What we believe

Every AI analytics tool talks about trust now. Most of them reached for the word when they learned it was do-or-die in deals. We obsess over it because we care deeply about data as a practice, and a data function the business doesn't trust might as well shut down shop.

Trust is foundational to how humans relate and collaborate, at home and at work. It takes years to build and can be destroyed in a second. Nobody sorts successes and failures into two jars and checks which one holds more coins at the end of the year; that is not how trust works.

It is on the line in every single interaction.

If trust is always on the line, and no one will ever be 100% correct, then how is it maintained? Not by accuracy, but by behavior.

Honesty when unsure, learning from mistakes, transparency when something went wrong. These are the things that build and keep trust. Translating this into product behavior is much harder than tuning a model for benchmark accuracy, but it's the path we believe necessary to build a truly trustworthy AI analyst.

We will not build a harness that competes on analysis benchmarks. We will build a harness that crosses our incredibly high bar for trustworthy behavior.

(Though we reserve the right to show off if we top benchmarks as a byproduct.)

Ask how many active customers you have. Marketing counts everyone using the product: their job was bringing in people with the problem, and usage is the proof. Finance counts everyone paying, because revenue is what they report. You can't force either team onto the other's number; both versions have to exist at the same time.

The faster your teams move, the faster new definitions appear, and the more they conflict. It's the result of each project or team tuning the number it's trying to improve.

We've seen data teams fight this to keep the metric layer clean. They end up on an island: far from the business, ignored by frustrated stakeholders, their metric layer gorgeous, and useless.

Sienna allows for competing definitions, each with its own wording and SQL. Any team wanting to use a specific version is free to, and warned of the alternatives.

The goal is stopping people from comparing apples to oranges, not making oranges illegal so you never have to worry about them.

The only place we have ever seen perfect data is a stack where every row is written by software, end to end, with no human anywhere in the path. Everything else has issues: CRM fields filled in by hand, spreadsheet uploads, free-text columns. However hard a team fights it, the issues come back, and most businesses aren't carrying a few. They are carrying piles.

So a data tool has to be honest with itself about that. A tool that assumes clean data assumes a dream.

We don't want the dream. We want the thing that works in the real, gritty world.

A strong analyst remembers business decisions, not templated queries. They then know how to turn those decisions into a query that instantiates them correctly for the current context. That's what makes them versatile, and able to answer follow-on questions.

So Sienna does not build memory fit for a compilation engine. It builds memory best adapted to store business concepts, with just enough connective tissue to point to the right areas of the warehouse that answer them.

Rigidity is a solution for stupidity, and we have already entered the age of cheap intelligence.

"A computer can never be held accountable. Therefore a computer must never make a management decision."

IBM training manual, 1979

Yeah, we've heard it, and disagree with it. A good leader does not make every decision in their team, yet they still hold themselves accountable for all the outcomes. They build systems where important decisions get flagged to them, so they can overrule them if needed.

That is how a decision-maker leverages themselves. Not by holding on to each decision.

AI will provide as much leverage as it is allowed to, and we strive to provide the highest leverage for those who want to embrace the opportunity.