AnswersReporting and delivery
Why business reports have looked the same for ten years
Rob CollieFounder and CEO, P3 Adaptive
Because the tools only knew how to make one kind of report. Every BI canvas has to work for every company in every industry, which means it ships the overlap – the handful of visuals and options useful to more or less everybody. That is the correct engineering decision, and it is also a ceiling. Your reports look like everybody else’s because there was never a way for them to look like anything else.
Row of number cards along the top. Bar chart, line chart, a table. Slicers down the side. You have seen that page ten thousand times, in companies that have nothing to do with each other.
I know exactly why, because I was in the room when we picked the parts.
I’ve been working on this problem for about sixteen years
In 2010 I was at Microsoft running the team that owned BI in Excel, and then I designed most of what became Power Pivot. The pitch we made – the one I believed, the one I left Microsoft over – was that we were going to take report building away from the priesthood and hand it to the people who actually had the questions.
Self-service BI. You’ve heard it. I said it a lot.
It half worked, and I want to be straight about which half. What we delivered was self-service consumption. Slicers, filters, drill-through, bookmarks. That was real and genuinely useful – a business person can interrogate a report somebody else built, and before us they couldn’t.
What we never delivered was self-service authoring. To make something new you still had to install a desktop app, understand data modeling, learn a formula language and pick up a visualization grammar. Four things to learn before producing anything at all, which is three more than a person with an actual job is ever going to get through. I have watched thousands of smart, motivated people bounce off step one and quietly conclude they weren’t technical enough.
They were fine. It wasn’t a failure of will and it wasn’t a design mistake either. Authoring was genuinely expensive, so restricting it to specialists and handing everyone else knobs was the right call. Both halves of what you see today – who gets to build, and what they can build with – come from that one assumption about cost.
The assumption stopped being true about eighteen months ago.
Am I trashing Power BI here?
No. I’d be trashing my own résumé.
The semantic model in Power BI is still the best in the business, and I’m not being polite about a former employer. It’s why this company has been built on the platform since 2013, and it’s why everything I’m about to describe sits on top of it rather than replacing it. If your models are in decent shape, you’re further ahead than you think.
The report canvas is a separate conversation. It is a very good answer to a question that stopped being asked. That’s not a knock on the people who built it – I was one of them – it’s just what happens when the ground moves faster than a product can be redesigned.
Two things changed, and the second one caught me off guard
Describing a report is now a way to build one
Those four things you had to learn first are simply off the path. Somebody who can say what they want to see, in the meeting where the question came up, can have the page.
The parts bin went away
I expected the first one. I did not expect this. A report no longer has to be assembled from a general-purpose kit, which means it is no longer capped at what that kit could express – and that ceiling turns out to have been holding down a lot more than anybody realized.
What a report can be when it isn’t the overlap
A page built for one question doesn’t have to serve every company on earth. It can be everything this question needs and nothing else. Three things open up at once, and the third one is the sleeper.
It looks like somebody designed it
Real typography. Real hierarchy. A layout built for this content instead of a grid of rectangles that will accept any content. You can put four times the information on the page and have it be easier to read, because density on purpose is legible and density by accident isn’t. And it looks like it came from your company, not from the same tool everybody else bought.
That sounds cosmetic. It isn’t, and I’ll say the quiet part out loud. When somebody circulates a page that looks considered, they get treated differently than the person who circulates the default dashboard. Everyone in every company knows this. Nobody in my industry will say it. The first time one of your people sends around something that looks like it came out of a design studio – and it came out of them– that is a promotion conversation waiting to happen.
It can show things no canvas has a button for
Not twenty-odd chart types with whatever options a vendor decided to expose. Anything you can describe. Small multiples across forty branches on one screen. A Sankey of how orders actually flow. A slope chart, a calendar heatmap, a bump chart, a waterfall that genuinely reconciles.
Better yet, the visuals that belong to your business, which no vendor is ever going to ship because the entire market for them is your industry. A production schedule drawn against actuals. A route map with the exceptions flagged. An aging pyramid. A plant layout with live readings sitting on it. Plus annotation where annotation belongs – the note that says this is the week the line went down, right there on the chart instead of in a comment thread nobody opens.
It behaves the way the question behaves
Interaction stops meaning slicers. A scenario control that recalculates instead of filtering. A prioritized work list you sort, act on and clear. A drill that reshapes the whole page rather than swapping out one visual. A comparison you build by clicking the two things you want compared.
And the page can explain itself. The finding written out in plain language next to the evidence, so your reader walks away with the conclusion instead of a chart they have to interpret. A summary up top that opens into the detail underneath it – one artifact, instead of four report pages and a bookmark somebody has to be trained on.
When you build a canvas that has to serve every company on earth, you ship the overlap. A report built for one question can be everything that question needs and nothing else.
Then all the questions nobody ever asked start showing up
Every company carries an invisible backlog far bigger than the visible one – the weekly view a regional manager would genuinely use and has never once asked for, because everybody learns fast what’s worth requesting around here and what isn’t. Drop the cost of asking, raise the ceiling on what an answer can be, and all of it arrives at the door at the same time.
