Tracewell
I designed the web app for Tracewell, an AML checker. Paste a wallet, learn whether to take the money.
Three months after launch, 8.6% of trial users pay for checks, up from 5.8%.
- RoleProduct Designer
- Timeline2 weeks
- Year2026
- PlatformWeb app
- Trial to paid
- +2.8 pp
- Score questions
- −35%
- Checks per paying user
- +37%
Challenge
A trader has a USDT deposit waiting. Take it, ask where it came from, or refuse? The old bot gave a score, 0 to 100.
38 of 100 tickets asked what it meant. The score came from a partner's engine, so I could change how it reads, not the math.
Approach
I read 124 support tickets and talked to 6 OTC traders. Most hard cases fit four wallets, so I tested every screen on them: sanctioned money, a clean exchange, a mixer, a new wallet.
The verdict says what to do
Each verdict ends in an action: accept, ask the owner, or refuse. Tracewell is for the trader taking the money, not a compliance team, so there is no case queue.
Incoming and outgoing stay apart
Here 95% of incoming funds are clean, but 38% of outgoing went to a mixer. A merged chart would call this wallet clean, so the two directions switch.
Every number adds up
Received minus sent equals the balance, and source shares add up to 100%. One wrong total and a trader stops trusting the score. Each source opens its transfers.
No score for a wallet without history
A score needs 10 transfers or 7 days. Below that the report says "recheck later", not 0, because people read 0 as safe.
Result
Traders see why a wallet is risky and what to do, so they check more deposits and ask support less. Three months of the web app against three months of the bot, about 9,400 checks.
| Metric | Telegram bot | Web app |
|---|---|---|
| Trial to paid | 5.8% | 8.6% |
| Score questions per 100 checks | 2.3 | 1.5 |
| Checks per paying user a month | 4.6 | 6.3 |
Next time
I would bring the verdict into the Telegram bot as well. Four in ten checks still start there.