Visual standards calculator: Paid, Flexible or MacGyver?
What does it cost to implement visual standards in Power BI with 3rd-party visuals, open source, Deneb or core visuals? The guiding principle: standard first, only as bespoke as necessary. So the calculator first checks whether there is a professional reason to depart from the core visuals, and then compares across three levels: Overview with three inputs, Narrow down via requirements and red flags, Details with every assumption and source. It gives an indication with ranges, not a price quote: every assumption is editable and marked with its evidence. If this is your first time here, answer seven short questions in the guided start — it only sets fields that you will find again, every one of them, in levels 2 and 3.
Start
Seven short questions for your first indication
Every answer only sets fields that you will find again in levels 2 and 3 and can change one by one. Nothing extra is calculated here, and nothing is final: the result below updates with every answer. If you already know the calculator, skip past it with “Straight to the calculator”.
Step 1 of 7
–
Level 1
What does a standard cost you, per viewer per year?
Three inputs, everything else sits at its default: a five-year horizon, volume tiering (the more licences, the cheaper each one) and a mid-range chart mix. In each approach's bar the band shows the P10 to P90 range, and the line inside it shows P50 — the middle number: in half of the simulation runs the cost falls below it. Levels 2 and 3 get more precise.
Presets:
People who build reports with the standard visuals.
With paid visuals, only the share of viewers who open class B or C reports needs a licence (default 85%, slider in level 3; with mandatory IBCS, all of them).
Operating model
Price anchor for paid
Per licensed user per year before volume tiering (licensed headcount: all report authors plus the share of viewers with access to standard reports). It flips the ranking more than any other assumption; the sentence on the right shows all three.
–
If you want to know first what drives this number: (level 2 — the click switches you there).
Level 2
What drives the cost
The map of the model: which quantities there are, where they come from and where they run to — and every element is clickable, so you can change it right here. The overview first, then the input sections one by one.
Impact map Band thickness = share of P50
In short: inputs → assumption groups → cost items — on the left what you set, in the middle the assumptions with their provenance (colour as in the evidence bar), on the right the cost items of the selected approach. Clicking an element highlights its paths, explains it underneath and opens a field for changing it.
evidenced = price list or snippetrule of thumb = literature without measurementopinion = panel or authors' estimate
In detail: how to read the bands, lines and colours
Your inputs on the left, in the middle the assumption groups with their provenance (colour as in the evidence bar), on the right the cost items of the selected approach. Thick bands carry amounts and add up to the P50; thin lines only show which input feeds which group; dashed lines scale something without carrying an amount of their own. Amounts and shares appear on hover.
The tabs above the map switch the approach (paid, open source, Deneb, core + SVG); the map always shows the P50 cost of the selected approach. Green = evidenced (price list or snippet), grey = rule of thumb, blue = opinion — the same split as in the evidence bar in the result.
Decision tree, general no scenario applied · printable
The same logic without your numbers: the order of the questions, and at each edge which approach drops out or comes first. The tipping-point range of 1 : 12 to 1 : 26 holds for the default assumptions, a blended price of €80, no template library, the default report classes and the default licensed headcount (all report authors plus 85% of viewers) — measured across the five presets, as of v0.19. Your own scenario works out the exact value above.
Level 2
Volume assumptions
Who builds, who reads, how much gets built? Paid visuals scale with the viewers who need a licence (since v0.18 no longer with all of them, but with the share who open standard reports), every other approach scales with chart types and report authors. This ratio decides more than any list price.
People who build and maintain reports with the standard visuals.
Microsoft scenario: around 9 viewers per report author. With paid visuals not every viewer needs a licence, only the share with access to standard reports (assumption “share of viewers with access to standard reports”, default 85%); governance, user support and reader time run across all viewers.
0 = no cap. Viewer growth (step 4) is capped at this value each year: viewers cannot grow beyond the workforce. Suggestion: your workforce, or failing that three times today's viewers. Affects licence, support, governance and demand.
Reports that are built to the standard (overview, analysis, detail).
Derived from the report classes: share per class × charts per report of that class. The classes themselves can be changed further down.
Variance chart, waterfall, table with Δ columns, KPI card, time series AC/PY/PL/FC.
