Assisted Power BI optimisation

Power BI reports that are faster and lighter, with figures that don’t move.

Honing analyses your model, lets you pick your goal, then guides you step by step to a corrected file. 50 detection rules, and 95 % of the problems it shows come with their fix. The file you get back is read by Microsoft’s official library before it reaches you.

Analyse my report How it works

Free to start · .pbip model, connected .pbix/.rdl reports · Your data never leaves your machine

Data accuracy guaranteed

detection rules

deterministic: same causes, same diagnosis, at every analysis.

95 %

of the findings shown carry their fix

measured on the reference reports on 18 August 2026 - the rest carry a how-to guide.

real Power BI reports act as a test bench

every fix is replayed on them before release (reading of 18 August 2026).

rows of your data leave your machine

the data cache is stripped in your browser, before upload.

None of your data is sent or stored

Before anything is uploaded, your browser strips the data cache from your .pbip. Only the “empty” semantic model reaches us: its structure and its formulas (tables, columns, relationships, DAX measures, Power Query queries) - never a single row of your data. For connected reports (.pbix, .rdl), we extract only the names of the fields used, then the file is deleted. Whatever goes to the AI is wrapped as data to analyse, never as an instruction: a measure name cannot give orders to the model. Everything is hosted in Europe and erased after 7 days by default, a period you set yourself (up to 90 days).

What Honing actually does

Where DAX Studio and Tabular Editor speak to experts, Honing puts optimisation within reach of business teams - by guiding you, and by staying faithful to your results.

You choose your goal

Slim down the model, make refresh reliable, speed up visuals or data extraction. You rank your priorities, the journey reorganises itself, and a filter lets you see only what serves your goal - it follows you from one step to the next.

Free up memory

The tool spots the columns nothing uses any more (by following every dependency) and moves calculated columns out of the model’s memory by having them computed upstream - in the query (Power Query), or even directly by your database (SQL). Each column states what it saves and what it costs, across five axes.

Fetching values elsewhere, counting, ranking

A column that reads a value from another table (RELATED, LOOKUPVALUE), that counts or sums rows from another table, or that computes a ranking, a running total or a group subtotal: all of it can leave memory. The tool infers the exact join from your relationships, including across several chained tables, and hands the work to your server. When a shape cannot be reproduced faithfully, it is refused with the precise reason.

Anticipate the refresh peak

The memory that makes a refresh fail is not the model at rest, but the peak during computation. The tool simulates it phase by phase, recalibrates it against the peak you actually measured, and compares it to your machine’s budget. It warns you if your selection would go over - including when an optimisation costs more than it returns.

The errors that raise no warning

A division by zero, a condition tested twice, two identical branches, a filter that is always true, a deduplication that keeps a row at random, a circular dependency. These faults don’t crash Power BI: they produce wrong figures, silently. The tool names them, but never fixes them on its own - only you know what you meant.

The right order, and what each move unlocks

Some columns can only leave once the ones they depend on are already out. “Prepare everything” runs the passes in the right order until exhaustion, a “where to start” card tells you which column to handle first, and the optimisations a fix will unlock appear from the very first analysis instead of being discovered afterwards.

Speed up extraction

A diagnosis shows the exact point where your query stops being run by the database and falls back onto your machine (broken “query folding”), safe rewrites with identical results, and an assisted optimiser that works piece by piece on long queries - interruptible, resumable, never paying twice for what is already done.

Measure rather than estimate

The tool cannot run your model: it estimates. But you can give it real figures. For a rewritten query, it produces a check script ready to paste into your database: an “identical” verdict and a measured gain. For the rest, Power BI Desktop can export what is needed - volumetry, memory peak, Performance Analyzer, Power Query diagnostics. A single reading proves nothing: it is the before/after comparison that counts, and the tool reminds you of it.

Formulas rewritten without changing your figures

For a measure or a calculated column (DAX formulas), the AI proposes a faster version, and a guard checks that it uses exactly the same fields. A proposal that fails the check isn’t thrown away: it goes back to the model with the discrepancies named, twice at most. Refusal has the last word - a formula that changes the fields is never applied.

