Data engineering · Platform and migration
A data platform sized for your business, not for a sales brochure
We choose and build your data platform, designing its structure, environments and running costs from the start. We then move your existing reporting onto it, without changing anyone’s figures along the way.
Where this starts
Chosen for its features, regretted for its cost
Data platforms are usually chosen after a feature comparison and a demonstration. Eighteen months later, they are judged on a bill nobody predicted and a structure nobody designed. Computing resources are left running overnight, every team builds its own copy of the same tables, and experiments are not kept apart from the figures the board reads. The platform is rarely the wrong choice. The problem is that it was simply adopted, rather than properly engineered.
What we usually find
- A platform bill growing faster than the amount of data it holds
- No separation between experimental work and the board’s figures
- Every team keeping its own copy of the same core tables
- An old reporting database that nobody dares to switch off
- A migration planned as a single switchover weekend
- Figures that changed after a move, with nobody able to explain why
Our position
The cheapest data platform is often the one you already have, run properly. We will tell you if that is the case before we recommend another.
What the work covers
Designed first, then moved
The structure comes first. Decisions made in the first month set the running costs and the maintenance effort for years afterwards.
Choosing a platform
We make the decision based on your data volumes, how you query the data, your team’s skills and your budget, rather than a feature list. Staying on your current platform is always one of the options.
A layered structure
Raw data, cleaned data and data modelled for reporting are kept in separate layers. A source can then be reloaded without disturbing the figures built on top of it.
Separate environments
Development work is kept apart from production, so experiments never affect the figures your leadership team is reading that morning.
Cost control built in
Storage and computing power are sized for how the platform is actually used, with automatic shutdown when idle and spending budgets. They are not set up for a peak that never comes.
Access by role
Who can read and change what is set by role from the first day. Protecting sensitive data is covered by data security, and we connect to that work rather than rebuilding it.
Moving your reporting
Existing reports and their logic are moved across in groups. Old and new figures are compared side by side before anything old is retired.
How the move stays trustworthy
The figures match before anything is switched off
A platform migration is judged on one thing: whether the figures people rely on are the same afterwards, or different for a reason someone can explain.
Usage review
Before the move, we measure how much each report and table on the old platform is used. The ones nobody opens are retired rather than carefully moved.
Matching figures
Key figures are calculated on both the old and new platforms and compared automatically. Every difference is explained and approved before the switchover.
Switchover
Reports are moved in groups, each approved by the people who use them, so no single weekend carries all the risk.
Cost alerts
Budgets, alerts and automatic shutdown mean an expensive query or a forgotten job is noticed within a day, rather than when the invoice arrives.
Switching off
A date for switching off the old platform is agreed at the start and kept to. The licence and hosting savings are recorded when it happens.
What you are left with
- A platform chosen to suit your own data volumes and budget
- Separate layers for raw, cleaned and modelled data
- Development and production kept apart
- Checks proving that the figures match
- The old platform switched off on an agreed date
Tell us what your current platform costs and who uses it
Those two figures usually decide whether you need a new platform or simply a better-run one.