FAQ
Frequently asked questions
Clear answers about wallet credit, usage, subscriptions, and how Tycoon charges for work.
How do you handle statistical significance for low-volume SaaS where signups per week are double digits?
Most founders at Tycoon's stage don't hit classical 95% significance on pricing tests because sample is thin. The AI CFO is explicit about this: it publishes both the frequentist p-value and a Bayesian posterior (e.g. 'there's an 82% probability the variant lifts MRR by ≥20%'). For small-N programs it also recommends tests with larger expected effect sizes (30%+ price changes, not 5%), accepts looser thresholds with wider confidence intervals, and uses sequential testing so you can call a test when the posterior converges rather than waiting for a fixed N. The honest framing: you're not going to P&G-level pricing rigor at 50 signups/week, and the AI is upfront about the confidence you actually have.
What happens if a test shows churn rising — do you auto-revert or wait for me?
Only guardrails with pre-defined harm thresholds auto-revert. You set these in the experiment design: e.g. 'auto-stop if free-to-paid conversion drops more than 30% for 3 consecutive days' or 'auto-stop if refund rate exceeds 8%.' Crossing an auto-stop flips the feature flag off within minutes. Anything ambiguous — churn ticking up 2 points over a week — escalates to you with the evidence and a recommendation. The design principle: automate protection against catastrophic outcomes, require human judgment for soft signals. Founders who want more automation can configure tighter auto-stop thresholds; founders who want more control configure looser ones with alerts instead.
Can it handle multivariate tests — testing price AND feature packaging at the same time?
Yes, with a warning. The AI will design 2x2 or higher factorial tests when you ask, but it's vocal about the sample cost: doubling factors quadruples the sample needed. Most of the time it pushes back and recommends sequencing: test the biggest hypothesis first (price), lock the winner, then test packaging. This is a real disagreement you'll have with the AI sometimes — you want to learn fast, it wants to learn cleanly. You can override and it'll ship the multivariate test, but the post-mortem will show you why the cleaner sequential approach usually beats it.
How do I run pricing tests without damaging SEO on the /pricing page?
The variant pricing gets served from a split URL (e.g. /pricing?v=b) with rel=canonical pointing to the main /pricing page. Googlebot sees the canonical version; test traffic gets the variant. The AI Head of Growth makes sure structured data (Product schema, Offer schema) on the canonical page stays pinned to the control price, so rich results don't flip mid-experiment. When a winner rolls out, the canonical page updates and Search Console gets a resubmit ping. You won't see ranking whiplash from the test because no search-indexable URL ever changed price during the run.
What about annual vs monthly — does the AI factor in payment terms?
It has to, or the reads are noise. Annual subscribers drop a year of revenue in month 1 and then nothing for 11 months; a test that pushes more annual signups will look amazing on 'first 30 days MRR' and terrible on 'net new MRR rolling.' The AI CFO always computes ARPA on an annualized, plan-adjusted basis — breaking out monthly vs annual cohorts, reporting each separately and as a blended LTV-adjusted number. It also watches for substitution effects: did we lose 20 monthly customers and gain 8 annual ones? That's a net win in revenue but you'd miss it looking at customer count alone.