Chart mix by complexity (%) calculated100%
Derived from the report classes (chart mix per class in level 3, weighted by share × charts per report). Simple: bars and columns with variances, lines, KPI cards. Medium: tables with embedded charts, small multiples, scenario notation AC/PY/PL/FC. Complex: waterfall with variances, combined charts, commentary, drilldown. The panel hours apply to a 50 / 35 / 15 mix; other mixes scale the hours per approach with the complexity factors from step 4.
Time horizon
Forrester works with 3 years, Gartner TCO with 5. Ten years show how licence and maintenance overtake the upfront investment.
Report classes: who actually needs an IBCS chart?100%
ClassShare %Charts/report
A · Operational, no IBCS needLists, tables, status overviews, simple bars — built with core visuals in every approach
B · IBCS standardVariance charts, time series with scenario notation AC/PY/PL/FC, tables with embedded charts
C · IBCS advancedWaterfall with variances, small multiples, commentary, drilldown
Operating modelSelf-service share of report authors (hybrid)
Who is allowed to build standard reports? The operating model often shifts more cost than the choice of tool: it determines how many report authors are trained in depth, how many chart variants appear and how much governance is needed.
If you stay with today's approach, the switch costs €0. Every other approach carries the rebuild of the existing estate.
Reports on today's approach that would be rebuilt on a switch.
Four fine adjustments have moved to level 3: semantic models, adoption rate (standard compliance), uncertainty in the volume assumptions and the hours anchor. They are not gone, just one level deeper — their defaults still count in unchanged.
A custom visual is registered once in the tenant. A DAX UDF, by contrast, sits in functions.tmdl in the definition folder of every semantic model and has to be rolled out and maintained there one by one: twenty models mean twenty libraries. A factor on setup and initial build.
Adoption rate (standard compliance)
Share of newly built charts that actually follow the standard. The rest is built ad hoc and costed at the rates of the do-nothing option (hours per chart, maintenance rate, follow-up queries). Default 100%; the suggested values (85 / 80 / 70 / 65%) are an authors' estimate with no source. Watch the direction: in the model a low adoption rate makes paid and open source more expensive above all — their standard hours are below the ad-hoc rate, so every workaround costs more there. With core + SVG the ad-hoc rate is below the standard rate, so in the model going around the standard is almost free. The reason is the same as with the do-nothing option: the loss of value from inconsistent charts is not priced in. Adjustable per approach in level 3.
Uncertainty in the volume assumptions
Draws report authors, viewers, reports, charts per report and chart types per scenario via PERT around your inputs — paired, so the same volume applies to all four approaches in each scenario. The P10–P90 band gets wider, the ranking stays stable. Expected value, do-nothing option and every chart that uses them stay on your inputs. The order of magnitude is the real message: ±20% on the volume assumptions widens the band by only about 1 to 2%. The uncertainty in this calculator sits almost entirely in the assumptions, not in your volume assumptions — if you want to get more precise, measure hours, not reports.
Hours anchor:
Level 2
Vendor quotes
Otherwise the calculator works with price anchors from search snippets. If you have a real quote, you hold the hardest evidence in the whole model — so use it. An active quote replaces the anchor for “3rd-party paid”: price, minimum term, price escalation clause and support then come from the quote, and volume tiering stays off.
Enter quotes up to three · exactly one can be active
A quote is commercially sensitive: vendor, price, term and note appear in plain text in the share link and in the Excel export — whoever gets the link reads your terms along with it. Without an active quote everything stays as before, and the three price anchors in level 1 still apply.
Level 2
Rate the requirements
–Prioritised by MoSCoW — must (cannot do without it), should (important), could (nice to have), won’t (not this time). On a 1-to-5 scale the boxes show how well each approach meets the requirement. Must + 1 = knock-out (knock-out = the approach is excluded but stays visible in the result with its cost, so that the price of the rule can be seen), must + 2 = only via a workaround. Should and could feed into the requirements score. Clicking a box overrides the rating with your own assessment.
Requirements and fulfilment level
1 = not possible · 2 = workaround only · 3 = limited · 4 = good · 5 = very goodDefault: Daten-WG assessment, as of 09/2026. Your own ratings have a dotted border and are included in the share link.
