Data
platforms
Data platforms
The single place where data from across the company arrives on its own, gets cleaned and is ready to query with confidence.
A data platform brings three things together. A central store, a data warehouse or a data lake, where the data from every system is kept. The data engineering processes (pipelines, ETL) that collect it every night from the ERP, the CRM, the online shop or the spreadsheets. And the quality and governance rules that make sure a customer is one customer and a sale is counted once.
It is what used to be called Big Data, with one practical difference: in a modern data platform, volume matters less than reliability. A company of 80 people doesn't need to process petabytes. It needs the margin leadership sees to come from the same data finance uses. And it needs an AI assistant to be able to query that data without making anything up.
The three that come up in every first meeting.
No. BI is what you see: dashboards and metrics. The platform is what sits underneath: where the data lives, how it arrives and who guarantees it is correct. Without it, every dashboard queries the ERP in its own way.
Yes, in two cases. If you have more than two systems that don't talk to each other and someone cross-checks their data by hand every week. Or if you are considering AI agents and already know your data needs work.
A first warehouse, with your two or three most important sources loading on their own, in 6-10 weeks. The other sources are added afterwards, one at a time.
Six separate pieces, from data architecture consulting to the data engineering that keeps every load running. You can hire them together or one at a time, depending on what your company actually needs.
ERP, CRM, production, e-commerce, banks and spreadsheets, loaded every night into a central store. We decide what goes into the data warehouse and what into the data lake, sized to your company.
Automatic rules: duplicate customers, amounts with and without VAT mixed together, orders with no date. A dashboard shows what percentage of the data is valid and who should correct it.
Which tables exist, what each field means, who owns it and who can see it. That way a new employee, or an AI agent, finds the right data without having to ask.
The processes that move and transform the data, under monitoring. If a load fails at 03:00, someone knows by 03:05, and nobody opens a report with data from the day before yesterday.
Sales, customers, product, finance and operations, modelled once and with the same keys. Crossing anything with anything stops being a project.
Metrics defined once and exposed in the same way to the dashboard and to the AI agent. This is where the platform becomes the foundation for the rest of the catalogue.
Business IntelligenceFive steps. The first one isn't choosing a technology.
Which systems there are, what data they hold, how they are accessed, who they depend on and what state they are in. This is where the timeline and the budget come from.
We agree with leadership on what the platform really needs to do. Answering the key questions is always cheaper than answering everything ‘just in case’.
With quality measured from day one, so it can be used, criticised and adjusted with the team in the room. Whatever isn't used in the first few weeks never gets used.
Every new system comes in with its owner, its quality rules and its documentation. No exceptions: whatever comes in without an owner ends up as noise.
The code, the model and the documentation are yours. We train whoever will maintain it; if you would rather we did it, that can be agreed. We don't work to make you dependent on us.
The platform stores and organises; AI queries and acts. The platform makes sure the data exists, is clean and means the same thing across the company. An AI agent reads it, reasons over it and carries out a task.
Neither replaces the other. The platform goes underneath, and you notice it with the very first question. ‘How much have we sold to this customer?’ only has a reliable answer if there is a single source where that customer is one customer and their sales are complete. Without it, the most expensive model on the market answers confidently and gets it wrong.
In practice, the AI projects that go well start here. First the warehouse with sales and payments. Then the dashboard that shows them. And finally the agent that warns the sales rep before the unpaid invoice reaches finance.
Your sources
Accounting, banks and invoicing in the same store, with the reconciliation done before the close arrives. The finance team stops chasing extracts.
ERP, CRM and online shop reconciled, so you know who each customer is, how much they buy in total and how much they owe. It often reveals a customer base more concentrated than anyone thought, or customers who are chronically late to pay.
Production, maintenance and quality data collected from machines and plant systems, and set against what they cost. This is where technical data turns into an economic decision.
Inventory, orders and suppliers in a shared model, so requirements are calculated every night without depending on each warehouse manager's judgement.
A warehouse with defined metrics and a documented catalogue, on which an agent can answer questions or carry out tasks without inventing figures.
The ones asked before the budget is approved. If yours is missing, write to us.
Let's talk about your projectAs with any data engineering consulting, it depends on the number of sources and the state they are in, not on the size of the company. A one-hour session looking at your systems is enough to fix a price. Before that, any figure would be a guess.
Almost always in the cloud: you pay for what you use, there is no server to maintain and scaling up takes a click. On-premise only when a legal or contractual requirement means the data cannot leave your premises.
The first useful warehouse, 6-10 weeks. The full platform, 4-8 months depending on the number of sources. Each source that comes in is useful straight away: you don't have to wait until the end.
It is the first question in any data warehouse consulting project. A warehouse if your data is tables (ERP, CRM, accounting) and the question is a business one. A lake if you store documents, images or machine data, or you want to feed AI models. Many mid-sized companies end up with both.
Yes, that is the usual way. The ERP isn't touched: the platform reads from it. Spreadsheets are loaded as they are and replaced by their real source as they get documented. Nobody stops working.
It is what makes AI work. An agent that queries company data needs a single, clean, documented source; otherwise it answers with figures nobody can defend. That is why AI agent projects usually start here.
In healthcare it always comes first, and it is the only industry where that happens. Until it is written down which data leaves, where it is processed and who can see what, the project doesn't get past the approval meeting. It comes third in manufacturing and food, to pull up the traceability of a batch instantly instead of rebuilding it by hand.
