Data you can get at
Data engineering services for small and mid-size businesses. We get your numbers out of the systems they are stuck in and into one place you can trust, with a run log that shows where every figure came from.
Our AI assistant asks about your project and can build you a free plan.
The problem
Every project on this site assumes you can get at your numbers, and often you can't. Sales are in one system, costs are in another, and the reconciliation that makes them agree lives in a spreadsheet on one person's laptop, kept up from memory.
You know you're there when a simple question, like how much you made on a customer last quarter, takes two days and gets two different answers.
How we approach it
We start small on purpose. The usual instinct is to build a platform that will answer every future question. That project takes a year and costs a great deal, and it is usually abandoned. We get one set of numbers into one place, correctly, and prove it by answering the question you asked.
In practice we pull from each source on a schedule and write every join down, including the ugly ones where customer names don't match. We check the result against something you already trust, and we log every run, which is what makes it a pipeline you can rely on.
You should be able to point at any number and see where it came from and when it last updated. Without that, the first argument about a figure is enough for people to stop trusting all of it.
Work we don't take on
We scope this work more carefully than anything else we do. We're a good fit for getting a handful of sources into one place you can trust. We're a poor fit for a company-wide data platform with many stakeholders and a migration deadline, and if that's what you need, we'll tell you. Selling you work that isn't right for us would be a disservice to you and a bad month for us.
When two teams have different numbers for the same thing, we go back to the process that produces them: what physically happens when each number gets recorded. There is usually a real-world reason one measurement sits closer to the truth, and once we find it, the argument ends.
A worked example
Three sources: a point-of-sale export, an accounting system, and a spreadsheet of supplier costs someone maintains by hand.
Each one is pulled on a schedule and saved unchanged, so there's always a copy of what arrived. The matching is done where anyone can see it. Supplier names differ from one source to the next, and the rule for matching them is written down, so nobody has to remember it. The result is one dated table with a run log beside it.
None of it is exciting. The weekly margin question now takes a minute and gets one answer, and when someone disputes it, finding where the figure came from takes seconds.
Why there is no package
How much work this is depends almost entirely on your systems. Two sources with good APIs is a straightforward job. One old system that drops a CSV every night, with keys that don't match, can take longer than everything else combined. So we scope it first and then quote a fixed price. We give you a fixed price before we start, usually within two business days.
Start here: Tell us the question that takes two days to answer today.
That tells us more than a list of your systems. From there we can usually tell you within one conversation whether this is a week of work or a bigger project.
How it usually attaches to a package
- Before a Dashboard Sprint When the three sources won't agree. Often it's handled in the first week of the dashboard work, as part of the same project.
- Before a Forecast Starter When the history exists but is scattered or inconsistent.
- After either When something that worked once needs to run reliably every week.
Need something bigger?
For pipelines that keep running, the care plan covers monitoring and fixes. With scheduled jobs that's the part that matters most: a pipeline that stopped without a word three weeks ago is worse than having none.
Every project is handled directly by a member of our team.
If you let us publish what we build together, you get a discount. The work has to be something we can publish, and you approve every word before it goes live. Our team applies the discount when we quote.
Questions people ask
Do we need a data warehouse?
Probably not yet. For most small businesses, a scheduled job that produces one clean, dated table answers the question they have. A warehouse makes sense when several teams need to ask questions nobody can predict.
Our systems do not talk to each other. Is that a problem?
It's normal. What matters is whether we can read each one reliably, through an API or a scheduled export that doesn't break. If one can only be read by a person clicking buttons, we'll tell you before you commit.
Who owns what you build?
You do. It runs in your accounts and on your infrastructure, and we hand over the code and the documentation. We don't hold your pipeline hostage.
What happens when a source changes?
It will. The pipeline is built to stop with a visible error, so a change doesn't turn into wrong numbers nobody notices, and the run log shows what broke. Fixes are included for thirty days, and the care plan covers them after that.
Can you just fix our spreadsheet instead?
Sometimes that's the right answer, and we'll say so. A well-built spreadsheet that one person maintains can be fine. It becomes a problem when the business depends on it and only that person understands it.
Which question takes two days to answer today?
Tell us the decision you're trying to make. If we're not the right people, we'll say so.
Our AI assistant asks about your project and can build you a free plan.