Ten years into my career, one thing keeps coming back to me. The most important data in most companies is the data nobody owns.
I've worked at Fivetran and Atlan. Both solve real problems for real customers. One gets your data from A to B, the other helps you govern the warehouse once it lands. But between the app code firing an event and the row landing in Snowflake, there's a gap neither of them fills. It's the gap where most analytics teams quietly lose their Monday mornings.
suky is what I wish I'd had.
What actually happens
I've yet to see a tracking plan that survives contact with reality. They live in a Google Sheet, a Confluence page, or a Notion database. They list checkout_started, signup_completed, feature_activated.
And almost universally, they're wrong.
Not maliciously wrong. Just quietly, inevitably wrong. An engineer renames a property from userId to user_id during a refactor. A PM adds a new event but forgets to update the spec. A mobile release ships with a typo in an event name that nobody catches for three weeks.
By the time someone notices, the damage is done. Dashboards break, attribution models skew, and someone on the data team spends Monday morning in a war room instead of building something useful.
The gap nobody fills
When I started digging into this, I assumed someone had already solved it. The data quality space is crowded.
There's Monte Carlo and Anomalo for observability. Soda and Great Expectations for testing. Avo and Iteratively for schema management. All good tools. None of them sit at the right layer.
Observability tools monitor tables in your warehouse. By the time they catch an anomaly, the bad data is already there — ingested, transformed, served to dashboards. You're doing forensics, not prevention.
Schema tools let you define naming conventions and generate SDK code, but they don't actually enforce anything at ingestion. The spec exists, and a malformed event still sails through.
CDPs (Segment, RudderStack, Snowplow) can enforce schemas at ingestion. The rails are there. They just need something pointed at them.
The missing layer sits between your tracking plan and your warehouse. It needs to do five things:
- Discover what events actually exist, not what the spreadsheet says should exist.
- Define enforceable contracts with types, required properties and SLAs.
- Enforce those contracts at the CDP, so bad events get blocked before they land.
- Monitor for drift, anomalies and SLA breaches in real time.
- Resolve violations by tracing them back to the PR that caused them, and where possible drafting the fix.
That's what suky is.
Why now
Three things had to be true for this to be buildable.
Snowflake made warehouse-native analysis practical. We don't ingest your events; we run read-only SQL against the data you already have. Your payloads never leave your infrastructure.
AI stopped being a gimmick. Writing event contracts by hand is tedious and thankless work. Claude can read your existing events, infer types, spot naming conventions, and draft contracts in seconds. It can also read your codebase and draft the PR that fixes a violation.
Schema registries matured. Segment Protocols, RudderStack and Snowplow Iglu all support real-time schema validation at ingestion now. The enforcement layer is finally there to point a contract at.
What suky is not
suky is not a general-purpose data observability platform. It doesn't monitor your dbt models or your ETL pipelines. It isn't competing with Monte Carlo.
suky is purpose-built for behavioural event data: the clicks, conversions, signups and page views that feed product analytics, marketing attribution and growth metrics. That's where schema drift hits revenue first, and where I think governance has the highest ROI.
And one more thing — we never ingest your raw event data. We connect read-only to your Snowflake warehouse. Only computed results (counts, rates, violation records) leave your environment. No PII. No payloads. Your data stays in your house.
What I hope this becomes
I've started companies that didn't work, and worked inside ones that did. The thing I keep coming back to: the companies that win with data aren't the ones with the most exotic stack. They're the ones where the analytics engineer, the PM and the growth marketer all trust that the same events mean the same thing.
That's the team suky is for.
If that sounds like yours, come take a look. I'd love to see what your event data actually looks like.
— Kevin