Everything an article about FlowVisual needs: edited boilerplate in three lengths, the facts behind it, images at full size, quotes cleared for use — and the section most press kits leave out: what the tool cannot do. Free to use, no permission needed.
Each text is edited and may be used as it stands — as a news item, a sidebar, a paragraph in your own copy. No sign-off, no embargo.
FlowVisual is a desktop process simulator for macOS and Windows: it runs a business process, names the constraint, and prices a planned change in money and time.
FlowVisual is a process simulator for macOS and Windows. Instead of drawing a flowchart, it runs the process: work flows through the model, load lands on the roles and systems that carry it, and a Monte Carlo stress test over thousands of working days names the constraint. Every planned change is re-simulated end to end and reported as Before/After in money and time — as a range, never as one flattering figure. It all runs locally, with no account and no cloud.
FlowVisual is a process simulator for the desks of consultants, lean leads and operations managers — macOS and Windows, built by Balane GmbH in Munich. A diagram shows the order of the steps; it cannot tell you which role runs out of hours first, what the third handover costs over a year, or whether a recommendation still holds in a bad week. FlowVisual answers that, because it runs the process: durations go in as ranges (“8 to 20 minutes”), a Monte Carlo stress test takes the model through thousands of days and reports cycle time as P50 and P90, throughput, and the odds of breaching the promise. For every change it re-simulates the whole model rather than adding up savings — which is how you see that improving the wrong step buys nothing and only moves the constraint along. What leaves the building is a proposal PDF on your own letterhead, with an assumptions register and a P10–P90 band. Modelling and simulating are free for good.
Balane GmbH is a software company in Munich. It builds tools for process work: FlowVisual for the simulation, FLOWREFY (flowrefy.com) for the method behind it. Managing director is Jonas David Höttler. The products are sold direct, with no investors behind them.
“FlowVisual”: one word, capital F, capital V, no space, not all caps. The company is “Balane GmbH”, not “Balane” alone where it matters legally. The method behind it is “FLOWREFY”, always in capitals, at flowrefy.com; it is not the product.
The short version, for someone deciding in two minutes whether there is a story here.
The process gets captured, drawn, a measure is recommended. Then the board asks: what is it worth? To this day the answer is a slide with a round number on it, and the first hard question in the room takes it apart. Between drawing tools at €15 a month and simulation software at five figures a year sits exactly the gap where most process work actually happens.
It computes. Work flows through the model as cases, queues build and drain, load lands on the roles, teams and systems that carry it. A Monte Carlo stress test takes the process through thousands of working days and says where it jams — on an average day and on a busy day, the tenth working day with a rush. Red means exactly one thing anywhere in the application: more work than capacity.
Nobody should ever have to guess where a change takes effect. So FlowVisual never adds savings together; it re-simulates the entire model for each variant, sub-processes included. That is the difference between a spreadsheet where two measures are summed and a tool that shows you: the second measure buys nothing, because the first already moved the constraint. Improving the wrong step buys roughly zero throughput — that is not an opinion, it is the result of the second run.
Beside today's constraint sits a fragility curve: at what volume does the process tip, and which step takes over as the constraint then? The options grid proposes variants on its own — relieve the constraint, automate the priciest step, close a media break, bundle the best two — re-simulates each one in full and ranks them by effect. Before and After stand side by side, and every figure moves while you type.
While inputs are estimates, the result is marked “provisional”, with the figure that ought to be measured first named beside it. In the proposal every number carries its basis — client statement, estimate or measurement — and behind it a register of every assumption the calculation stands on. Plus a plausibility check that contradicts the cover page when a measure saves money but lengthens cycle time or lowers throughput. A tool that argues with its own number is worth more in front of a CFO than one that only agrees.
A proposal PDF, on your own letterhead with Pro: the constraint, the measure, the value in hours and money as a P10–P90 band, with a register of every assumption behind it and a plausibility check that contradicts the cover page when a measure saves money but lengthens cycle time. Plus the process manual with RACI, documentation, work packages, workshop minutes, Word, Visio (.vsdx), draw.io, SVG, JSON and CSV.
The section for the question every good article asks: where does the number come from? All of this is in the shipped source as described — checkable against the application itself.
Steps, decisions, drop-offs, sub-processes. Per step: who does it, which system, how long — as a range, not a figure.
400 runs. Each run is a working day with its own case volume and its own service times.
The step that binds — with load on an average day and on a busy day, and beside it the curve showing at what volume it tips.
Proposed variants, each re-simulated in full and ranked by effect. Nothing is added up.
The basis of every figure, a register of all assumptions, a plausibility check against your own recommendation.
Proposal, process documentation, raw data — on your own letterhead.
