Power BI consulting is priced three defensible ways: time and materials, fixed price per deliverable, or a managed retainer. EPC Group scopes every engagement across five delivery layers — capacity, data, semantic model, reporting, and governance — because proposals that price only reports are the proposals that overrun.
EPC Group is a Houston-based Microsoft consulting firm operating since 1997, with six Microsoft Solutions Partner designations and 216+ M&A tenant migrations covering 1.83M users.
Last updated: 2026-07-31
Key facts
- Power BI Pro lists at $14.00 per user per month and Power BI Premium Per User at $24.00 per user per month, paid yearly, as published at time of writing (July 2026) on Microsoft's Power BI pricing page.
- Power BI Premium per-capacity SKUs (P1–P5) are retiring at the end of each customer's current agreement term. No new purchases, add-ons, or renewals. Fabric F SKUs, billed through Azure, are the replacement.
- F64 is the licensing cliff. On F64 and larger, users with a Fabric Free license and the Viewer role can view Power BI content. On F2 through F32, every single viewer needs Pro or PPU.
- Fabric capacity uses regional pricing and pay-as-you-go billing (per second, one-minute minimum), with reservations that Microsoft's Azure pricing page states can save approximately 41% versus pay-as-you-go.
- ZipRecruiter published a US average of $54.60 per hour for a Power BI consultant as of 27 July 2026, with a full range of $29.09 to $72.60. That is a practitioner wage, not a firm's billing rate — the gap between the two is the thing most buyers never interrogate.
- Capacity throttling is staged, not binary. Fabric allows 10 minutes of future capacity consumption as overage protection, then applies a 20-second interactive delay, then rejects interactive jobs, then rejects everything.
Quick facts
| Question | Answer |
|---|---|
| What is being bought | Design and delivery across capacity, data pipeline, semantic model, reports, and governance — not "dashboards" |
| Who signs | Usually a CFO, CIO, or VP Data; the technical evaluator is rarely the budget owner |
| Dominant pricing models | Time and materials; fixed price per deliverable; managed retainer |
| Largest recurring cost | Licensing (per-user and capacity), not professional services |
| Licensing decision point | Pro → PPU → Fabric F SKU; F64 is where free viewers become possible |
| Most common overrun cause | Data readiness discovered after the reporting phase has been priced |
| Hard prerequisite for Copilot | A paid Fabric capacity F2 or higher, plus tenant-level Copilot settings enabled |
| What to demand in writing | Named delivery team, semantic model ownership, exit and handover terms |
What "Power BI consulting services" actually covers
The phrase is imprecise, and vendors exploit the imprecision. Four genuinely different offers hide behind it.
Report production. Visuals built against data someone else already modeled. Cheapest work, least durable, and what most agency pages are quietly selling.
Semantic model engineering. The tabular model — relationships, DAX measures, row-level security, refresh strategy, storage mode. Performance and correctness are decided here, and this layer survives when the reports get rebuilt.
Platform and capacity engineering. Gateways, capacity sizing, workspace topology, deployment pipelines, source control. Cheap to ignore and expensive to retrofit. See the Power BI gateway guide for where on-premises sources complicate this.
Program-level BI advisory. Licensing strategy, governance operating model, adoption, and whether Power BI is the right destination at all — Power BI vs MicroStrategy and whether Power BI replaces Microsoft Access are legitimate questions to settle before signing.
A proposal that does not tell you which of these four it is priced for is not a proposal. It is a rate card with a cover page.
The Five-Layer Power BI Delivery Stack
This is EPC Group's scoping model. Every Power BI engagement decomposes into five layers. Price all five or accept that the unpriced ones become change orders.