Level 2
Red flags and governance switches
Hard constraints from the tenant, IT security or procurement. They act as knock-out filters before anything is calculated. Excluded options stay visible in the result, below the line and with a reason.
Distribution channel for the visual
Assumptions with ranges and a source badge
Level 3
Assumptions with ranges
Every assumption as a three-point estimate (min · likely · max). From these the result is computed as P10 / P50 / P90 using a PERT distribution and 1,500 simulation runs. The fields are pre-filled and do not need touching. If you change only two numbers, change the internal hourly rate and the hours per chart type. Evidence behind all assumptions:
Global assumptions
Apply to all four approaches.
Assumptions per approach
Licence indication:
Report classes: chart mix per class each row normalised to 100%
How many reports belong to which class, and how many charts they have, is set in the volume assumptions in level 2. Here you set how complex the charts are within a class. From that the calculator derives the global chart mix and the charts per report. Simple: bars and columns with variances, lines, KPI cards. Medium: tables with embedded charts, small multiples, scenario notation. Complex: waterfall with variances, combined charts, commentary, drilldown. The panel hours apply to a 50 / 35 / 15 mix.
Classsimple %medium %complex %
A · Operational, no IBCS needLists, tables, status overviews, simple bars — built with core visuals in every approach
B · IBCS standardVariance charts, time series with scenario notation AC/PY/PL/FC, tables with embedded charts
C · IBCS advancedWaterfall with variances, small multiples, commentary, drilldown
Vverified (Microsoft Learn, GitHub, vendor)Ssearch snippet of the primary source, not checked directlyFrule of thumb from blogs / consulting practiceEauthors' estimate by Daten-WG, no sourcePexpert panel: 20 role profiles simulated by a language model, not a surveyNyour own assumption (changed by you, origin in the tooltip)Avendor quote (level 2), the hardest evidence in the model
Level 2
Result and assessment
Total cost over the horizon, cost per viewer per year, upfront investment against running cost. The ranking is an indication under uncertainty in the inputs, not a price quote. Alongside P50 the cards also give the net present value (NPV) — later payments discounted at the discount rate; over three years that barely makes a difference, over ten it does. The reading guide below says in four steps what is where.
Reading guide: four blocks, in this order click to jump there
If you are seeing the result for the first time, read it from top to bottom in this order. Clicking a row jumps to that block and highlights it briefly.
Key terms explained
P10 / P50 / P90
In 10%, 50% and 90% of the 1,500 simulation runs the cost was below this value. P50 is the “middle” number, P10 to P90 the range you should plan with.
PERT
A three-point estimate (min, likely, max) from which a probability distribution is formed. The expected value is (min + 4·likely + max) / 6.
MoSCoW
Must (cannot do without it), should (important), could (nice to have), won't (not this time). A must with fulfilment level 1 is a knock-out.
Knock-out
The approach is excluded but stays visible with its cost, so that the price of the constraint can be seen.
Net present value (NPV)
Later payments discounted at the discount rate. Over 10 years it makes a difference, over 3 hardly any.
Volume tiering
Volume discount on the licence: the more users, the cheaper the individual seat. A negotiated vendor quote is calculated without tiering, because the discount is already in it.
Red flag
A hard constraint from the tenant, IT security or procurement that rules out an approach before anything is calculated.
Price anchor
The assumed list price per licensed user per year, as long as there is no real quote. The licensed headcount is all report authors plus the share of viewers with access to standard reports, not all viewers. It flips the ranking more than any other assumption.
–
These ranges come out of the chosen model assumptions. They are not an empirically validated forecast of actual project cost. The ranking holds only for the scope described (standard charts, no data model, no data preparation), the product families considered and the constraints shown. If you change an assumption, the badge still shows where the original value came from.
Ranking: cost ⟷ requirements met
100% = ranking by cost alone. Move it to the left to give weight to soft benefits (should/could fulfilment) as well. Weighted-scoring caveat: weights are subjective, which is why they are visible and adjustable.
Perspective of the charts and key figures
“Per user per year” divides every amount by the selected user base and by the horizon. The break-even then shows how far build-out cost spreads across the number of viewers.
