← Back to Pulse PULSE. brief
I Helped a Complete Stranger Build an AI Ecom Store

I Helped a Complete Stranger Build an AI Ecom Store

Davie Fogarty30 min2026-09-29 ▶ Watch on YouTube
ℹ️
Partly verified — 5 of 37 specificsA few specific details here couldn't be independently confirmed against the video. The overall summary is sound, but double-check exact numbers or names before you rely on them.
What this video is
⚡ a 30-minute video, readable in 60 seconds

Davey, who says he has done over a billion dollars in e-commerce sales, mentors first-time founder Kelsey through launching a dropshipping store from scratch: finding validated products with an ad spy tool, generating AI product photography and UGC-style ad creative with Higgsfield and Claude, and building a Shopify store using Claude connected via MCP. The brand, named Veloris and anchored by an anti-theft crossbody bag, crossed $1,000 in sales on day two of running ads and grew to over $42,000 in sales across the following eight days, prompting the mentor to call it 'a $3 million store' based on its run rate. Along the way the video covers the technical workflow in detail (ad-spy filters, JSON-prompted image generation, brand naming rules, MCP store-building) as well as the personal and business realities of launching a brand, including Kelsey sharing the loss of her father and closing advice to know the numbers before scaling.

[01:24] Product research goal when using the ad spy tool: find a product that hasn't been scaled too far but has shown product validation.
Key takeaways
+ 29 more takeaways
  • [04:43] Anti-Theft Wallet/Bag sells for around $40 retail, sourced on AliExpress for around $10 USD.
  • [05:13] Cellulite leggings sell for over $70 AUD and can be sourced for as cheap as $1.40 USD.
  • [06:13] Team decides to start with the anti-theft bag and begins creating product images for it.
  • [06:19] AI image method: find a product category on Pinterest, gather inspiration images, and feed them into an LLM like Claude to write a detailed JSON prompt for image generation.
  • [06:55] Tip: run 4 variations with the same prompt but different seeds to pick the most believable skin texture.
  • [07:07] Major issue found: the AI-generated bag in the image was not the actual product being sold.
  • [07:18] An inpaint tool is used with an uploaded reference image to replace the AI-generated bag with the real product.
  • [07:43] The corrected product image was generated within 30 seconds.
  • [08:04] Real AliExpress listing for the bag: priced AU$1.40 (marked down from AU$8.48), 4.4 stars from 146 reviews, 11,000+ sold.
  • [08:44] Recommendation: create 10 ads per product (50 total); since the Facebook Andromeda update, each ad must be individualized rather than duplicated creative, or it serves the same audience.
  • [11:35] Kelsey shares that her dad passed away two years ago at 58, close to retirement.
  • [14:24] Xero, used as accounting software for every business Davey has launched, is recommended to Kelsey once she gets her first sale, currently offering 90% off for 6 months for new customers.
  • [14:51] Brand naming must follow the 'DTF' framework: a buyable Domain, no Trademark conflicts, and a Flexible name (citing 'Calming Blankets' as an example of an inflexible name).
  • [15:37] The chosen brand name is Veloris, which Davey says reminds him of Polaris/4x4 vehicle branding.
  • [16:17] Claude is connected to Shopify using an MCP (Model Context Protocol) so the AI can talk directly to Shopify and build the store.
  • [17:21] AliExpress listing details (Fengdong anti-theft cross body bag, waterproof nylon, 0.17KG, ultra-thin) are fed to Claude to help draft polished product copy.
  • [18:15]-[18:19] The published 'Anti-Theft Cross Body Bag' product page goes live in the Veloris store, priced at $0.00 AUD and marked sold out at that stage.
  • [18:54] What normally takes about two to three weeks was done in less than a day, starting around 9:30am, building the entire Shopify store.
  • [20:28] The vest uses UPF 50+ knit fabric, the same rating used in premium swimwear, blocking 98% of UV-A and UV-B rays.
  • [21:36] After ads ran for weeks without converting, the team used Higgsfield via an MCP connector into Claude to create UGC-style ads, generating seven or eight variants and sending them live within about five minutes.
  • [24:00] On day two (August 8, 2026), the store crossed $1,000 in total sales with ad spend raised to $266.58/day, yielding a 3.63 purchase ROAS.
  • [25:02] On August 10, 2026, ad spend was $1,079.02 with 56 purchases, 2.36 ROAS, $2,551.17 purchase value, and $19.27 cost per purchase.
  • [27:38] Over the following eight days (from around August 10), the store did $42,159.88 in sales.
  • [27:52] The mentor calls it 'a $3 million store,' referring to its projected run rate.
  • [28:52] Kelsey's main advice for e-commerce beginners: fundamentally know the numbers first, including cost of goods (cogs) and net margins, before scaling a product.
  • You know, I thought we could get it to a million dollar store, but, you know, doing $5,000, $8,000 days in eight days, that is insane.
  • They're selling roughly around $15 USD, which is more of a lower value product, but we can source them on AliExpress for around $8.
  • So you can already see that we're starting to get some things that have spent, you know, $167, $700.
