One pitch-deck DM turned into $17K MRR

The Case: $17K MRR, ~$200K ARR run-rate

What he does. Runs MagicSlides, an AI tool that converts prompts, PDFs, Word docs, and YouTube videos into editable PowerPoint files in 136+ languages.

Backstory. In late 2022, someone DM'd him for help with a pitch deck. Instead of doing the work, he read the request as a market signal and shipped a generator. The window was narrow and brutal: GPT was finally good enough to outline a coherent deck, but Office still owned the corporate desktop. He ranked early in long-tail keyword spaces and stayed bootstrapped while VC-backed rivals burned cash redesigning the slide.

Product. - Prompt to 10-slide editable .pptx in under 60 seconds - Input: PDF, DOC, PPTX, JPG, PNG, MP4 (YouTube), raw topic - Output: native PowerPoint file, opens cleanly in Office and Google Slides - 136+ languages, including Hindi, Arabic, Indonesian, Portuguese, Vietnamese - Distributed via Google Slides extension and ChatGPT plugin

Pricing. Free tier with watermark. $8/mo Essential, $14/mo Pro, $19/mo Premium. No annual contracts, no enterprise calls.

Key numbers. - MRR: $17K - ARR run-rate: ~$200K - Total presentations generated: 10M+ - Free-to-paid conversion (inferred): ~1.5% - Languages supported: 136 - Input formats: 6 - Founder count: 1 - Support load: zero retainers, zero enterprise calls - Time from inbound DM to paid users: under 30 days - Competitors funded above $40M: Gamma, Tome, Beautiful.ai

Stack. Next.js, GPT-4o-mini, python-pptx, Stripe, Vercel, Postgres, Cloudflare R2, Chrome Web Store.

Why this works (and what most readers will miss)

1. He picked the boring file format. Gamma, Tome, and Beautiful.ai all bet on browser-native slide formats that look gorgeous and edit nowhere. MagicSlides exports a .pptx the user's boss can open in Outlook, mark up in Office, and email back. The corporate slide is not a design artifact, it is a deliverable in an email thread. He won by refusing to redesign the deliverable.

2. 136 languages equals SEO arbitrage in keyword spaces nobody fights for. "AI PPT generator" is a bloodbath in English. "Generador de presentaciones IA" or "AI প্রেজেন্টেশন মেকার" has near-zero competition and a billion users behind it. He is not winning Google. He is winning every Google.

3. The DM was the product spec. The non-obvious move: he ran zero customer interviews. He treated one cold DM as a leading indicator and shipped against it before validating. When strangers email you asking for a service, that is the highest-fidelity demand signal in software, and most operators waste it by doing the work instead of building the tool.

The Distilled Read

Productize the request. Never deliver it. When a stranger DMs you asking for a pitch deck, a logo, a contract review, or an SEO audit, they have just handed you a working spec for free. The wrong move is to quote them $500 and disappear into Figma for a week. The right move is to ask how many other people are sending that exact DM to other strangers right now, and ship the cheapest possible tool that answers it. The deck this guy did not build was worth maybe $1K. The tool he built instead is on a $200K ARR pace and growing while he sleeps. Service-trap operators trade hours for dollars. Tool operators trade weekends for compounding cash.

A presentation is a template wearing a costume. Every deck is title slide, agenda, three problem slides, three solution slides, traction, ask. The AI's job is not creativity. It is filling slots. Once you see that, MagicSlides stops being an AI product and starts being a templating engine with an LLM bolted on the front. That reframing is what lets you ship in a month instead of two years. The same playbook is sitting unbuilt for resumes, cover letters, sales one-pagers, board updates, contracts, term sheets, and lesson plans. Pick the document, find the latent template, wrap an LLM around it, ship.

$17K MRR with one founder beats $5M ARR with thirty. Run the unit math. Solo operator. AI inference likely under $2K/mo at this volume using GPT-4o-mini for outlines and pre-rendered templates for layout. Stripe takes 3%. Hosting on Vercel and R2 is rounding error. Net margin is plausibly 80%+. He is pulling roughly $13K/mo in profit out of a tool he does not have to staff, sell, or onboard. Tome raised $43M and pivoted twice. The bootstrapped solo bet on a boring use case beat the VC-backed solo bet on a sexy one, not because the founder was smarter, but because the structure of the bet was different. Capital constraints forced him toward the most cash-generating version of the product. Capital abundance forced his competitors away from it.

Distribution is languages and integrations, not Twitter threads. He is not posting threads. He is ranking on long-tail "AI PPT" queries in 136 languages, and he is living inside Google Slides and ChatGPT as a plugin. Each language is a Google. Each integration is a distribution surface he did not have to pay for. While English-language indie hackers fight each other for the same fifty keywords, he owns the equivalent of those keywords across the rest of the planet. If you are an English-only operator in 2026, you are voluntarily competing in the hardest league for zero extra revenue.

The moat nobody named is file format compatibility, and it is about to widen. When Microsoft Copilot finally ships native deck generation inside PowerPoint, every browser-native AI slide tool dies overnight, because they do not export to Office cleanly and the corporate user just abandons them mid-quarter. MagicSlides survives because its output is already a native .pptx file. The user does not switch tools, they just get the same file faster. The AI-native slide companies built their products on the assumption that the deliverable would change. It will not. The .pptx format has survived 30 years of "PowerPoint killers" and will survive this one too. The operator who bets on the boring file format wins the decade.

Steal-the-Playbook

1. Pick one document type, not the category. Pitch deck. Resume. Sales one-pager. Board update. Lesson plan. One.
2. Mine your DMs and your competitors' Discords for the cold-request pattern. When 5+ strangers ask for the same artifact, that is your spec. Stop validating. Build.
3. Map the document to a 6-12 slot template. LLM fills slots. You render natively to .docx, .pptx, or .pdf. Never ship browser-only output.
4. Launch in 5 languages on day one using DeepL plus your prompt template: English, Spanish, Portuguese, Hindi, Indonesian. SEO surface area equals revenue surface area.
5. Free tier with a watermark, plus a one-click ChatGPT and Google Workspace integration. The free tier is the marketing budget. The integration is the distribution.

Stack: Next.js + GPT-4o-mini + python-pptx + Stripe + Vercel. ~$40/mo to run. Weekend to v1, 14 days to first paid user.