel.

1. Persona + copy in ChatGPT. Define 3 target personas. Generate subject + body variants in the most minimal form possible — the per-recipient image is doing the personalization work, the text is short on purpose.
2. Buy Apollo for one month. Apply the persona filters ChatGPT suggested. Export contacts to Google Sheets manually (Apollo blocks API export — don't waste an afternoon trying to script it).
3. Per-domain branded-image endpoint. Wire BrandFetch API → for each work email, pull the company's brand assets (logo, brand colors). Feed into gpt-image (or equivalent) with a fixed prompt template that places the logo onto your offer mockup. ~2-3¢ per image at non-top-tier quality settings — that's the unit cost driver.
4. Spam-script — Claude writes it. Tell Claude: walk the Google Sheets contact list, hit the image endpoint per row, create one Gmail draft per row with minimal text + the generated image as the clickable CTA. Save as drafts in Gmail, not sent — you'll batch-review then bulk-send.
5. Review drafts early; stop reviewing once prompts stabilize. First few hundred drafts: review every one. After the template hardens, the operator stops reviewing entirely.
6. Retarget the engaged. ~30% open/click but sales lag on first touch (long sales cycle for the source product). Cheap retarget the engaged contacts until they convert.
7. Cost-control the image generation. gpt-image at "not the highest quality" is 2-3¢/image — that's the lever for total cost. Don't reach for the top-quality tier unless your average order value justifies it. Total all-in lands at ~6-7¢/email, ~20¢/click.

The operator's reflection on what this kills:

What's beautiful about this case: it's pure vibe-coding — I haven't even
looked at what's under the hood, because "why bother". It's a very specific
tool for a specific task, not a product or a platform.

The personalized image triggers interest immediately — only possible with
image gen.

If someone offers me a cheaper contact base than Apollo I'll cancel that
subscription instantly.

Spam-script architecture:

For each row in contacts.csv:
 1. Extract work email → domain
 2. POST domain to BrandFetch API → brand assets (logo, colors)
 3. Compose image-gen prompt:
 "Render <our offer mockup> with <logo> applied in <brand colors>"
 4. POST to gpt-image (low/mid quality tier) → ~2-3¢/image
 5. Pick subject + body variant by persona
 6. Create Gmail draft via Gmail API:
 to: <row.email>
 subject: <variant.subject>
 body: <variant.body>
 inline_image: <generated_image, clickable to landing>
 7. Save as draft (do not send)

Batch-review drafts in Gmail, bulk-send the approved batch.

• Apollo blocks API export. You will manually download the CSV from Apollo. Don't waste cycles trying to scrape — just plan for it in the workflow.
• Open/click ≠ sales on first touch. Reported ~30% open/click but sales were "still warming" at writeup time. Long sales cycles need a retarget loop — bake it in or the campaign reads as a flop at 30 days.
• Image quality is the cost lever, not the script. ~75% of unit cost is gpt-image; the rest is Apollo amortized + Gmail. Don't optimize the script — optimize the image quality tier against your AOV.
• Minimum-text, one-image format only works with a visual offer. Source was branded corporate gifting (cookies with a company logo). For non-visual products (B2B SaaS without a UI screenshot story) the format degrades — switch to a per-prospect screenshot of their product / website with your tool's annotation overlaid.