Decision measure
Ranking, headline, key figures and the overview sentence all use this quantile. Ranking by P90 = conservative: the approach with the wider distribution loses its lead. As measured, the order only flips in the corporate preset (as of v0.19: by P50 and P75 Deneb leads, by P90 paid does — Deneb's wider distribution costs it the lead there); in the other four presets only the gap changes.
Draw assumptions paired across approaches
Paired: an optimistic hours assumption applies to all four approaches at once in each scenario, so differences are comparable pairwise. Independent: every approach draws its own random values, and then part of the difference comes from the sampling.
Charts and cost tableCost structure · break-even over viewers · sensitivity · cumulative · table
Cost structure per approach
Bars = categories (expected values), label = P50, line beneath = P10–P90. Build-out, migration and training are upfront investment; licence, maintenance and governance run every year.
Break-even over the number of viewers
A reference calculation with mid-range assumptions as a function of the number of viewers. Paid rises with every licence, the others with support, demand and governance tiers. The dots mark the number of viewers from which an approach becomes cheaper than paid.
What drives the result?
Sensitivity: each assumption set to its minimum or its maximum, all the others at their expected value.
Cumulative cost over the years
Year 0 = upfront investment (build-out, training, rollout). After that licence, maintenance, support and a share of the risk.
Cost table P50 · in €
Assessment per approachAdvantages, disadvantages, hidden costs, Fabric outlook
Assessment
Advantages, disadvantages and hidden costs
Prepared text per approach, filled out with your inputs. The hidden costs are the items that licence comparisons usually leave out.
A side note on the horizon: Fabric Apps
In public preview since May 2026 (Rayfin SDK and CLI since June 2026): data models written in TypeScript automatically create their own SQL database in Fabric plus a GraphQL API, on top of which React apps run, with Copilot support while building. It is not a reporting layer on the existing semantic model. A Fabric capacity is required, and billing runs through capacity units. Microsoft positions Fabric Apps as a complement to reports, not as a replacement. For visual standards that means a fifth route, with development cost instead of licence cost, which this calculator does not yet model. Source: Microsoft Learn, Fabric Apps overview.
Level 2
What this calculator cannot do
So that the numbers above are read correctly: where the assumptions come from, which causal chains the model does not know, and in which direction it gets things wrong.
What these numbers are
Indications under uncertainty. Not a quote, not a business case, not a price statement. The one exception: numbers from a vendor quote that you entered yourself in level 2 — those are as solid as your quote.
Where the assumptions come from
…
The assumptions the ranking hangs on — hours per chart, per approach — are among the unevidenced ones. We found no solid literature for them; the values come from a simulated expert panel and from our own experience. The evidence bar shows, per approach, how much of the cost depends on them.
Disclosure
ChartKitchen byDatenWG (the example for “open source / in-house”) and the Deneb templates are projects run by the organisers. So we are comparing our own work with purchased products here too.
Six known gaps from an independent review
An independent review of the model (09/2026) found seven causal chains that shift the result noticeably and that the calculator either does not know or gets wrong. One of them has been fixed: since v0.18 the licence volume knows about the report classes. Six are still open.
Gap
Magnitude
Works in favour of
The number and ambition of reports are price-inelastic — in reality, expensive approaches build fewer and simpler ones
10–40% of the volume
3rd-party paid
The rollout has no speed limit: the calculator warns about missing capacity but does not price it
delays payback by years
Deneb, core
No key-person risk: staff turnover is linear and identical for every approach
5–15% in small teams
Deneb, open source
Price escalation without bargaining power: a smooth annual rate instead of a jump when the contract rolls over
tail risk, hits P90
3rd-party paid
What has been learned does not decay along with staff turnover
10–20% of the rollout hours
Deneb, core
Extra capacity consumption does not grow with the existing estate and the number of viewers
2–5× where a capacity is entered
Core + SVG
Four of the six work in favour of the approaches we stand for ourselves. So the model is not neutral; it has a known bias in our own direction. The full review report, including suggestions for modelling these gaps, is openly available in the repository.
What is not included
Power BI base licences (Pro, PPU, the base price of a capacity) — they apply equally to all four approaches.
