The example scenario
A machine builder, 250 employees. Data lives in an ERP, a CRM, and a sprawling Excel landscape in the service department. The goal: an automated sales dashboard with an early warning for at-risk customers — the classic first case, because that’s where the pain is most visible: until now, controlling spends two days a month assembling the report by hand.
Build costs
Five line items, realistically priced for delivery by an experienced freelancer or a small consultancy:
Concept and as-is analysis (reviewing systems, sharpening the use case, designing the data model): 3,000 to 6,000 euros. Connecting the source systems with a standard loading tool: 2,000 to 5,000 euros — the range depends almost entirely on whether your ERP has a decent interface. Data model and transformations in dbt including quality tests: 5,000 to 10,000 euros; this is the heart of the build and the point where master-data chaos takes its revenge — duplicate customers and inconsistent article numbers can double this item. Dashboard build in Power BI: 2,000 to 4,000 euros. Documentation and handover: 1,500 to 3,000 euros — and whoever strikes this item from the proposal is buying themselves a dependency.
In total: roughly 15,000 to 30,000 euros for a cleanly delivered first use case over two to three months. Anyone quoting far below that is usually leaving out documentation or testing. Anyone far above it should be able to explain why.
Running costs
Database or warehouse hosting: 50 to 300 euros a month depending on class and volume. Power BI licences: around 10 to 20 euros per analysing user. dbt Cloud, if desired: from about 100 euros. Plus maintenance — realistically two to four hours a week, in-house or as a small service agreement. All in, most setups of this size land at 200 to 500 euros a month plus that support effort.
The costs no proposal mentions
Now for the part you rarely read elsewhere: at least half of the real effort isn’t technical. Your business departments have to give up time — because no outsider can decide which of three customer numbers is the right one, or why revenue is defined differently in the ERP than in the sales spreadsheet. Budget several person-days of internal time across the project, and plan them like money. Projects that fail on this fail quietly: the provider delivers, but nobody answered the substantive questions, and in the end nobody trusts the numbers — again.
The second hidden item is upkeep after go-live. Data is a living system; source systems change, new fields appear. Without named ownership, every pipeline rots — not immediately, but reliably.
What stands against it
Back to the example: two days of manual month-end closing add up to roughly 190 working hours a year — at full cost, quickly a mid-four to five-figure sum, just for copying and pasting. Then comes the harder-to-quantify part, which usually weighs more: a single agreed set of numbers instead of departmental truths; errors that never happen in the first place (Raymond Panko’s research puts the share of complex spreadsheets containing errors at around 88 percent); and decisions based on figures rather than the loudest gut feeling. For a sense of the stakes: Gartner estimates the average cost of poor data quality at 12.9 million dollars per organisation per year, and research by Thomas Redman suggests 15 to 25 percent of revenue is lost to bad data. These come from large, mostly enterprise samples — not a forecast for your business — but the mechanism doesn’t stop at any company size.
The three cost drivers you can avoid
First: buying too big — a data platform where a database would have done easily costs five times as much to run. Second: your provider’s home-grown tooling instead of standards — the maintenance lock-in arrives with a delay, but it arrives. Third: wanting everything at once. One use case, brought properly to the finish line, beats three half-finished ones — including in its effect on your own team.
What it would be in your case is what the on-site Data Readiness Check clarifies: two to three days, fixed price, ending in a prioritised recommendation with an effort-benefit assessment. The first indication comes from the online version — in five minutes.