| Layer | What it contains | Who typically owns it | What happens when it is left unpriced |
|---|---|---|---|
| L1 — Capacity and licensing | SKU selection, F SKU sizing, reservation vs pay-as-you-go, Pro/PPU allocation, region | Client IT + finance, advised by the partner | Deployment lands on a capacity too small to refresh; throttling appears in week six and reads as "Power BI is slow" |
| L2 — Data and refresh | Source connectivity, gateways, pipelines, incremental refresh, data quality remediation | Client data engineering, or the partner | The single largest overrun driver. Data readiness is assumed, then discovered |
| L3 — Semantic model | Tabular design, DAX, RLS, storage mode, certification readiness | Partner, handed over to client | Reports get rebuilt every 18 months because the model underneath them was never engineered |
| L4 — Reporting and distribution | Report design, apps, subscriptions, embedding, mobile | Partner, then client self-service | Visible, demoable, and the only layer most proposals actually price |
| L5 — Governance and adoption | Workspace topology, endorsement, sensitivity labels, tenant settings, training, support model | Client, enabled by the partner | Sprawl. Six versions of revenue. The reason the next consulting engagement gets funded |
Use the stack as a red-line test. Ask the vendor to map every line of their proposal to L1–L5 and to state explicitly which layers they are not pricing. A firm that will not do that on a call is telling you something.
What a Power BI engagement costs
Nobody publishes a credible average, and there is a reason. The variance driver is L2 — data readiness — and no vendor can price it before seeing the sources. Any page quoting "a typical Power BI project costs $X" is either quoting a report-production job or making it up.
What can be published are the inputs. Here they are with sources.
| Cost input | Published figure | Source |
|---|---|---|
| Power BI Pro | $14.00 per user / month, paid yearly | Microsoft Power BI pricing page, as published at time of writing (July 2026) |
| Power BI Premium Per User (PPU) | $24.00 per user / month, paid yearly | Microsoft Power BI pricing page, as published at time of writing (July 2026) |
| Fabric F SKU capacity | Regional pricing; pay-as-you-go billed per second with a one-minute minimum | Microsoft Learn — Fabric licenses / P-to-F migration FAQ |
| Fabric capacity reservation | "~41%" saving versus pay-as-you-go on a 1- or 3-year commitment | Azure Microsoft Fabric pricing page, as published at time of writing (July 2026) |
| Practitioner labor benchmark | $54.60/hour US average; $29.09–$72.60 range | ZipRecruiter, Power BI Consultant hourly pay, 27 July 2026 |
| Free viewer threshold | F64 or larger; below that every viewer needs Pro or PPU | Microsoft Learn — Fabric licenses |
Two things follow from that table that vendors rarely volunteer.
First, the licensing decision usually dominates the services decision. For an organization with a large read-only audience, the arithmetic between "F32 plus a Pro license for every viewer" and "F64 with free viewers" is often worth more than the entire consulting fee. Model it before you scope the work, not after. Our Power BI cost and licensing guide and the Premium vs Premium Per User comparison carry the full SKU-by-SKU treatment.
Second, the practitioner wage is a floor, not a rate. A published $54.60/hour US average is what an individual earns. A firm's rate additionally carries recruiting, bench time, delivery management, quality review, insurance, and margin. That gap is legitimate. Refusing to explain it is not. Ask any vendor to state their blended rate and what percentage of the engagement the named people in the proposal will personally deliver.
The cost worksheet
Build the number yourself. Five lines:
- L1: annual capacity cost (F SKU size × regional rate, minus reservation discount) + (Pro or PPU seats × $14 or $24 × 12).
- L2: estimated data remediation hours × blended rate. If the vendor will not do a paid discovery to size this, that is the finding.
- L3: semantic model hours × blended rate, plus a named handover deliverable.
- L4: report hours × blended rate.
- L5: governance and enablement — usually the smallest line and the one with the highest return.
Then add the line nobody writes down: the internal hours your own people will spend. A Power BI program consumes client-side time in data ownership, business-rule adjudication, and UAT. Budget it or it will be borrowed from somewhere else.
