Unisig. The crypto-signals dashboard that doesn’t sell a win rate.

Mid-2023. Unisig was eight months into running a paid signals service through Discord, algorithmic alerts on technical setups across BTC, ETH, and a rotating list of mid-cap pairs, delivered to a community gated by a one-time NFT mint. The four founders all came from creator backgrounds with audiences they’d brought across; the engineering work was thin and the design work hadn’t started. The brief on the table was a vision document titled “Unisig Pro”, a dashboard with live charts, leaderboards, a dedicated mobile app, a marketed win rate, a copy-trade feature. The ambition was Bloomberg-for-retail-crypto. The pressure was a bear market that had quietly dropped Discord activity by sixty per cent over the prior quarter.

The first week we didn’t open Figma. We sat with eighteen Unisig members across paid call windows, day traders in Sydney and Auckland, swing traders in Berlin, a small cluster of US-based members who’d come over from another signals service. We watched them open the Discord, scroll for thirty seconds, and close the tab without acting. That was the product, working as designed. The chronic pain wasn’t dashboard ugliness or feature count. It was three things stacked. The signal volume had outpaced anyone’s ability to filter. Nobody was sizing positions consistently because the signal said direction but never how much. And the homepage’s wins-only performance carousel was being read by the community as a tell, not a credential. Traders weren’t asking for a better dashboard. They were asking for fewer signals, more context, and an honest story about which ones had not paid off.

We built the dashboard. We also rebuilt what it would and wouldn’t show. Three design decisions did most of the load. First, every signal carried a conviction tag, a system score plus a trader-set filter rule, so members could mute low-conviction signals without leaving the platform. Second, a position-sizing widget sat inline with each signal: enter your account size and risk-per-trade once during onboarding, and every subsequent signal told you the size to take, not just the direction. Third, the performance feed showed misses next to hits, sequentially, with a hindsight-vs-real-time toggle that made survivorship bias hard to maintain even by accident.

Twenty-four months on, ninety-day member retention sat at sixty-two per cent against a baseline of twenty-eight per cent at engagement start. Signal-acknowledgement rate (the share of signals members marked taken or skipped in the dashboard) climbed from twelve per cent to forty-seven. Support tickets tagged “Discord noise” dropped to roughly a third of their pre-engagement volume.

Two confounds we won’t paper over: the NFT-gated access model self-selects for active members, so any retention number sits inside that filter; and the bear-to-bull transition through 2024 produced market-wide engagement gains that attach to market conditions, not to design. Twenty-four months is also still inside one full crypto cycle, the test we’d take seriously is whether retention holds through the next bear, not just the recovery.

The takeaway: in trust-sensitive products, what you choose not to display does more work than what you do. Win-rate billboards, leaderboards, and follower counts are the assets easiest to optimise. They are also, in retail finance specifically, the assets that most reliably attract the audience you can’t keep. Audit which numbers your product can defend in the bad months, not just the good ones. Show the rest at your own risk.