  • Slowly I've increased ad spend from like $100 a day yesterday to launch and test.
Tip: run 4 variations with the same prompt but different seeds to pick the most believable skin texture.
Tip: run 4 variations with the same prompt but different seeds to pick the most believable skin texture. ▶ 7:00
AliExpress listing details (Fengdong anti-theft cross body bag, waterproof nylon, 0.17KG, ultra-thin) are fed to Claude
AliExpress listing details (Fengdong anti-theft cross body bag, waterproof nylon, 0.17KG, ultra-thin) are fed to Claude ▶ 17:33
The vest uses UPF 50+ knit fabric, the same rating used in premium swimwear, blocking 98% of UV-A and UV-B rays.
The vest uses UPF 50+ knit fabric, the same rating used in premium swimwear, blocking 98% of UV-A and UV-B rays. ▶ 20:29
Shown on screen — grab and go
PROMPTCoastal Editorial e-commerce hero-shot JSON prompt
{
  "title": "Coastal Editorial - Women's E-commerce Hero Shot",
  "intended_use": "Lifestyle / hero image for women's accessories or apparel e-commerce",
  "aspect_ratio": "3:4",
  "recommended_model": "Higgsfield Soul (photoreal) or equivalent realism model",
  "main_prompt": "Editorial fashion photograph of a woman walking along a quiet beach at early golden hour, framed from nose-down to mid-thigh, carrying a cognac-brown half-moon crescent leather shoulder bag with soft pebbled grain and a thin adjustable strap, wearing an oversized ivory silk high-collar blouse with billowing long sleeves and a matching flowing cream midi skirt. Soft diffused natural sunset light, pastel peach and lilac sky, gently blurred ocean waves in the background. Shot on 85mm prime lens at f/2.0, Kodak Portra 400 color palette, subtle 35mm film grain. Calm, refined, quietly aspirational mood.",
  "subject": {
    "gender": "female",
    "age_range": "late 20s to early 30s",
    "ethnicity": "warm medium skin tone, ethnically ambiguous",
    "hair": "natural voluminous curly afro, slightly windswept, visible flyaway strands, not glossy or over-styled",
    "expression": "calm, neutral, face partially cropped or turned away from camera",
    "pose": "side-profile mid-stride, bag resting naturally at hip, one hand soft at side, relaxed shoulder drop, weight on back leg",
    "skin_tone_notes": "warm olive-bronze with natural undertone variation (not uniform)"
  },
  "wardrobe": {
    "top": "oversized cream/ivory silk blouse, mandarin high collar, long billowing sleeves with wide folded cuffs, subtle textured weave, natural soft creases and drape",
    "bottom": "matching cream flowing midi skirt, light silk or crepe, gentle movement from breeze",
    "bag": "cognac / burgundy-brown leather half-moon crescent crossbody bag, soft pebbled grain, slim gold-tone ring hardware, thin leather strap with small dangling tie. NO visible logo, No text, No brand markings of any kind.",
    "accessories": "none - minimal and clean styling"
  },
  "setting": {
    "location": "quiet empty sandy beach, soft ocean waves in middle distance",
    "time_of_day": "early golden hour / soft sunset"
  }
}
Merged from two scrolled reads of the same VS Code JSON file at 06:52 and 06:54; each frame filled a gap the other left, so the combined field text has no [...] breaks. shown at 6:52
PROMPTInstruction asking AI to build the JSON prompt
I plan to create an image inspired by the above using an image creation tool like NanoBanana and Higgsfield. Create a detailed JSON file that I can copy directly. Include code to ensure skin isn't smooth and the photo has slight imperfections to make it not feel like AI. The image will be for my WOMENS ECOMMERCE STORE.
Transcribed from the on-screen chat prompt at 00:20 (spacing restored from crushed OCR); the same instruction reappears at 06:33/06:41 confirming the wording. shown at 0:20
CODESeed and prompt-tuning tips for the shoot
{
  "tip_1": "Run 4 variations with identical prompt but different seeds - pick the one with most believable skin texture.",
  "tip_2": "If the first pass still looks too smooth, append to the main prompt: 'visible skin pores, realistic skin texture, matte skin, film still, unretouched'.",
  "tip_3": "If the bag looks wrong, describe its shape in physical terms rather than by brand (e.g. 'half-moon, curved bottom, flat top opening').",
  "tip_4": "For e-commerce consistency, save the seed of the winning generation and reuse it for follow-up shots in the same campaign with prompt variations (different outfit, different bag color, etc.)."
}
Transcribed from the on-screen JSON tips block at 06:55, legible apart from spacing repairs. shown at 6:55
CODEAI-image rejection criteria for a wireless device
{
  "infrared_glow": "Must be warm amber-orange — reject if it renders white, blue, or absent",
  "device_chrome_reflection": "Red metallic band may over-reflect — acceptable if warm-toned, reject if it looks like chrome mirror",
  "ear_anatomy": "If ear is visible, check it looks natural — AI sometimes distorts ear cartilage"
}
Shown fully legible on screen at 00:19-00:20 as part of a separate product's image-validation JSON. shown at 0:19
Their links, sorted & clickable
📢 Sponsored / affiliate1Get 90% off for 6 months with Xeroreferrals.xero.com
🔗 Other links4Learn how to build your AI-powered ecommerce brand herelp.dailymentor.coFor more educational videos, check out my second channelyoutube.com1️⃣youtube.com2️⃣youtube.com
← Back to Pulse Dashboard
Was this brief useful?