Harsha Guggilla's list of 19 Fabric mistakes is mostly about engineering. The ones that have cost our clients money were settled, or left unsettled, before the build started. Here are the five, and a checklist to go with them.
Harsha Guggilla, a data solution architect, put a list on LinkedIn this week of 19 mistakes he keeps seeing on Microsoft Fabric projects. It is worth a read. Most of it is about how the platform gets built: notebooks, Spark, Direct Lake, package management, CI/CD. We agreed with nearly all of it.
What struck us, though, is that the mistakes that have cost our clients real money sit further up his list and are not engineering mistakes at all. They are things the business decided, or never got round to deciding, before anyone opened a workspace. Five of them come up again and again.
Starting with the platform instead of the problem
This was his number one and it would be ours too. We still get asked to build Fabric platforms by businesses that, when you push a bit, cannot say which decision it will improve or which number it will make trustworthy. The reason usually given is that Fabric is where Microsoft is going. That is true. It is not a reason to spend six figures this year.
For a lot of mid-market businesses, a properly built Power BI model over the systems they already run gets them most of what they want for a fraction of the cost. Fabric starts to make sense when several systems need to share one data layer, when volumes get large, or when there is data science or AI to run on top. We wrote a short guide to that decision: Do you actually need Fabric?
The simplest test is to ask what a real user will have in front of them in the first 8 to 12 weeks. A landing zone is not an answer to that.
Not knowing whose data it is
A Fabric platform sits downstream of your ERP, your CRM, payroll and whatever else the business runs on. When one of those changes, a supplier upgrade, a new dimension in finance, a renamed field, something in the platform will break, and that is to be expected. The problem comes when nobody on the client side owns the relationship with each source system, so the first anyone hears of a change is a wrong number in a report.
The other half of this is definitions. Revenue, margin, headcount and utilisation usually have two or three meanings inside one company. If those are not agreed and written down before the build, the platform reproduces the disagreement very accurately and the board goes back to the spreadsheet it trusted before. There is more on this in how long a Fabric implementation takes and what to decide first.
Assuming compute and storage are cheap
Fabric is billed on capacity, and everything that runs on it uses capacity: refreshes, pipelines, notebooks, queries, Copilot. We have seen platforms that were comfortable on an F8 at go-live needing an F64 a year later. The business had not grown. Reports, pipelines and a Copilot trial had been added one at a time, and nobody was looking at what each one cost to run.
The build cost gets negotiated hard. The running cost is the one that turns up in year two, so ask for it up front, ask what the assumption rests on, and ask who will be watching the capacity metrics once you are live. Microsoft provides the signals, but somebody still has to read them. For a starting figure on your own estate, the Fabric licensing calculator will give you one.
Leaving security and residency until the end
Once payroll, margin by customer and pipeline by salesperson are all in one place, who can see what stops being an afterthought. Row level security, workspace roles and who is allowed to create what need deciding before the first dataset lands, because bolting them on afterwards usually means rebuilding. In our experience the security work that gets pushed into the last fortnight before launch is the work that gets cut.
Residency is a similar story. Fabric capacity sits in the region you choose, and some features will process data elsewhere unless you tell them not to. For most UK and Irish businesses this is a quick check. It is still a check somebody has to make, so ask for the region, the settings that keep data in it, and who confirmed them.
Building for go-live, not for year two
This is the one that turns most of the engineering items on his list into a business problem. A platform made of notebook chains that only one person understands, with no version control, no tests and no documentation, is not something you can leave the engineers to worry about. You inherit it the day the project team leaves.
The same goes for the refresh that fails at 6am on a Monday. Something will fail at some point. What matters is whether your finance director finds out from a wrong number in the board pack, or whether somebody was already fixing it before anyone opened the report. Monitoring, alerting and a named person at the end of it are part of the platform, and so is an agreed basis for changes and support once you are live. Agreeing that in month four, when you have no other option, is not a strong position to be in. The warning signs are described in the Fabric rollout that quietly goes wrong.
Why this matters for the other 14
The rest of his list, from package management to skipping Git and CI/CD, is what a competent partner does without being asked, and you should not have to check. Those things do slip, though, and they slip most often on projects with no clear outcome, no named data owners, nobody watching the cost and no plan for year two, because nobody on the business side is holding the work to a standard they can see. Get the five above right and the rest is much easier to insist on.
The checklist
We have put this into a two page checklist: 12 questions a business leader can put to their own team, or to a partner they are considering, grouped into why, data, cost, security and living with it, each with a note on why it matters and what to do if the answer comes back vague. You do not need a technical background to use it, but you should have written answers to all 12 before you commit to anything.
Download the checklist: Questions to ask before you commit to Microsoft Fabric
If you would rather we worked through it with you, that is what the free Data Assessment is for: a 60 minute call with a senior consultant, then a written summary within 3 working days, whether or not Fabric turns out to be the answer. Book a free Data Assessment.
Related reading
How long does a Microsoft Fabric implementation take, and what to decide first
The Fabric rollout that quietly goes wrong
Fabric capacity: a plain English guide to not getting a shock on your invoice
How to choose a Microsoft Fabric consultancy
James Morley
Analytics Solutions Consultant
Part of the Hopton Analytics team, delivering governed analytics programmes for UK mid-market organisations.



