Show HN: MarginDash – See which AI customers are profitable

Show HN: MarginDash – See which AI customers are profitable(margindash.com)

1 points by gdhaliwal23 94 days ago | 1 comment

I built MarginDash because I couldn't answer a basic question: which of my customers actually make me money after AI costs? My Stripe dashboard showed revenue going up. My OpenAI bills showed costs going up faster. I had no way to connect the two at the customer level.

MarginDash is a few line SDK integration (TypeScript, Python, or REST) that tracks model usage per customer and connects it to revenue — either through Stripe sync or by passing revenue directly in the API call. You get a per-customer P&L showing revenue, cost, and margin.

A cost simulator lets you pick any feature, swap the underlying model, and see projected savings. Models are ranked by intelligence-per-dollar using public benchmarks (MMLU-Pro, GPQA, AIME) so you can find cheaper options that aren't actually worse. Budget alerts email you before a customer or feature blows past a threshold.

The pricing database covers 100+ models across OpenAI, Anthropic, Google, AWS Bedrock, Azure, and Groq with daily updates — so cost calculations stay accurate without you maintaining a spreadsheet. The SDK only sends model name, token counts, and customer ID. No prompts, no responses.

Solo founder, built the whole thing for $239.72 in AI costs (wrote about that too). Currently free while I get feedback — would love to hear what you think, especially about the cost simulator.

gdhaliwal23 94 days ago |

Founder here. Gemini actually suggested I build this. I was brainstorming SaaS ideas and it pointed out the gap between AI cost tracking and revenue attribution. I dug into it, agreed, and built it.

The hardest part was the cost simulator. Comparing price-per-million-tokens across models is misleading — different models burn different amounts of tokens for the same task. So we normalize token counts to estimate what a swap would actually look like. When we recommend an alternative, we filter out anything that drops more than 10% on any benchmark or can't handle your context window size. Still improving this.

The SDK never sees your prompts or responses — just model name, token counts, and a customer ID. Limitations: simulator recommends from six vendors only, no custom/fine-tuned models, USD only.

Stack is Rails and Postgres. Happy to answer anything.