A step counts as done only when a real artefact exists: a model with steps, a run with a result, a constraint that this run actually named. A bar that fills up as you click would be the fake progress meter this product refuses everywhere else.
The day's volume is drawn, not set: a lognormal around the stated figure, steered by variability, peak chance and peak factor. The same variability also enters the queue term as arrival variance.
“8 to 20 minutes” is read as the P10–P90 of a lognormal. That way the occasional monster case beyond the upper bound can occur — a uniform draw between 8 and 20 could never produce it, and it is the one that ruins the day.
Waiting follows Kingman (G/G/1): the factor (arrival plus service variance)/2 times utilisation/(1−utilisation). Near the capacity line the term is linearised and above it driven into saturation — beyond that line a queue has no steady state left to compute.
How often a step is visited per case is solved as a fixed point rather than by a topological sort. That is the only way rework counts: a “send it back for correction” loop disappears in a sort, and half the step's load with it. Cycles are found by Tarjan's algorithm; parallel branches that re-converge count the case once, not twice.
Several steps can compute against one team, one system or one licence pool. When the pool is short, the individual step is only the symptom — the finding names the pool. Sub-processes roll their capacity and service time up into the parent step, which is why an edit three levels down moves the headline figure.
Cycle time as P50 and P90, throughput, the share of days over capacity, cost per day as a P10–P90 band — plus the average of the worst five percent of days, and flow efficiency: hands-on work divided by lead time. In real processes that tends to sit in the single digits, and it is the most uncomfortable number in the whole analysis.
The same run again at 0.8 · 0.9 · 1.0 · 1.1 · 1.2 · 1.35 · 1.5 times today's volume. What is measured is not the level but the curvature: harm grows faster than load, which is how a step that is green today can be the first to go non-linear in a busy week.
A step nobody stated a capacity for cannot win the constraint ranking — and is not painted as slack either. If not one step states anything, the result reads “constraint undetermined” rather than an invented figure.
The thresholds live in the source, not in a brochure: 85 % utilisation is the knee past which waiting inflates sharply; from ten percent of days over capacity the verdict reads “strained”. And the randomness has a fixed seed — the same model gives the same figure when you open it again. A proposal whose result changes when you recompute it would be no proposal at all.
Everything that fits in a fact box. Figures that change are pulled from the same source as the rest of the site, so they are current here.
Prices in euro, consumer prices including VAT, billed annually. One licence key covers Windows and macOS at the same time.
Every shot comes out of the real application window, rendered at 3520 × 2000 pixels — no mockup, no rebuilt interface. What you see is a demo process; there is no client data in it.
A 3520-pixel-wide screenshot is unreadable in a 700-pixel column. These three crops are cut from the same files and stand at full size — each shows exactly one thing.
The icon is an F whose crossbar is red: the same colour that means “overloaded” inside the application. The accent is violet (#6B5BB0), the typeface is IBM Plex. The wordmark ships as an outlined SVG so it sets correctly without the font installed.
Use them verbatim, no further sign-off. For an interview or a question of your own, the contact is below.
A process diagram is a claim. The moment you run it, it becomes checkable — and in most workshops the result surprises the person who drew the diagram first.
The sentence the whole application hangs on: nobody should ever have to guess where a change takes effect. That is why we never add savings together but re-simulate the whole model for each variant. It is the only way to see that the second measure buys nothing because the first already moved the constraint.
We never report a single number. A saving without a range is the first thing a sceptical room takes apart — and rightly so.
The tool has to hold up in the room where the decision is made. That is why there is a presentation mode and no slide export: the question from the table gets answered on the running model, not deferred to the next meeting.
The limits as you will find them when you check. It saves you research.
We publish no install counts, no reference logos and no case studies with client names.
FlowVisual reads no event logs from ERP systems and watches nobody's screen. Volumes and times come from the people who do the work, with no data project up front.
The application reads documents only through an AI endpoint you set up yourself. Every suggestion quotes the sentence it came from and enters the model only once you accept it.
No account, no server, no simultaneous editing. With Pro, a team shares a project folder on its own drive or in SharePoint.
While figures are estimates, the result is marked “provisional”, and the proposal states what each number rests on: client statement, estimate or measurement.
The tool finds the constraint and prices the change. Which process to touch and what the organisation can absorb is decided by people. The approach for that is the FLOWREFY method on flowrefy.com.
Review keys for the Pro version are free for editorial staff and trade writers — with no strings attached to your coverage and no question about what you intend to write.
Both stores carry the same version, and one licence key covers both. Modelling and simulating need no account and no key — for a review, the free scope reaches all the way to a finished PDF.
macOS 12 or newer (Apple Silicon and Intel) · Windows 10/11 (64-bit)
The shortest route to the point where you can judge whether the tool is any good: