m.a.d labs

Data · Power BI · Snowflake

Follow one number.

Every decision in your company is a number somebody trusted.

m.a.d labs makes the number worth trusting: modelled once, owned by someone, read without a manual, and ready for whatever you point AI at next.

Scroll to follow it

Net revenue, Q3

  • €4.2MFinance deck
  • €3.9MSales dashboard
  • €4.0MWarehouse query

Same question. Three answers.

“Why isn’t yesterday in yet?”The analyst, every morning

An analyst frowning into an open laptop
lands once

Sources

The pipes eat the team.

What’s going wrong

One hand-built pipeline per source, each written by someone who has since left. Large enterprises report 4.7 failures a month, each taking nearly 13 hours to resolve. Your engineers keep the pipes alive instead of building.

Figures fromFivetran, enterprise data infrastructure benchmark, 2026

The fix

Ingestion that is configured, not hand-coded. Raw data landed once, unchanged, in an open table format, with freshness checked at the door and schema changes caught where they enter.

What we useSnowflake, Fabric or Databricks as the landing place; SQL.

“Why did the bill double?”The finance director

A finance director reading a bill as long as he is tall
cost has an owner

The platform

The warehouse bill grew faster than the insight.

What’s going wrong

Three in four organisations exceeded their cloud budget last year, and eight in ten data teams say they cannot forecast their data platform costs. Compute runs for queries nobody is waiting on, and everything is rebuilt every night.

Figures fromForrester, via CIO Divedbt Labs, State of Analytics Engineering, 2026

The fix

A warehouse designed around the questions actually asked: right-sized compute, incremental builds, and the cost of every model and query visible to the people who own it. The platform is chosen on your workloads, and open table formats let one platform read another’s tables instead of copying them.

What we useSnowflake, Databricks, Fabric.

“Who changed this number?”The financial controller

A financial controller checking figures on a clipboard
tests on every model

The models

Bad data has a price nobody measures.

What’s going wrong

Gartner puts the average cost of poor data quality at $12.9 million a year, and finds that six in ten organisations do not measure it. The logic lives in stored procedures, notebooks and report formulas, with no tests. AI now writes SQL faster than anyone checks it: 72% prioritise AI-assisted coding, 24% AI-assisted testing. The wrong number is usually discovered in the room where it matters most.

Figures fromGartner, via ZoomInfodbt Labs, State of Analytics Engineering, 2026

The fix

Every transformation is a SQL model in dbt, under version control, with tests on every model, CI on every change, a named owner, and lineage you can follow from the source to the screen. A wrong number fails a build before it reaches a board pack.

What we usedbt, SQL.

“Which revenue is right?”The board

Three board members, each holding a different chart
one definition

The meaning

Two versions of the truth.

What’s going wrong

Fifty-eight percent of business leaders say their organisations make important decisions with inaccurate or inconsistent data. The usual cause is not bad data but two definitions: “revenue” in Power BI, “revenue” in Excel, “revenue” in the warehouse, each a little different, each defended in the same meeting.

Figures fromSoftServe survey, 2025

Further readingGartner, predictions for data and analytics in 2026dbt Labs, the Open Semantic Interchange specification

The fix

One semantic model. Every metric defined once, in code, with a named owner. Power BI reads the definition; it does not invent one. The definition lives where your platform keeps it: a Power BI semantic model on Fabric, the dbt Semantic Layer, semantic views in Snowflake, metrics in Unity Catalog on Databricks.

What we usePower BI, dbt, Snowflake, Databricks, Fabric.

“Where do I look?”The head of operations

A head of operations scratching his head over a tablet
answers a question

The dashboards

The dashboards nobody opens.

What’s going wrong

Active use of business intelligence tools has sat near 29% for seven years, while organisations keep reporting more analytics. People open a dashboard once, find it does not answer the question they had, and never come back.

Figures fromGartner, via Business Analysis

The fix

Fewer dashboards, each built to answer a question somebody actually asks, readable without a manual. Retire the rest.

What we usePower BI, Fabric.

“Can I just ask it?”The chief executive

A chief executive asking her phone a question
reads the model

The agents

AI is waiting on your data.

What’s going wrong

Gartner expects organisations to abandon 60% of AI projects that are not supported by AI-ready data, and 63% say they do not have, or are not sure they have, the data practices AI needs. The model is not the problem. The data under it is. Without a governed model an agent writes its own SQL, and its mistake looks like a plausible number.

Figures fromGartner, 2025Gartner, May 2026

Further readingdbt Labs, Semantic Layer vs. text-to-SQL, 2026

The fix

The same governed, documented model that feeds the dashboards feeds the agents. Lineage, quality and ownership are what make data AI-ready; nothing else does.

What we usePower BI, Snowflake, Databricks, on the one model.

  1. Sources
  2. Warehouse · dbt
  3. Semantic model
  4. Dashboards & agents

How m.a.d labs works

From source to screen.

How m.a.d labs works: the lineage is drawn first, then built. Nothing is built until the drawing is agreed.

First, we understand.

Working sessions with the people who own the numbers: finance, operations, the product lead, the analysts who answer the questions today.

  • Not a requirements capture; a joint look at what the business needs to know, who needs to know it, and what a wrong answer costs.
  • The output is a model of the problem (the metrics, their owners, their sources, their grain) and a shape for the solution: the warehouse layout, the semantic model, the order in which dashboards are rebuilt.
  • A shape both sides agree is the best answer. Nothing is built until it is.

Then, we build.

Senior engineers go to work on that shape, and only senior engineers.

  • Modelled in dbt with tests on every model and CI on every change; one versioned semantic model that Power BI reads.
  • Documentation is a first-class deliverable: a metric dictionary, lineage from source to screen, runbooks, onboarding a new analyst can follow on day one.
  • The platform is built so your team can add a metric without m.a.d labs, whenever you choose.
  1. lands once
  2. cost has an owner
  3. tests on every model
  4. one definition
  5. answers a question
  6. reads the model

The number, trusted

One number, one owner.

What the board pack looks like when we are gone, and what you are holding.

Net revenueQ3
€4.2M

Invoiced revenue net of refunds and credits, recognised on invoice date.

Owner
Head of FP&A
Source
Snowflake · dbt · fct_revenue
Freshness
refreshed 6 min ago
Quality
14 of 14 checks passed

Finance warehouse on Snowflake

  • 38 → 1reports into one model
  • 6 minrefresh
  • 100%of metrics with a named owner

What you leave with

  • A Snowflake warehouse modelled in dbt, with tests and CI on every change
  • One semantic model for Power BI: every metric defined once, owned by someone
  • Dashboards people open, each answering a named question
  • Lineage from source to screen, and quality checks that fail before the board pack
  • A metric dictionary and runbooks, in a repository that is 100% yours
  • A team of yours that can add a metric without us
Talk to the data team

Warehouses, semantic models and dashboards your whole company can read without a manual.

Data consultancy. Power BI, Snowflake, dbt, Fabric, Databricks, semantic model, semantic layer, metric definitions, single source of truth, data quality, lineage, data governance, self-serve analytics, AI-ready data, warehouse cost, FinOps for data, senior data engineers, documentation, handover.

Services: Power BI & semantic models, Snowflake, dbt & Fabric, Governance & lineage, Self-serve analytics.

Stack: Power BI, Snowflake, dbt, Fabric, Databricks, SQL.