Your file read in full

Tables, relationships, measures, roles, the calculation functions you wrote, and the shared Power Query queries your tables reuse. Before analysing, the tool warns you if your project holds changes made in the Power Query editor but not yet applied in Power BI Desktop - otherwise it would analyse a state that is no longer yours.

Shared models taken into account

If your model feeds several reports, add them as .pbip, .pbix or paginated .rdl. We extract the fields they use: a column used elsewhere is never wrongly flagged as unused. A relationship change, in turn, shows you which visuals in which reports would need checking.

Optimised .pbip generated

In a single file: unused columns removed, calculated columns moved out, joins replacing lookups into another table, calculated tables converted to queries, duplicate measures reduced to one, snowflaked tables flattened, relationship loops broken, text keys converted to integers, join columns hidden, missing formats and descriptions filled in, the date table marked. A summary states what was applied and what was set aside, with the reason. Your original file is never modified.

What is left to do by hand

Not everything can be written into a file. A toggle permanently separates what goes into the .pbip from what falls to you, and a final step gathers those actions into a downloadable action plan, with instructions. This list shrinks as you validate: that is not a flaw, it is the point.

You stay in control of costs and of your files

Every AI call is costed and capped per day: you see what a report cost you, and a rewrite already refused isn’t resubmitted until something around it changes. On the file side, you set the retention period yourself (7 to 90 days), you can delete the source immediately, and a log keeps a trace of every deletion and extension.

How it works

From upload to optimised .pbip in four steps.

Drop in your model

Save your report in .pbip format and drag it into your workspace. If your dataset is shared, add the connected reports too (.pbip, .pbix, .rdl). Your files stay in Europe and your data never leaves your machine.

Measure, then choose your goal

First import your real table sizes (the “volumetry”, without sending any data): it is what allows ranking by real impact rather than on principle. You then say what you want to optimise first - memory, refresh, extraction, visuals - and the journey reorganises itself around your goals.

You move forward step by step

Five steps ordered from the most profitable to the most detailed: clean, slim down, restructure, refine, polish. Each proposal states its effect on five axes (model size, refresh peak, visual speed, extraction time, formula editing lag), gains AND costs. One button prepares everything in the right order, and nothing stops you taking back control column by column.

You generate, you finish, you verify

You get an optimised .pbip, you open it in Power BI and refresh. A final step lists what was left to do by hand, with its action plan. An accuracy check then compares the figures before and after, keeps a record that it was run, and flags anything falling outside the guaranteed accuracy.

“Faithful to your results”: the proof

Any tool can promise not to break anything. Here are the four checks that keep that promise here, and what each one actually caught.

The file you get back still opens

Every .pbip produced is read by Microsoft’s official library, not by our own reader - which we have proof accepts files that Power BI refuses. This step is mandatory before every release: close to 1,000 fixes replayed on the reference reports, no refusal attributable to the tool (reading of 18 August 2026).

The SQL produced is accepted by a real database

The queries the tool writes are submitted to real database engines, which compile them without ever running them. A file can open perfectly well and still carry a query the database will refuse at refresh time, weeks later, at your end. It happened once; this check was born that day.

A fix that fixes nothing gets caught

A test bench degrades a model on purpose, lets the tool detect the problem, applies the fix, then re-checks that the finding really disappeared. Two fixes were reporting “applied” without fixing anything: it found them on its very first run.

Nothing newer than your Power BI

A fix using a function your version of Power BI doesn’t know would produce a file that won’t open, or a refresh that fails - at your end, later, with nothing having warned you. The vocabulary the tool allows itself to write is declared and checked automatically, and the one recent function we accept is flagged to you where you meet it.

What the tool does not do

It cannot run your model: the gains it announces are estimates until you give it your own readings. It never fixes a logic error on its own - it shows it to you. And when it cannot faithfully reproduce a calculation, it refuses and names the obstacle, rather than producing a plausible-looking result.

Private by default

Before upload, the data cache is stripped in your browser: only the semantic model (structure and formulas, without any data) is transmitted and stored. Hosted in Europe (GDPR), automatically deleted after 7 days by default - adjustable up to 90 days, immediate deletion on request, and a log that keeps a trace of every deletion.

Ready to speed up your reports?

Create your account: every sign-up is approved manually, and you are notified by email as soon as it is activated.

Start for free