RiskOps Lablive
A simulator of Fraud/AML analyst work: the user investigates a synthetic case from alert to decision in an interface that looks like an internal risk admin.
FinCrime has plenty of theory: courses, certifications, typologies. What is missing is a way to do an analyst's job by hand: open an alert, find signals in transactions and events yourself, make a decision, and argue it.
In companies people learn this only after they get access to real customer data, and training cases mostly end in a test with a right answer. RiskOps Lab lets you go through that process earlier and without risk: on data from a fictional fintech, with a written rationale instead of picking an option.
This is a deliberately niche product: its audience is limited to people entering Fraud, AML, or KYC, or already working there at the start of a career.
RiskOps Lab was my first product. I built it without product experience and without discovery: the idea grew from a gap between theory and practice that I noticed myself. The most important thing I took from the project: what should have been tested before development, I learned after launch.
First such insight: products with a similar idea already existed (AML simulators, practical labs, some of them paid). That changed the positioning: not “an idea with no analogues,” but a category in which RiskOps Lab differs by format. It is a working environment, not a course or a test.
Second: an AML specialist I showed the product to said they would pitch a tool like this to the business as a training environment for analysts. That is not B2B validation, it is a scenario I had not planned: onboarding junior analysts inside companies. So I want to start the next iteration by testing B2C and B2B hypotheses, not by adding features.
Launch 20.04.2026: a LinkedIn post and basic SEO. There was no paid promotion. Amplitude data as of 15.09.2026. The numbers are not final: ad blockers on the user's side cut some of the events.
Channel. External referrals come from Google, Bing, ChatGPT, and LinkedIn. After April I did not develop the product, but people still find it and register every month (in June-September, 10 to 17 real visitors a month). Conclusion: the organic channel works without support, but for a niche this narrow search alone is not enough to produce a noticeable flow.
Audience. Of 840 unique visitors, most were bots and accidental hits. After cutting visits shorter than 3 seconds, 176 real visitors remained. Of those, 23 registered (13%, N = 176), and 15 created a case review. Conclusion: filtered traffic converts reasonably well, and those who register mostly reach the core action. The bottleneck is not entry, it is the volume of targeted traffic. The sample is small and self-selected, so this is a signal for the format, not proof of demand.
In the first version, after sign-up the user landed straight in the working environment: client profile, transactions, events, and alerts at once. I noticed that for a candidate or junior specialist this was overload: unclear where to start and what to look for.
Instead of simplifying the interface, I added a step-by-step tutorial on the working pages of the alert and the profile. The user walks the first case with hints in the same environment they will later use alone. The environment stays close to a real risk admin, and the barrier to entry gets lower.
I did not measure the effect separately: there is no before/after comparison. There is a signal in favor: over the last two months, every user who completed the tutorial submitted a rationale for review.
A fictional fintech: synthetic users, transactions, events, alerts
An interface in the style of an internal risk admin: alert, client profile, operations history
A step-by-step tutorial on the working pages
A case decision with a required rationale
Live feedback on every rationale (I review by hand)
Accounts and a progress dashboard
Before development I write a product spec (tutorial, dashboard, onboarding), then build it myself, using AI agents for the code. After launch I watch the funnel in Amplitude and review every case myself.
Now: the project is on pause, the product is available and still gets organic traffic.
Validated: the organic channel works without support; those who register mostly reach review (small sample).
Hypotheses for the next iteration (not started): packaging and channels that bring more targeted traffic into the niche; automating review with AI; a separate check of B2C (practice for candidates) and B2B (a sandbox to train analysts without real data); monetization (planned).
Next.js · React · TypeScript · Supabase · Tailwind CSS · Onborda