Engagement models: which one, and when
| Model | Choose when | Do not choose when | Who carries scope risk | Change-control trigger |
|---|---|---|---|---|
| Time and materials | Data quality is unknown; discovery is genuine; scope will move | Procurement requires a fixed number; you have no capacity to govern burn | Buyer | Weekly burn report vs plan |
| Fixed price per deliverable | Scope is genuinely knowable — a defined set of reports on a modeled source | L2 is unassessed. Fixed price on unknown data is a change-order machine | Vendor, until the first ambiguity | Written change order per deviation |
| Managed retainer / BI operations | Steady-state run: refresh monitoring, capacity tuning, model maintenance, new-report intake | You have no built platform yet. Retainers do not build, they sustain | Shared, defined by SLA | Monthly service review |
| Staff augmentation | You have a working delivery method and need throughput | You need a method. Augmented staff inherit your problems, they do not fix them | Buyer, entirely | None — you own it |
| Outcome-based | A measurable, agreed business metric exists and both sides can attribute movement to it | Attribution is contested. Most BI outcomes are contested | Vendor, nominally | Metric review board |
The honest default for a first enterprise engagement is a paid, fixed-price discovery that produces a data-readiness assessment and a capacity model, followed by fixed price for L3 and L4 and time and materials for L2. That structure puts fixed price where scope is knowable and T&M where it is not. A vendor who insists on fixed price across all five layers before discovery is pricing the risk into a number you cannot audit.
The 10-question vendor qualification call
Run this on the first call. Ten questions, twenty minutes. The weak answers are not hypothetical — they are the answers you will actually get.
| # | Ask this | Weak answer | Strong answer |
|---|---|---|---|
| 1 | Which of the five delivery layers are you pricing, and which are you excluding? | "We do end-to-end" | Names L1–L5, states exclusions in writing |
| 2 | What capacity SKU are you assuming, and why that one? | "We'll figure that out later" | States SKU, viewer count, and the F64 free-viewer arithmetic |
| 3 | Who specifically will be on the delivery team, and what percentage of hours will they personally deliver? | "Our team of certified experts" | Names, LinkedIn profiles, a percentage |
| 4 | How do you assess data readiness before you quote L4? | "During the project" | A defined paid discovery with a written deliverable |
| 5 | Who owns the semantic model at the end, and what is the handover artifact? | "You'll have full access" | Source-controlled model, documentation, a handover session, a named owner |
| 6 | What is your position on certifying semantic models, and who certifies them here? | Blank look | Explains Fabric endorsement, promoted vs certified, and who the authorized reviewers should be |
| 7 | What will you do when capacity throttling appears? | "We'll scale up" | Explains smoothing, the 10-minute overage window, surge protection, and optimization before scaling |
| 8 | What have you had go wrong on a Power BI program, and what did you change? | "Nothing comes to mind" | A specific failure and a specific process change |
| 9 | What does your exit look like if we terminate at 90 days? | Contractual deflection | A defined transition package and a stated notice period |
| 10 | What are you going to tell us not to do? | "Whatever you need" | Says no to something in the first call |
Question 10 is the highest-signal question in the list. A firm that cannot say no in a sales call will not say no during delivery either. Question 8 is second — see delivery partner and vendor risk for why failure-honesty predicts outcomes better than logo slides.
Red flags
| Red flag | What it usually means | What to ask next |
|---|---|---|
| No pricing structure published or explained | The rate is set per prospect | "What is your blended rate and how is it derived?" |
| Proposal prices reports only | L1, L2, and L5 will arrive as change orders | "Map each line to the five delivery layers" |
| No individual author or named delivery lead anywhere | Delivery will be assigned after signature | "Name the delivery lead in the SOW" |
| Claims a Microsoft partner tier that Microsoft retired in October 2022 | The site has not been maintained since before the Solutions Partner model | "Which Solutions Partner designations do you hold today?" |
| Refuses paid discovery, offers free assessment | The assessment is a sales artifact, not an engineering one | "What is the written deliverable, and can we buy it standalone?" |
| No mention of licensing in a services proposal | The largest recurring cost is being left to you | "Model Pro vs PPU vs F SKU for our viewer count" |
| Dashboards demoed against sample data only | The data problem is unexamined | "Demo against one of our real sources under NDA" |
| Every question answered yes | Nobody has thought about tradeoffs | Ask question 10 |
What breaks in Power BI engagements
| Symptom | Root cause | Fix |
|---|---|---|
