RecipeMemoir Unit Economics
Enter hub password to continue.
That's not it. Try again.
What it costs to set up a new recipe site, and what to charge
← Back to hub
Short answer: setting up a brand-new recipe website costs you almost nothing. For a typical 300-recipe collection it's about 30 cents in one-time AI costs, then pennies a month to keep running. Your $4/month price isn't just affordable, it's roughly an 80 percent margin. The thing that actually limits your pricing isn't API cost at all: it's the payment-processor fee (Stripe takes about 42 cents on a $4 charge), which is why you don't want to price below about $3. The one cost that could run away is AI photo upscaling, so that's the piece to put a cap on.
1. What one recipe costs to process
These come from what we actually spent running the engine on your own collection, not estimates:
| Setup step (per recipe) | Cost | Basis |
|---|---|---|
| Macros (calories + protein/fat/carbs/fiber/sugar/sodium) | $0.0003 | measured: $0.005 / 20 recipes |
| Micronutrients (10 vitamins + minerals) | $0.0001 | measured: ~$0.25 / 1,700 recipes |
| Auto-categorize (tags from the taxonomy) | $0.0003 | measured: ~$0.005 / 20 recipes |
| Buffer (retries, a little commentary) | $0.0003 | safety margin |
| Total per recipe | $0.001 | about a tenth of a cent |
The food database (USDA) is a free one-time download that lives in Cloudflare. Storage, database, and hosting all sit inside Cloudflare's free tiers at this scale. Login emails are free too. So the only real per-recipe cost is the DeepSeek calls, and DeepSeek is deliberately cheap (it's why we route the bulk AI work through it instead of Claude).
2. What a whole new website costs to set up
| Collection | Recipes | One-time setup cost | Covered by one month of $4 |
|---|---|---|---|
| Small starter | 100 | $0.1 | 40x over |
| Typical cook | 300 | $0.3 | 13x over |
| Serious collection | 700 | $0.7 | 5x over |
| Annette's own | 1,700 | $1.70 | 2x over |
Even your own 1,700-recipe collection, the biggest realistic case, costs under $2 to fully process end to end. One month of a single $4 subscription pays for it twice.
3. The real cost drivers, ranked
| Cost | Per $4/mo user | Notes |
|---|---|---|
| Payment processing (Stripe) | ~$0.42/mo | 2.9% + $0.30 flat. The flat 30 cents is why small prices hurt. |
| AI setup (one-time) | ~$0.30 once | typical 300-recipe collection |
| Ongoing hosting/storage/email | < $0.10/mo | free tiers cover almost everything |
| AI photo upscaling | $0.005–0.02 each | opt-in per photo; the one thing that can add up |
The one watch-out: photo upscaling. Every other cost is fixed and tiny. But if one enthusiastic user upscales 200 photos in a month at ~2 cents each, that's ~$4, which wipes out their whole subscription. Cloudflare gives 10,000 free AI credits a day across your whole account, so light use is free, but heavy use is not. Recommendation: include a monthly upscale quota (say 20 on Home Cook, 100 on Family) and sell a small credit pack for more. That turns your only unbounded cost into a bounded, even revenue-generating, one.
4. Margin at your current pricing
| Scenario | Revenue | Cost | Margin |
|---|---|---|---|
| Home Cook $4/mo, first month (300 recipes) | $4.00 | ~$0.73 | ~82% |
| Home Cook $4/mo, every month after | $4.00 | ~$0.50 | ~88% |
| Family $8/mo, ongoing | $8.00 | ~$0.65 | ~92% |
These are healthy software margins. The subscription absorbs the setup cost in the first few days and is nearly pure margin after that.
5. AI photo generation, the one big cost
Generating brand-new photos is different from everything above. Our quality model (flux-2-klein-9b, the same one the upscaler uses, and the right one for a gorgeous hero shot) costs a flat $0.015 per generated 1024x1024 image. In the 500-photo blocks you asked about, definite numbers, no ranges:
| Photos generated | Cost | Per extra 500 |
|---|---|---|
| 500 photos | $7.50 | $7.50 |
| 1,000 photos | $15.00 | +$7.50 |
| 1,500 photos | $22.50 | +$7.50 |
| 2,000 photos | $30.00 | +$7.50 |
This is now your biggest AI cost by far. Filling one 500-recipe cookbook with generated photos costs $7.50, about 15 times the ~50 cents of all the text and nutrition setup combined, and roughly two months of a $4 subscription. So generated photos cannot be a free, unlimited, bundled feature. Treat them as a paid add-on: either a one-time "fill my cookbook with photos" pack (e.g. $19 for up to 500, which covers the $7.50 cost with room to spare), or a photo-credit system with a small included monthly quota and packs beyond it. There's a cheaper fast model (flux-1-schnell, about $0.32 per 500) if you ever want a budget tier, but for cookbook hero images the quality model is worth it, and pricing it as an add-on keeps your margins clean.
6. What to actually charge
- Keep the $4 / $8 tiers. They cover costs many times over and leave ~80–90% margin. You do not need a separate setup fee; the AI cost is too small to bother charging for.
- Don't price below about $3. Not because of API cost (you'd still profit at $1), but because Stripe's flat 30-cent-per-charge fee eats a painful share of anything smaller.
- Cap AI photo upscaling with a monthly quota per tier plus an optional credit pack. That is the single change that protects your margin from a power user.
- Sell AI photo generation as a paid add-on, never bundled. At $7.50 in real cost per 500 generated photos it's your biggest AI expense; a one-time pack (around $19 per 500) or a credit system turns it from a margin risk into its own revenue line.
- If you ever want a cheaper entry point, an annual plan (e.g. $36/year = $3/mo effective) is better than a low monthly price, because you pay Stripe's 30-cent fee once a year instead of twelve times.
Bottom line: the API and LLM cost of standing up a new recipe website is not a constraint on your business. It's about 30 cents. Price for the value of a gorgeous personal cookbook and for payment-processing efficiency, cap the one runaway cost (upscaling), and you're comfortably profitable at the pricing you already sketched.