Which brings a consequence I want to be honest about, because it cuts against the usual sales pitch. This makes your semantic models more important, not less – and more work, not less. A person opening a report patches the model’s gaps by instinct without noticing they’re doing it. They know which measure is the real one. They know to skip the two test calculations somebody left in there in 2019. Now put ten times the questions through that model, asked by people who don’t have the instinct, by way of an assistant that reads exactly what’s written and nothing else.
Your ticket queue shrinks. Your modeling work grows, and it becomes the most leveraged thing in the whole estate.
The part nobody warns you about
This is usually where the story stops, and it shouldn’t, because the genuinely hard problem starts right here.
Somebody with no BI training just built something good against real company numbers. Now they want a colleague to see it. Today there are three ways to do that and every one of them is bad.
Send the file
Stale the second it was made, and your company’s numbers are now loose in an attachment carrying whatever security the sender happened to feel like applying.
Host it as an app
Now an end user is running an application that talks to company data – on their behalf, or worse, on behalf of whoever opens it. Every one is a fresh security review, which is exactly why nobody lets people publish them.
Don’t share it
The winner by a mile, and the reason all of this work stays invisible. It sits on one laptop and the company learns nothing from it.
Option three is what actually happens. It’s why “anyone can build a report now” hasn’t changed how most companies operate yet. Open up authoring without solving distribution and all you’ve done is move the bottleneck onto somebody’s hard drive.
What has to be true for there to be an option four
Sharing is hard for a real reason. A report that reaches live data is a program running against your numbers on somebody’s behalf, and the entire question is whose.
So a published report has to re-run its queries when it’s opened, under the identity of the person opening it, with their row-level security applied before a single row comes back. If they can’t see it in the source, they can’t see it in the report, and a forwarded link grants nobody anything. That one property is the difference between something you can let people publish and something that needs a review every single time.
The second requirement is the one people miss, and it’s a difference in kind rather than a setting. The thing a published report runs inside has to be a sandbox built for reports – not a general application platform that somebody has gone around locking down.
What that buys you is everything it takes out of the author’s hands. It has no way to reach the network, so nothing can be sent anywhere. It has no way to get data except through queries that already ran under the reader’s own identity, so there is no back door around permissions. It can’t hold a credential, because there is nowhere to put one. Your author gets to build the most capable thing of their career, and there is no version of it that ships your numbers somewhere they shouldn’t go – not because they were careful, but because the capability was never in the room.
That is the entire difference between a platform where everything published is a fresh security review, and one where “anybody here can publish a report” is a sentence you can say out loud in front of your security team.
And your governance people come out ahead, which surprises them. You’re opening up who can build, not who can see. In exchange they get a record of every report, who owns it, who it reaches, every query and who ran it. Including the thing no BI administrator has ever been able to answer: which semantic models people actually use, and which ones you’ve been paying to refresh while nobody opened them.
And yes, you still want pros building some of them
Let me deal with this quickly, because it does not need a whole essay. The handful of reports your entire company runs on should still be built and tested by people who do this for a living, and a number that lands in a board deck should have had somebody’s name on it. None of that changes.
What changes is everything else. And everything else is most of the queue.
Four things you can check this afternoon
Diagnostic
How much is the ceiling costing you?
Each one tells you something different, and none of them needs a tool.
Open the last five reports your team shipped. How many are the same layout with different fields in it?
That’s not a lack of imagination on their part. It’s the parts bin. Whatever they built had to come out of it.
Ask a manager to describe what they’d most want to look at. Not a chart type – the actual thing, in their own words.
Then ask your BI lead whether the tool can draw it. The gap between those two answers is the ceiling, measured in the only units that matter.
Ask three managers what they’d look at every week if it existed. Count how many exist.
The ones that don’t were never requested, because everybody learned what’s worth asking for. That’s your invisible backlog, and it’s almost always bigger than the one you can see.
Open the last month of email from your leadership team and count the attached spreadsheets.
Every one is a report somebody needed, couldn’t get, and rebuilt by hand – now circulating completely outside your security model. It’s the most honest measure of unmet demand in any company I know of.
Where I’d start
The instinct after reading something like this is to go plan a program. Select a platform, run a readiness assessment, stand up a governance council, sequence a roadmap. That’s two years, and at the end of it the questions have changed on you.
Take one model your people already argue about. Put it in front of ten or twenty of them who have real questions and no way to answer them – not a pilot committee, and not your power users, who have already bent themselves around the limitations. Then watch what they build. What they reach for is the only trustworthy evidence about which pipes are worth laying, and in my experience it contradicts the roadmap somebody would have written in a conference room.
Ten business days is enough to know whether any of this is real for you.
I’d rather show you than keep typing.
The sandbox has reports built this way on real data – pages we could not have produced on any canvas two years ago, including mine. Open one, change it, ask it something else. No sign-up, no form, no call with anybody.