Seeking Feedback: 2% of our signups pay. 40% of our weekly actives do.
Constants (constants.io) is a creative studio with a AI tool builder bend. Make music, images, video, UGC, podcasts and decks, or build your own tool and re-run it. It works standalone, or inside Claude and ChatGPT over MCP.
I want to show you what we built, share the numbers, and then ask for honest feedback. Please roast away!
The numbers
- 2,505 creators
- 4,004 tools made
- 31,954 creations
- $11k+ ARR, from monetization we switched on four weeks ago
- 2% of signups convert, but 40% of weekly actives do
Almost half of the people who use us in a given week are paying. Very few actually stick around.
A bit about how we got here
Constants started as an agent-agnostic capability layer: build your own tools, skills and connectors, invoke them from any AI.
Most our our users repeatedly generated media and saved the recipe. So we pointed the product at that. I wouldn't call it a pivot, but more of focusing the platform at an initial vertical. A very crowded vertical.
What our most active users actually do
When I sorted repeat and paying users into groups, I got seven. They overlap at the edges, but not in a way that gives me one market:
- Lyricists without a band. They write the words themselves and need someone to perform them. They want their lyric sung well.
- Catalogue builders and labels-of-one. Running a real or invented act as a body of work. They want one consistent sound across many tracks.
- Occasion and tribute songwriters. One song about one named real person, for a birthday, a funeral, a thank-you.
- Builders. Turning a repeated job into a tool. They want the workflow to persist so they stop re-describing it.
- Art directors of one. Many images on-model to one locked character, brand or style.
- Owner-operators making their own ads. One-person businesses producing their own marketing, wanting brand-accurate assets without hiring anyone.
- Worship and church media. Music and imagery for a congregation against a weekly deadline, tied to a specific passage.
Where we are stuck
We are trying to find our blue ocean. The traction with these tribes is nice to have. No tribe is big enough to be a market. None of them are also defensible enough. There are alternative solutions and generation is commodity.
We believe the direction to go is focus on tool building and chat experience that beats Claude and ChatGPT in workflow type work. Think Perplexity's focus on research. However, that's a product that does "everything", which is hard to market to anyone.
What we're trying next
- Attract more tool builders, on the theory that the tools they publish bring organic traffic.
- A community space where users can share the tools they've made with each other and make these tools easier to discover.
- Targeted ads per tribe, lean into the specific avatars we see on our platform already.
What I'd love your take on
- Roast constants.io. The page, the service, the idea, the strategy. What does not make any sense to you?
- If you've built something horizontal, what strategy did you find to grow? How did you choose?
- Has anyone treated an MCP or agent-directory listing as a real acquisition channel yet? What did it convert like, and how did you attribute it?
Happy to answer anything about numbers, stack or margins in the comments.
Taking the "roast the page" part, since the activation and tribe comments above already cover the rest.
The first screen is written for nobody in particular, and that is the 2% right there. "Your Creative Studio, In One Chat. Every tool you need to bring ideas to life. From music and video to UGC and custom apps." A lyricist without a band, a church media volunteer and a one-person shop making their own ads all read the same sentence and none of them sees themselves in it. A page that promises everything lets everyone sign up, and 98% of them wanted nothing specific enough to come back for. Your 40% are the people who arrived already knowing which of the seven jobs they came for; the page did not do that for them, they did it themselves.
The proof on the first screen is activity, not outcome. "Creators 2,615 / Tools Made 4,021 / Creations 32,444" tells a stranger that people press buttons here. The number you actually have that a stranger would care about is in this post and not on the page: almost half of weekly users pay. Say that, in plain words, next to the button.
Every "Try this prompt" card ends with "I'll attach the source audio / image / PDF". So the zero-effort first action, the one a stranger takes in the first minute, does not exist: each example needs an asset they have to go and find. That is a plausible reason session one ends before the first creation. One card that needs nothing attached, at the top, is the cheapest test of that.
And the first screen asks a visitor to decide where to use the product ("On Constants or inside Claude and ChatGPT") before they know what it does. Move the MCP line down; it is a reason to stay, not a reason to start.
To your horizontal question: you already have the answer in your own data - seven doors, one house. Each tribe gets a first screen written in its own words, all of them into the same product. The page you have now is the hallway.
Thank you! That is very good feedback. I will incorporate some of it in our home page.
The 2% vs 40% gap is pure measurement boundary visualization. You're measuring two populations in different scopes: "people who ever clicked a button" (2%) vs "people who voluntarily came back in the last week" (40%). These aren't different rates of the same thing - they're completely different measurement domains.
This is why founders see themselves as failing when they're actually filtering. The signup rate looks terrible until you add the "activated and returning" boundary. Your real product is serving the 40%, not the 2%. That's not a conversion problem - that's a measurement scope problem.
Most product teams optimize to improve the 2% when the real insight is "we have excellent retention among returners, so our bottleneck is getting people to session 2." Those are opposite problems with opposite solutions. The 2% metric is measuring distribution of ambition (how many people tried once), while the 40% metric is measuring actual product-market fit (what happens to engaged users).
What would be your recommendation?
The 2% vs. 40% conversion gap is the most interesting signal here.
Curious whether the challenge is really finding a blue ocean, or deciding whether one of those seven tribes is already strong enough to become the wedge.
The 2% vs 40% gap reads less like a pricing problem and more like an activation problem: people who reach a second session pay at a great rate, so the leak is between signup and first repeat use. I'd instrument the exact step where first-session users drop (before first generation, after first generation but before saving a recipe, etc.) and treat "saved a reusable tool in session 1" as your activation metric — that's the behavior your payers share.
On the horizontal problem: I wouldn't try to pick the blue ocean analytically. Of your seven tribes, occasion/tribute songwriters and worship media are the two with a real deadline and an emotional payoff, which usually means less price sensitivity and easier messaging. Building a dedicated landing page and onboarding flow per tribe (same backend, different front door) is cheap and gives you real conversion data instead of a guess.
The community/tool-directory bet is fine but slow; published tools only bring organic traffic if each tool page is indexable and answers a specific query. Worth checking whether those pages actually render for crawlers before you count on that channel.