Paginated reports (RDL) — Power BI custom visuals do not exist there at all, neither purchased ones nor Deneb. A printed report pack gets built a second time in every approach.
The benefit side — only as a requirements score and as “€ per requirements point”, not as a return.
Tax questions (e.g. German withholding tax under § 50a EStG on licences from foreign vendors). Capitalisation and amortisation are included as orientation, not as accounting or tax advice.
And the most important point
The tipping point — the core number of this calculator — has moved by more than 30% twice since v0.13, both times through a correction to the model, not through new data. The direction of the statements is robust; the exact number is not. If you use it as a decision threshold, recompute it with your own numbers instead of copying it from here.
Generate a Monte Carlo pageChoose a variation band, get your own results page in a new tab
Level 3
How certain is the result? Generate a Monte Carlo page
Builds a results page of its own from your inputs: distribution of the cost per viewer, certainty of the ranking, cash against internal time, building against using, sensitivity, paid against open source, and a scale simulation at your report authors : viewers ratio. Everything is calculated locally in the browser; nothing is transmitted.
Variation band around your values±15%
Every quantity is drawn per scenario via PERT: most likely value = your input, minimum and maximum = your input minus or plus the variation band. Report authors, viewers, reports, charts per report and chart types are rounded to whole numbers.
What varies
“Volume and model assumptions”: every three-point assumption from step 4 is replaced by ± the variation band around its likely value. “Volume assumptions only”: the assumptions keep their ranges from step 4, and only report authors, viewers, reports, charts and chart types vary.
Draws
2,000 are enough for stable means and ranks, 5,000 for smooth distributions, 10,000 for shares below 1% and for comparing close placings (a few seconds of computation). The scale simulation uses a quarter of that per point.
The page opens in a new tab (you may need to allow pop-ups). It contains only the numbers from this simulation and no input fields; to pass it on, use “Save as HTML”. The ranking there is a pure cost ranking across all four approaches: requirements and red flags do not feed into it, and approaches excluded in step 5 are only marked on the page.
Sources, method, honesty noteAppendix
Appendix
Sources, assumptions, method
Honesty note: For “hours per chart” there is no solid figure anywhere in the literature or in blogs. The defaults for build-out, reuse, setup, training, maintenance, breaking updates and user support come from a simulated expert panel (20 role profiles from consultant to internal audit, estimated independently by a language model, median per value; raw data in the repository under sessions/). That is a structured opinion, not a measurement, and it is the model's most important lever. Your own experience belongs in here. Figures such as “46% faster” or “10x” were deliberately not adopted as defaults, because they conflate reading speed with build time; the “46%”, incidentally, comes from the study “More than just a standard” (blueforte / TU Munich 2019) on the IBCS standard, and not from a visual vendor. Two assumptions still live only in the code and appear in no table of assumptions: the normalisation mix of the chart mix (cx0 = 50 / 35 / 15 — the mix that the panel hours per chart apply to) and the learning-curve anchor (“the panel value is the 10th chart per report author”). Both are noticeable: move the anchor from 10 to 20 and the hours for the code-based approaches rise by around 18%; assume a panel mix of 40 / 35 / 25 instead of 50 / 35 / 15 and the complexity factor for core drops to 0.78. Prices come mostly from search snippets of the vendors' own pages (as of 09/2026) and are to be checked with the vendor before any decision.
Limits of the comparison:Paginated reports (RDL) are not included. Power BI custom visuals do not exist there — neither purchased ones, nor Deneb, nor a core visual taken from the report. If you print a report pack or send it out pixel-perfect, you build it a second time in RDL, and equally so in all four approaches; the calculator does not reflect that second build in any approach. The do-nothing option is open in the same way: an independent blind estimate (09/2026, with no knowledge of the values set here) arrived at a maintenance rate of 15 · 30 · 50% (here 20 · 30 · 45), 3 · 8 · 15 minutes of reader time per viewer per week (here 0.5 · 2 · 6) and 40 · 120 · 300 hours of follow-up queries per 100 viewers per year (here 6 · 12 · 20). All three of the higher values would support the claim that “a standard pays off” — which is precisely why they appear here as a range and not in the defaults.