| Reports slow for everyone at 8 a.m. | Scheduled refreshes and interactive queries collide on one capacity; smoothed background load pushes the capacity into the 20-second interactive delay stage | Move refreshes off peak; enable surge protection; isolate executive content on its own capacity; optimize the model before scaling the SKU |
| Refresh fails intermittently on one source | On-premises data gateway resource contention or credential expiry | Cluster the gateway, monitor it, and put credential rotation on a calendar rather than an incident |
| Six versions of "revenue" across the tenant | No certified semantic model and no endorsement policy; every team modeled its own | Stand up one certified model per subject area, restrict certification to authorized reviewers, and retire duplicates on a published schedule |
| Costs jump after go-live | Viewer population grew past what the SKU supports, or the capacity was sized for the pilot | Re-run the F64 free-viewer arithmetic; consider a reservation once the steady-state size is known |
| Copilot returns confident nonsense | It is grounded on unmodeled, undocumented semantic models | Fix the model — descriptions, relationships, measure naming — before enabling Copilot for a business unit |
| Consultants leave and nothing is maintainable | No source control, no documentation, no named internal owner | Require Git integration or deployment pipelines and a named internal model owner in the SOW, not the retrospective |
What changed in 2026
- Power BI Premium per-capacity (P1–P5) is retiring at the end of each customer's current agreement term, with no new purchases, add-ons, or renewals. Fabric F SKUs are the replacement, purchased in the Azure portal rather than the Microsoft 365 admin center. Premium Per User is not affected. If a vendor's proposal still says "P1", the proposal is stale.
- Billing moved to Azure. F SKUs are pay-as-you-go per second with a one-minute minimum, use regional rather than global pricing, and can be paused, resumed, and resized on demand. That changes the FinOps conversation entirely: capacity is now a variable cost you can manage, not a fixed subscription you absorb.
- Capacity overage (preview) lets a capacity admin set a rolling 24-hour overage limit so that excess usage is paid off automatically instead of triggering throttling. It prevents throttling; it does not add performance. Microsoft recommends it only for F16 and higher during preview.
- Workspace-level surge protection lets admins cap CU spend per workspace within a rolling 24-hour window, with a "Mission Critical" exclusion. This is a governance control that did not exist under P SKUs.
- Master data endorsement (preview) joins Promoted and Certified as a third endorsement badge, applicable to items that contain data. Certification enablement can now be delegated to domain administrators, so each domain can name its own reviewers.
- Copilot's floor is F2. Copilot in Fabric and Power BI requires a paid F2-or-higher capacity (or P1+ where still in place) plus tenant-level enablement. Pro and PPU workspaces do not support Copilot directly without a Fabric Copilot capacity.
For how this interacts with the broader platform decision, see Microsoft Fabric consulting services, the Fabric consulting guide, and the Fabric vs Databricks comparison.
How EPC Group scopes a Power BI engagement
Discovery is paid and produces artifacts you keep whether or not you continue: a data-readiness assessment against your actual sources, a capacity model with the F SKU arithmetic run against your viewer count, a semantic model inventory, and a governance gap list. From there, L3 and L4 are fixed price, L2 is time and materials against the assessment, and L5 runs as an enablement track.
We publish the same qualification checklist we expect to be run against us. Delivery detail sits on the Power BI consulting services page, with local teams in Houston and Phoenix. Industry work such as Epic and Cerner data in Power BI is scoped the same way, with source-system complexity priced into L2 rather than discovered there. Technique references such as the running total and cumulative sum DAX guide sit under L3.
Building a shortlist? See best enterprise Microsoft consulting firms, top Power Platform consulting firms in North America, and best data governance consulting firms. Our delivery model is documented on the Microsoft Frontier Company page and at epcgroup.net.
Talk to the team that will actually deliver
If you want the five-layer scope mapped against your own sources, EPC Group runs a paid discovery that produces a data-readiness assessment, a capacity model, and a governance gap list you keep regardless of what you do next. Start at Power BI consulting services.
Frequently asked questions
How much do Power BI consulting services cost?
There is no credible single figure, because the variance sits in data readiness, which cannot be priced before assessment. What is publishable: Power BI Pro lists at $14.00 per user per month and PPU at $24.00, paid yearly, as published in July 2026, and ZipRecruiter put the US Power BI consultant average at $54.60 per hour on 27 July 2026. Build the total from the five delivery layers rather than accepting a blended number.
