Blog

The arguments behind the product.

Why decisions go unmeasured, what a semantic layer actually does, why hiding a column is not access control, and when you should not buy from us. Filter by the problem you came in with.

Twenty-five pieces. Pick the one that sounds like your business.

BI cost and trust

The decisions nobody measures

Every reporting tool takes you to a number and stops. The gap after that is where the value went.

Governance and AI readiness

Governance that runs at query time

A policy in a document is a hope. A policy in the query path is a guarantee.

BI cost and trust

What a semantic layer actually buys you

One definition of revenue, enforced everywhere, and why the alternative costs more than it looks.

Building something data-heavy

Leaving should be boring

Open formats, an open catalog, and what you can still read on the day the contract ends.

Starting out

When not to buy Antvia

Four situations where we are the wrong answer, written down so you find out now rather than later.

Governance and AI readiness

Why the AI project is really blocked

The model was never the hard part. Nobody can say what it is allowed to see, or prove it afterwards.

Governance and AI readiness

What an auditor actually asks for

Not an architecture diagram. A specific answer about a specific field, with evidence nobody drew by hand.

Governance and AI readiness

Agents need identities, not API keys

A shared key tells you a query happened, not who asked or whether they were allowed to.

Governance and AI readiness

Catching bad data before anyone sees it

Monitoring tells you after the meeting. A gate stops the load and keeps yesterday's good data in place.

Governance and AI readiness

Mapping access to how the business works

Rules built around tables need rewriting every reorganisation. Rules built around the business survive one.

BI cost and trust

Why two dashboards disagree

Nobody built a wrong one. Two people built reasonable ones from different definitions.

BI cost and trust

What per-seat pricing does to access

Once every reader is a line item, who gets data becomes a budget conversation.

BI cost and trust

Four hundred dashboards, no trust

Sprawl is a symptom: making a new dashboard is easier than finding the right one.

BI cost and trust

Self-serve without a specialist for every chart

Most self-serve hands people raw tables and calls it freedom. That is why it fails.

BI cost and trust

Discipline loses to structure eventually

Certified models work as long as everyone follows them, which is until somebody is in a hurry.

Building something data-heavy

A platform without a platform team

Hire, then build, then ship costs quarters. And the first hire arrives with nothing to run.

Building something data-heavy

Why bronze, silver and gold hold up

It looks like ceremony until you need to rebuild against history you did not throw away.

Building something data-heavy

Analytics inside your own product

Different constraints to internal reporting: isolation, latency, and definitions you cannot quietly change.

Building something data-heavy

Isolation without a stack per customer

Separate deployments isolate obviously, and cost linearly. There is a middle option.

Building something data-heavy

The eighteen months nobody budgets

Estimates cover the parts you can imagine. The overrun is in drift, backfills and on-call.

Starting out

Why first attempts get rebuilt

The first version optimises for a quick dashboard. Those shortcuts are why it gets replaced.

Starting out

The spreadsheet is a source, not a failure

Every project has one somebody maintains by hand. Excluding it does not remove it.

Starting out

The first ninety days

Not a tool selection. Pick one argument the business keeps having, and follow the numbers behind it.

Starting out

Hiring a data team, or buying one

Usually framed as cost against cost. The useful frame is which risks you can carry.

Starting out

The smallest platform worth building

Below a certain point, a spreadsheet really is the right answer. Here is roughly where the line is.