The chosen tier was wrong on first inspection more often than not, because seats were bought where a shared capacity node ran 25 to 40 percent cheaper
Capacity ships compute and the per user tiers ship seats. The right answer depends entirely on how many people consume against how many actually author.
Prepared by Redress Compliance · August 19, 2026 · Analytics tier reviews. 40 to 55 reviews run, 2024 to 2025.
Executive summary
Enterprises over bought the per user premium tier where a shared capacity node ran 25 to 40 percent cheaper at their viewer count. The break even is a distribution question, not a feature one.
Author populations were over counted: the true author seat count ran 30 to 50 percent below the licensed count. Everyone who opens a report is not an author.
Capacity nodes sat under 35 percent utilized, so moving to a smaller node released budget without losing a single feature.
Capacity ships compute; the per user tiers ship seats. Choosing between them on feature lists rather than on consumer distribution is the common error.
What are the three commercial tiers?
Two license individual seats and one licenses a defined compute node any user can consume. They carry different features, different costs and completely different distribution patterns.
| Tier | List rate | Distribution model | Best fit |
|---|---|---|---|
| Pro | 14 USD per user per month | Per user, all users licensed | Small to medium teams, standard reports |
| Premium per User | 24 USD per user per month | Per user, all users licensed | Mid sized teams needing advanced features |
| Premium Capacity P1 | 4,995 USD per node per month | Capacity, consumers use a free viewer | Large enterprise, broad distribution |
| Premium Capacity P3 | 19,978 USD per node per month | Capacity, consumers use a free viewer | Very large enterprise, complex workloads |
The entry tier licenses everybody who touches the content
It covers report authoring and publishing, workspaces and apps, standard refresh at up to 8 a day, and models up to 1 GB. It bundles inside the top enterprise suite and sells standalone at 14 USD. The rates are published on the pricing page.
What the entry tier does not reach
The 1 GB model cap, no paginated reports, no advanced machine learning features, no deployment pipelines and no third party authoring endpoint. Each of those is a real reason to move up, and none of them is a headcount reason.
When does the per user premium tier fit?
When you need the premium features but do not have enough users to justify a capacity node. It adds those features to the per user model at 24 USD.
What the extra ten dollars buys
- Paginated reports for pixel perfect operational reporting.
- Models up to 100 GB, against the 1 GB cap below it.
- Up to 48 refreshes a day rather than 8.
- Deployment pipelines and the third party authoring endpoint.
The trap is the viewer population
Every viewer needs a licensed seat on this tier, which is exactly the cost curve capacity is designed to break. The published mechanics sit in the per user premium reference.
The analytics platform negotiation guide
How the capacity nodes price, where the break even sits, and the buyer side moves before the tier is chosen.
Get the brief →What 40 to 55 tier reviews showed
Across roughly 40 to 55 Power BI tier reviews Morten Andersen ran in 2024 and 2025, the chosen tier was wrong on first inspection more often than not. Three patterns recur.
- Enterprises over bought the per user premium tier where a shared capacity node ran 25 to 40 percent cheaper at their viewer count.
- Author populations were over counted, with the true author seat count running 30 to 50 percent below the licensed count.
- Capacity nodes sat under 35 percent utilized, so moving to a smaller node released budget without losing a feature.
Capacity ships compute. The per user tiers ship seats. The choice depends on the consumer distribution, and almost nobody measures that before buying.
- Entitled vs deployed vs active seats compared, priced at your actual contract terms
- Usage exports analyzed: inactive accounts, tier right sizing, per user reassignment
- Your renewal quote benchmarked against real closed transactions
How does the capacity tier change the math?
By licensing a compute node instead of people. Any user inside the tenant can consume content hosted on it, and consumers need no paid seat at all.
| Node | Compute | Memory | List per month |
|---|---|---|---|
| P1 | 8 virtual cores | 25 GB | 4,995 USD |
| P2 | 16 virtual cores | 50 GB | 9,995 USD |
| P3 | 32 virtual cores | 100 GB | 19,978 USD |
| P4 | 64 virtual cores | 200 GB | 39,938 USD |
| P5 | 128 virtual cores | 400 GB | 79,891 USD |
The free viewer is the whole economic argument
Consumers do not need a paid license to view content on a capacity, which is why the node beats the per user tiers past a certain viewer count. The capacity model is documented at the capacity reference.
