Show IH: Saemi — AI data assistant for PLG SaaS, free tier, looking for 10 testers
Hey IH 👋
Solo founder from Korea, 3 months in. I kept watching small SaaS teams (5-50 people) burn their analyst's week on "what's our W4 retention by plan?" — because the PMs and CSMs couldn't query the data themselves. So I built Saemi.
What Saemi actually does — the part most "AI Slack bots" skip:
Multi-step causal analysis, not single-query answers. Ask "why did pro plan churn spike in May?" — Saemi joins Stripe + PostHog, runs 4-5 SQL queries, surfaces the correlated feature rollout, builds the chart. ~90 seconds.
Verified knowledge layer. Your team's metric definitions (ARR, NRR, activation) get cited inline so answers don't drift between teammates.
Datasets + dashboards + wiki, not ephemeral chat. Every analysis becomes a reusable artifact you can pin and rerun.
Skill Builder. Turn recurring questions into named workflows (
/weekly-cohort-report,/anomaly-check) anyone on the team can trigger.Read-only DB role + SQL shown inline. Your CTO can audit every query Saemi ran. Not stored, not used for training.
Why it's different from existing tools:
vs Genie / Cortex — warehouse-agnostic (Postgres, Snowflake, BigQuery, Databricks, Athena). Most small SaaS aren't single-cloud.
vs Hex / Mode — chat interface PMs and CSMs actually use, not analyst-only.
vs other Slack data bots — handles real analysis (joins, cohorts, anomalies), not
SELECT COUNT(*).
Free tier: 200 questions/month, unlimited charts/datasets/dashboards, 1 Slack channel. No card.
What I need: 10 testers. In exchange:
1:1 30-min onboarding (I screen-share, watch you use it)
Lifetime 50% off Team tier ($79 → $40/seat/month)
Direct line to me — broken? fixed in 48h
Honest admissions: solo from Seoul (async-friendly), 0 paying users yet, onboarding has rough edges, bus factor = 1.
Reply here or DM. I'll respond to every one.
What stands out to me isn't the onboarding or even the lack of paying users yet.
It's the fact that several very different problems could produce the same early feedback from testers.
That's what makes this stage tricky.
A product can get positive reactions, useful conversations, and even repeated usage while still leaving the most important decision unresolved.
The part I'd probably spend the most time on is figuring out which conclusions actually deserve confidence before the feedback starts pointing the roadmap in a particular direction.
That's exactly the trap I'm trying not to walk into. The same "yeah this is helpful" can come from "you solved my actual bottleneck" or "cool toy, happy to support a solo founder" — both feel identical in the moment.
Heuristics I'm holding onto for the first 10 onboardings:
Ignore week 1 (novelty). Watch which questions they ask unprompted in week 2.
Ask what they did last Friday afternoon before showing the product — were they actually unblocked, or did they just give up?
The hesitation before they answer "what would you pay" matters more than the number itself.
Curious what's worked for you — is there one tester behavior you've learned to trust over the rest?
What makes that difficult is that the behaviors I end up trusting tend to depend on the decision sitting underneath the feedback.
That's why I hesitate to answer it casually.
The useful part usually isn't the behavior itself.
It's understanding which decision that behavior should increase confidence in.
I wouldn't try to unpack that properly in a thread.
If you're curious, drop your email and I'll send over the tighter version.