Data engineering · vectorel

What it looks like

Three systems, three revenue numbers.
One of them is the one everyone uses.

1

CRM$1.31MClosed-won deals, including ones not yet invoiced

Billing$1.24MInvoiced, before refunds and credits

Finance spreadsheet$1.19MManual export, last updated 9 days ago

2 Revenue · Q3 · modelled layer

$1.22M

Invoiced, net of refunds and credits. One definition, written once in metrics/revenue.sql and used by every report.

3

Reconciled to billing
Fresh as of 06:00
0 duplicates

1
Every tool counts differently
Nobody's wrong, and nobody agrees. The difference is in the definitions, not the arithmetic.
2
One definition, agreed and in code
Agreed with the people who use it, version-controlled, and changed in one place when the business changes.
3
Tested on every load
Freshness, duplicates and reconciliation are checked before anyone sees a number. When a check fails, someone is alerted.

Illustrative example

What we build

What breaks in reporting,
and what we build to fix it.

Today
Every tool has its own version of the truth.
Your CRM, billing system and spreadsheets each count customers differently.

We build
A modelled warehouse.
Raw data turned into tables that match how your business works: customers, orders, subscriptions, margin. Versioned, tested, and readable by a human.
Today
The definitions live in someone's head.
What exactly is an "active customer"? Ask three people, get three answers.

We build
Agreed definitions, written once.
Revenue, active customer and churn defined in code, in one place, and documented in plain language anyone can read.
Today
Reporting is somebody's Monday.
Exports, lookups, manual joins, a spreadsheet rebuilt every week. It works until that person is away.

We build
Pipelines that run on their own.
Scheduled, monitored ingestion from your CRM, billing, product, ads and operational systems. When something fails, it retries and tells you.
Today
Nobody finds out when it breaks.
A failed sync or a changed field goes unnoticed for weeks. The dashboard still loads. It's just wrong.

We build
Data quality tests.
Automated checks on the things that quietly break: duplicates, unexpected blanks, totals that don't reconcile, rows that stopped arriving.
Today
The dashboard tool didn't fix it.
BI on top of messy sources produces fast, confident, inconsistent answers.

We build
Reporting on the modelled layer.
Dashboards and scheduled reports built on tested tables, so the numbers agree regardless of who opens them.
Today
People copy data between tools.
Export from one system, paste into another, and reconcile two exports every Monday.

We build
System-to-system sync.
Data flowing between tools on a schedule, with validation and logs, and an alert when something doesn't match.

Use cases

Reports it fixes

Leadership reporting
Revenue, pipeline and retention assembled by hand every week, then defended in the meeting.
One weekly view built from the source systems that nobody has to assemble or defend.
Finance
Billing reconciled against operations in a spreadsheet at month end, with exceptions hunted for.
Reconciled automatically every day, with the exceptions listed.
Sales and marketing
Each channel reports its own numbers, so channel performance is an argument.
Spend, pipeline and closed revenue joined end to end, so channel performance is one number.
Operations
Problems show up in the weekly report, after they've already cost something.
Live metrics from the systems that run the work, with alerts when something moves outside its normal range.
AI and ML projects
Pilots stall because the data underneath isn't clean, joined or trusted.
A modelled, tested layer that AI and machine learning can build on.

How data projects run

The same three phases as every engagement,
starting with the sources that matter most.

01 / Assessment

Map the disagreement

We map your sources, the reports people rely on, and where the numbers diverge. You get a written scope, a fixed price, a running-cost estimate, and a clear picture of what's causing the disagreement.

About one week

02 / Delivery

Most important sources first

The sources that matter most, landed, modelled and tested, with definitions agreed with the people who use them. More systems are added in stages, each delivering something usable.

First version in 4–8 weeks

03 / Handover

Ready for your team to extend

Every table and metric documented. If your team is comfortable with SQL, they can extend it without calling us. Least-privilege access, and full control handed back at the end.

Support optional

Built on

PostgreSQL
dbt
Airflow
Airbyte
BigQuery
Snowflake
DuckDB
Metabase

How every engagement works Full stack

FAQ

Before you rebuild reporting

Do we need a data warehouse?

We already have dashboards. Why isn't this solved?

Will it break when we change a tool?

What does it cost to run?

One set of numberseveryone works from.

Three systems, three revenue numbers.
One of them is the one everyone uses.

Every tool counts differently

One definition, agreed and in code

Tested on every load

What breaks in reporting,
and what we build to fix it.

Every tool has its own version of the truth.

A modelled warehouse.

The definitions live in someone's head.

Agreed definitions, written once.

Reporting is somebody's Monday.

Pipelines that run on their own.

Nobody finds out when it breaks.

Data quality tests.

The dashboard tool didn't fix it.

Reporting on the modelled layer.

People copy data between tools.

System-to-system sync.

Reports it fixes

Leadership reporting

Finance

Sales and marketing

Operations

AI and ML projects