Columns
Ask your spreadsheet a question, get an answer on every row
Start new thread

Columns - Ask your spreadsheet a question, get an answer on every row

by•
Upload a CSV, type a question in plain English, and every row answers. "Is this a fit for commercial B2B services?" across 5,000 scraped leads, in minutes. Every answer carries a confidence value, so Columns tells you which rows to check instead of asking you to trust all of them. All your questions about a row are answered in one request, so the 6th column costs almost nothing. Re-run after adding rows and you only pay for the new ones. Running on JEV! (Typsafe AI)

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt 👋 Columns came out of a boring, expensive problem: you scrape 5,000 businesses, and you have no idea which 200 are worth calling. So you pay a VA for two days, or you skim it and guess, or the list just sits there. Upload a CSV. Add a column. Type the question you'd have asked the VA — "Is this business a good fit for commercial B2B facilities services?" Every row comes back answered. Three things I'd point at specifically: **It tells you which answers to distrust.** Anything can fill 5,000 cells. The problem is you can't trust a list you didn't check, so you check it, so the tool saved you nothing. Columns keeps the actual probability behind every answer — not a label invented afterwards. Turn on "Needs review" and the sheet collapses to the rows where the judgment was genuinely close, usually a fifth of the list. You check those 200 and trust the rest for a reason. **Extra questions are nearly free.** Every question you ask about a row is answered in the same request — the row is read once no matter how much you want to know about it. So don't add one column, add six: who they sell to, whether they're big enough to have a budget, is there a decision maker, how strong a fit, any disqualifiers. The list stops being names and becomes a briefing. **You never pay for the same row twice.** Results are keyed to the content of the row. Add 50 rows to a list of 5,000 and re-run, and you're billed for 50. And at your limit, runs *pause* rather than billing past it — no overage, no surprise invoice. There's a live demo on the landing page — a real spreadsheet, really being judged, that you can type into. No signup. Free tier is 500 row-questions and takes no card. Paid starts at $29. It's priced on row-questions, never on seats — put your whole team on it. Built on TypeSafe Jev rather than an LLM, which is why every answer comes with a real probability instead of a vibe, and why the unit costs work at all. Also why it's so fast! I'd genuinely like to know what question you throw at it first, and whether the answer held up. Reply here and I'll be around all day.