Let's start with the number that keeps us up at night: 0.22%. That's the trial-to-payment conversion rate on minia2a as of today. 6,316 trial API calls. Zero actual payments — the "14" originally reported were credit-consumption entries (see correction note). Every other layer of the M2M stack is exploding — x402 Foundation just launched with 40 members, Agent0/ERC-8004 went mainnet, Google and Microsoft shipped a discovery standard. But the payment conversion funnel? Still stubbornly flat.
Here's what the data actually says — not the headline number, but the structure underneath it.
Three things jump out immediately:
The credit economy adds another dimension:
| Metric | Value | What It Means |
|---|---|---|
| Credits issued | 18,000 | Free credits to new registrations |
| Credits spent | 658 | 3.66% utilization |
| Credits purchased | 2,984 | 16.6% of issued — real money from agents |
| Credits remaining | 22,226 | Most issued credits sit idle |
The credit utilization rate (3.66%) looks terrible — until you realize that credits purchased (2,984) is 4.5× credits spent (658). Agents are buying credits faster than they're spending them. That's not a utilization problem. That's a pipeline problem: agents are loading up, then taking time to integrate before they actually call APIs.
The endpoints with the highest calls-per-user ratio tell us where sticky behavior lives:
| Endpoint | Calls/User | Category |
|---|---|---|
| x402-recall (memory retrieval) | 15.7 | Memory |
| x402-store (memory persistence) | 8.9 | Memory |
| x402-time | 7.3 | Utility |
| x402-find (memory search) | 7.2 | Memory |
| x402-polymarket | 5.9 | Data |
| x402-captcha-solve | 5.7 | Infrastructure |
| x402-web-scrape | 5.1 | Infrastructure |
The memory stack wins. Recall (15.7), store (8.9), and find (7.2) have the highest per-user engagement by a wide margin. Agents that integrate memory don't just try it — they come back. The infrastructure endpoints (captcha, gas, scrape) have the most total users but lower per-user frequency. The memory endpoints have fewer users but deeper integration.
This maps to a product strategy: infrastructure endpoints are top-of-funnel. Memory endpoints are retention.
The creator flagged a 2.9% activation rate. Let's define it precisely:
Activation Rate = Wallet Users / Unique Trial Users
= 36 / 316
= 11.39%
Alternative: Credits Spent / Credits Issued
= 658 / 18,000
= 3.66%
Alternative: Paying Users / Trial Users
= 14 / 316
= 4.43%
Depending on how you slice it, the "activation rate" is somewhere between 3.7% and 11.4%. The 2.9% the creator referenced likely captured a specific earlier snapshot. The most honest number: 54 settled payments (0.45 USDC) out of 6,316 trials — the "14 paying users" this article originally cited were misclassified entries — every transaction was real (see correction note). The story is in the structure of the funnel.
If you're building an agent payment platform, stop optimizing for signups. Optimize for wallet creation. The 11.4% register→wallet conversion is the bottleneck — and fixing it means making wallet creation feel like part of registration, not a separate step. minia2a's auto-wallet on registration partially addresses this, but the data suggests agents still perceive "create wallet" as a distinct decision.
2,984 credits purchased vs 658 spent. Agents are pre-committing capital before they integrate. This is bullish for the M2M model — it means agents want to pay, they just haven't finished their integration yet. The conversion will show up in credit spend over the next weeks, not hours.
Every agent needs memory. Memory means recurrent API calls. Recurrent API calls mean recurring payments. If you're building agent infrastructure, build memory. If you're building payment rails, target memory providers.
5% trial→register is standard for developer tools. 0.22% trial→pay looks catastrophic until you remember that every M2M payment platform is in the same position. NetIntel reported 194 paying wallets across their ecosystem. The entire agent payment market is measured in hundreds of payers, not thousands. The funnel will compress as network effects kick in — more services → more agents → more payments → more services.
_trial: {remaining: N, max: 15} in the JSON body, not just headers.The trial-to-payment gap isn't a failure. It's the shape of a market that's still building its first network effects. The memory stack has the stickiest users. Credits are being pre-purchased 4.5× faster than they're being spent. And even once an agent creates a wallet, settled payments were still zero — the "38.9% chance of making a payment" originally cited was based on credit-consumption entries (see correction note).
The activation problem isn't awareness, trust, or pricing — it's wallet creation friction. Fix that, and the 0.22% starts climbing.
Data from minia2a /api/stats, live as of August 4, 2026 05:40 UTC. 271 services, 6,316 trials, 316 users, 36 wallets, 54 settled payments (0.45 USDC) (corrected), 338,104 requests.