Under 35 percent utilized is a smaller node
Nodes sitting below that ceiling can move down a size without losing a feature, and the newer platform pricing on the unified platform pricing gives finer grained sizes to move into.
Watch the briefing · 4:44The Tier DecisionWhere the tier choice actually gets made, and what the bundle argument hides when it is made for you.
Why is the author count always too high?
Because everyone who opens a report gets counted as somebody who might build one. The true author seat count ran 30 to 50 percent below the licensed count.
Three populations, one licence type
- Authors, who create and publish content and genuinely need a seat.
- Collaborators, who edit occasionally and are often reassignable.
- Viewers, who consume only and cost nothing on a capacity.
Pull the publish activity, not the login activity
Logins tell you who opened the service. Publish and edit events tell you who authored, and the gap between those two numbers is the 30 to 50 percent. The reclamation method sits in the seat optimizer.
Where the common advice on the tiers is wrong
The common advice is to choose the tier from the feature comparison table. We disagree.
Features decide whether you move up, distribution decides how
The feature gap tells you the entry tier is insufficient. It says nothing about whether the answer is a per user premium seat or a capacity node, and that second question is where 25 to 40 percent sits.
The buyer side move is to count real authors from publish activity, count viewers separately, price both models against those two numbers, and check node utilization before renewing a capacity. The timing lever sits in the fiscal year timing guide and the wider agreement in the agreement series.
What the reviews measured, 2024 to 2025
Two cuts of the review file, and the second is usually what makes the first true.
At the same viewer count, where the estate had bought per user premium seats for a population that only consumes.
Measured from publish and edit activity rather than from logins, which count everybody who opened the service.
Neither figure is a discount. Both come from counting two populations that had been treated as one.
Your first five moves
- Count real authors from publish and edit activity, not from logins, because the true author seat count ran 30 to 50 percent below the licensed one.
- Count viewers separately, since they cost nothing on a capacity node and a full seat each on the per user tiers.
- Price both models against those two numbers, where shared capacity ran 25 to 40 percent cheaper at the same viewer count.
- Check node utilization before renewing a capacity, because nodes under 35 percent can move down a size without losing a feature.
- Let the feature gap decide whether you move up, and distribution decide how. The Microsoft practice runs the two counts before the tier conversation opens.
Frequently asked questions
What are the three tiers?
Two license individual seats, at 14 and 24 USD per user per month, and one licenses a defined compute node that any user in the tenant can consume.
Which is usually chosen wrongly?
The per user premium tier, bought where a shared capacity node ran 25 to 40 percent cheaper at the same viewer count.
Why does capacity win past a point?
Because consumers need no paid licence to view content hosted on it. The free viewer is the entire economic argument for the node.
How over counted are authors?
The true author seat count ran 30 to 50 percent below the licensed count, because everyone who opens a report gets counted as somebody who might build one.
How should authors be counted?
From publish and edit activity rather than from logins. Logins tell you who opened the service; publish events tell you who authored.
What does the entry tier not reach?
A 1 GB model cap, no paginated reports, no advanced machine learning features, no deployment pipelines and no third party authoring endpoint.
What does the per user premium tier add?
Paginated reports, models up to 100 GB, up to 48 refreshes a day, deployment pipelines and the third party authoring endpoint, at 24 USD per user.
How utilized are capacity nodes?
Under 35 percent in the reviewed estates, which means moving down a node size released budget without losing a single feature.
Should the feature table decide the tier?
Only whether you move up. It says nothing about whether the answer is a per user seat or a capacity node, and that is where the 25 to 40 percent sits.
What is the simplest test?
Count authors and viewers as two separate populations, then price both models against those numbers. Almost nobody measures the distribution before buying.