Should I hire a firm or an individual contractor?
A contractor is cheaper per hour and sufficient for report production against a modeled source. A firm is worth the premium when you need semantic model engineering, capacity design, governance, and continuity across staff turnover. If your requirement is genuinely L4 only, do not pay firm rates for it.
Is fixed price safer than time and materials?
Only where scope is knowable. Fixed price on unassessed data sources transfers risk to you through change orders while appearing to remove it. The defensible structure is fixed price for discovery, semantic model, and reports; time and materials for data remediation.
Do we still need Power BI Premium?
Power BI Premium per-capacity SKUs (P1–P5) are retiring at the end of each customer's current agreement term, with no new purchases or renewals. Fabric F SKUs replace them and are purchased through Azure. Premium Per User is a separate per-user license and is not affected by the retirement.
What is the F64 threshold everyone mentions?
On a Fabric capacity of F64 or larger, users with a Fabric Free license and the Viewer role can view Power BI content. On F2 through F32, every viewer needs Power BI Pro or PPU. For organizations with large read-only audiences, that single threshold often decides the entire licensing strategy.
How long does a Power BI implementation take?
Duration is a function of data readiness, not report count. Any vendor quoting a timeline before assessing your sources is quoting the reporting layer only. Insist that the schedule be issued as an output of discovery rather than an input to the sales process.
What should be in the statement of work that usually is not?
Named delivery team with committed hour percentages, semantic model ownership and handover artifact, source-control requirement, exit and transition terms with a notice period, and an explicit list of the delivery layers the vendor is *not* pricing.
How do I know if a consultant actually understands capacity?
Ask what they will do when throttling appears. A weak answer is "scale up." A strong answer explains smoothing, the 10-minute overage-protection window, the staged progression from 20-second interactive delays to interactive rejection to full rejection, and optimization before scaling.
Can consultants get us ready for Copilot?
Copilot requires a paid Fabric capacity of F2 or higher plus tenant-level enablement, but the capacity is the easy part. The work that matters is semantic model quality — descriptions, relationships, measure naming, and endorsement — because Copilot grounds on the model. Enabling Copilot on unmodeled data produces confident, wrong answers.
What is the single best question to ask a Power BI vendor?
"What are you going to tell us not to do?" A firm that cannot decline anything during a sales call will not decline anything during delivery, and unbounded scope is the most common way a BI program fails quietly.
Sources and verification
- Microsoft — Power BI pricing (Pro $14.00/user/month, PPU $24.00/user/month, paid yearly; as published at time of writing, July 2026)
- Microsoft Learn — Understand Microsoft Fabric licenses (per-user licenses, capacity licenses, F64 free-viewer threshold)
- Microsoft Learn — Power BI Premium P SKU to Fabric F SKU migration decision guide
- Microsoft Learn — Power BI Premium to Microsoft Fabric migration FAQ (P SKU retirement, regional pricing, pay-as-you-go)
- Microsoft Learn — Power BI service features by license type
- Microsoft Learn — Power BI Premium Per User
- Microsoft Learn — The Fabric throttling policy (smoothing, overage protection, throttling stages)
- Microsoft Learn — Capacity overage (preview) in Microsoft Fabric
- Microsoft Learn — Surge protection
- Microsoft Learn — Evaluate and optimize your Microsoft Fabric capacity
- Microsoft Learn — Endorsement overview (promoted, certified, master data)
- Microsoft Learn — Enable and configure Copilot in Microsoft Fabric (F2 minimum, tenant settings)
- Microsoft Learn — Roles in workspaces in Power BI
- Microsoft Learn — Best practices for lifecycle management in Fabric (Git integration, deployment pipelines)
- Azure — Microsoft Fabric pricing (reservation savings of approximately 41% versus pay-as-you-go; as published at time of writing, July 2026)
- ZipRecruiter — Power BI Consultant hourly pay, United States (average $54.60/hour; range $29.09–$72.60; retrieved 27 July 2026)
- Target citation under review — iTransition, Power BI consulting services
- Secondary target under review — Beyond Intranet, Power BI consulting services