The 79-point idea I'd never build, and the 74-point one I'd start tonight

36 new cards, 144 signals, and my whole feed is product hunt launches about AI agents. Nothing against them. But the scoreboard rarely matches the money, and today's list is a pretty good example.

The top card is AI finds leads and writes outreach at 79 — the highest score on the board. Pitch is: "I spend hours every day finding prospects, researching them, and writing cold emails, yet most get ignored."

Here's my problem with it. That's not a demand signal, that's a complaint about a saturated category. There are maybe 40 companies doing this right now, and the ones that win aren't winning on better AI, they're winning on data. Better emails don't fix bad lead lists. The actual moat is knowing that a specific person at a specific company just did a specific thing — raised a round, hired three SDRs, swapped out their checkout — and that requires scraping and enrichment infrastructure that costs real money before you charge a single user.

Also, cold email as a channel is genuinely hostile right now. Deliverability is a full-time job. You'd spend more time babysitting domains and inbox warmup than building product. So no, I wouldn't build it. If I did, I'd charge $99+/mo and go after one narrow vertical where the trigger event is obvious — say, Shopify stores that just added a new app category. Even then I'd be nervous. Score high, sell hard.

The 74 I'd actually start tonight is Stripe revenue leak audit tool. Failed payments, duplicate subs, proration mistakes, dunning that never fires. Every subscription business has this and almost nobody has looked. The pitch writes itself: "you're leaving $4k a month on the table, here's the invoice number."

What I love is the monetization shape. Free scan → $49-99 one-time full report → $19-29/mo continuous monitoring. That's a wedge, not a subscription ask. You show the number before you ask for money, which is the only reliable way to sell to a finance-adjacent buyer who doesn't trust you yet. The one-time report is under most people's card-limit-without-thinking threshold, and the ones who see real leakage will convert to the monthly without a fight.

The catch: read-only Stripe access is a trust ask, and you're dealing with money, so the technical bar is higher than the usual micro-SaaS. You can't half-*** idempotency or webhook handling when your tool's whole promise is "we find mistakes." But the work is a Stripe API integration and a rules engine. I could have a scan running this weekend. Stack: Next.js, Supabase for storage, Stripe's API for the reads, Resend for the report email, and a very plain HTML report — no dashboard, no charts, just a list of problems with dollar amounts. Ship the ugly version.

The third one that got me is Shopify store localization at 74. The pain is real and specific: theme copy, product copy, and checkout strings all live in different places, and translating a store properly means touching three systems plus app embeds. Anyone who's tried it knows the misery.

But this is an app store grind, not a build. You're one of several hundred translation apps, Shopify takes its cut, and reviews are brutal and titled "broke my theme." The pricing is tiered by languages or order volume, which sounds fine until you realize bigger stores are the ones who need it and they're the ones who want a human translation agency, not an app. I'd build the product but I wouldn't want to be the one doing support for it. Someone should though — the localization space is full of legacy apps from 2018 with bad UX and no AI. A clean, opinionated, AI-assisted version that actually handles checkout strings could take a real bite. Just not my bite.

For what it's worth, the 72-card audio/video to searchable text is fine, but the pitch is "every transcription tool butchers names," which is a feature complaint, not a business. Accurate proper nouns is a nice wedge for podcasters and legal, but you're competing with the cheapest commodity on the internet. And the support inbox idea at 68 is a $19-49/mo knife fight with Intercom's free tier.

Two 74s and a 79, and I'd take the Stripe one every time. Highest score isn't the same as highest odds.

Anyway — I'm going to go poke at a Stripe test account and see how many leaks a fake SaaS business can generate in an afternoon.