"How much will it cost?" is the first question we get about a data platform, and the honest answer is that for a company of 20 to 500 people the cloud bill is rarely the expensive part. This article explains where the money actually goes, why Google Cloud is a good fit at this size, and the handful of habits that turn a cheap platform into an expensive one.

Three cost drivers, in the wrong order of importance

Owners usually picture the bill as "servers". On a modern cloud data platform there are no servers you keep running. There are three things you pay for.

Storage. How much data you keep. For an SME this is measured in gigabytes, not terabytes: a few years of invoices, stock movements, orders and CRM activity, from systems that produce a row per transaction rather than per click. Cloud storage for that volume is cheap enough that it rounds down. It is almost never the driver.

Queries. How much data you read, and how often. This is where design matters. A daily job that reads yesterday's rows costs almost nothing. A dashboard that re-reads three years of history every time someone refreshes it costs something every time, and if it refreshes every five minutes for forty people, it costs something all day.

People. The time to build the pipelines, define the numbers, keep the loads running and answer "why does this not match the ERP". This is the real cost of a data platform at any size, and it is the part that no pricing page shows. A platform that takes two days a month to keep alive costs more than one whose cloud bill is ten times higher but runs itself.

Rule of thumb for an SME: storage is noise, queries are what you design for, and people are what you budget for.

Why BigQuery's pricing suits small volumes

BigQuery, Google's data warehouse, is what we use for most SME platforms, and the reason is its pricing model rather than any feature.

You do not pay for an idle machine. Traditional warehouses, and a good number of "modern" ones, bill for a cluster that exists whether or not anyone is querying it. BigQuery has no cluster. Nothing runs between queries, so a platform that is used for twenty minutes a day costs twenty minutes a day. For a company whose data team is one part-time person, this is the single biggest difference.

You pay per data read, with a free monthly allowance. On-demand pricing charges for the amount of data a query scans, and the first slice each month is free. At SME volumes, a well-designed set of daily loads and a handful of agents reading recent data sits inside or near that allowance. The bill becomes a rounding error and stays there until the business changes, not until the calendar does.

Storage has a free tier too, and old data gets cheaper on its own. Tables that have not been modified for a while drop to a lower storage rate without anyone doing anything.

Everything around it is pay-per-use as well. Scheduled jobs, the small serverless services that run the pipelines and the agents, the notification channels: none of them need a machine kept on. A complete platform can be made of components that cost nothing at rest.

We can say this with our own numbers, because the systems we run for clients and our own products live on exactly this stack. One of our platforms serves its users in two languages on Google Cloud and runs for €0-2 a month in cloud cost. That is not a trick of the free tier; it is what a correctly designed small workload looks like, and it is the rule across our projects, not the exception.

What makes a small platform expensive

The same pricing model that rewards good design punishes a few habits. These are the ones we see most often.

An always-on virtual machine "for the ETL". Someone sets up a VM to run the loads at night. It runs all day. It is the most common line in an SME cloud bill, and it is rarely needed. Scheduled serverless jobs do the same work and cost nothing between runs.

Dashboards that scan full history on every refresh. A BI tool pointed at a raw table with auto-refresh on is a query bill with a timer. The fix is to pre-aggregate: a daily summary table of a few thousand rows that the dashboard reads instead of the millions behind it.

Unpartitioned tables. BigQuery can limit a query to the days it needs, but only if the table is organised by date. Loading everything into one unpartitioned table means every "yesterday" query reads every year. It is a design decision made in week one and paid for monthly.

Copies of the same data. A staging copy, a "clean" copy, a copy for the dashboard, a copy someone exported to a sheet. Each costs little; the real cost is that nobody knows which one is right.

Licences that cost more than the platform. Per-seat BI subscriptions, per-connector integration tools, per-user "AI add-ons". It is common for the tools around a small platform to cost several times what the platform itself does. Count them in the budget, and ask whether a scheduled query and a message to Slack would do the same job.

Nobody watching the bill. Cloud providers let you set a budget and an alert in five minutes. Every platform we build has one from day one, with a threshold low enough that a mistake is noticed in days, not at the end of the quarter.

What a sensible budget looks like

For a company of this size, a reasonable way to think about it is in three lines: a cloud bill that is small and predictable, with a budget alert on it; a build cost that is mostly design and pipelines, concentrated in the first weeks; and a running cost that is mostly people, which should fall over time as the platform runs itself and the definitions settle. If the proposal in front of you has the first line large and the third line missing, ask why.

If you want a plain estimate for your own case, send us a note describing your source systems and how many people would use the platform. We will tell you what we think it should cost, and if the answer is "you do not need this yet", we will say that too.