ModelChoice: Decision tree software for Excel
Decision trees that solve themselves, inside Excel.
Draw the decision on an Excel worksheet, with the probabilities and payoffs in cells you can link to the rest of your model. ModelChoice finds the best choice, and finds it again every time an input changes. The same tree then shows what could happen under each option, which estimates matter, how far each one can move before the answer changes, and what better information is worth.
What can ModelChoice do?
ModelChoice turns the decision you would sketch on a whiteboard into a working model in Excel. You draw the choices and the uncertain outcomes, put the probabilities and payoffs in cells, and ModelChoice finds the best strategy and its expected value. Each analysis then answers a question the board will ask about it. Every example below uses the same decision, the Product Launch model that comes with ModelChoice: launch at full scale, launch a minimum viable product first, or don’t launch.
How do you build a decision tree in Excel with ModelChoice?
You draw it on the worksheet. ModelChoice places decision, chance and outcome nodes on the sheet as Excel shapes, and every probability and payoff is a real cell, so you can link it to your financial model or an assumptions sheet. Change any input and ModelChoice solves the tree again straight away by backward induction. There is no Run button.
- Add choices, outcomes and payoffs in the Node Editor, or click any node for the actions that fit it.
- Build Control Panel lifts every probability and payoff into one labelled input block, with the tree linked to it.
- Copy, paste, append or replace whole subtrees, with the cells behind them, and undo any edit.
- The tree is saved, shared and versioned with the workbook. Colleagues without ModelChoice can still open it and follow the logic.
Swipe the sheet sideways to see the whole tree →
Performed By: Analyst
Date: Tuesday, October 13, 2026
Output: Decision Tree ‘Product Launch’ — Starting Node: Entire Model
| Statistic | Launch Full Scale | Launch MVP | Don't Launch |
|---|---|---|---|
| Expected Value | $70,000 | $178,200 | $0 |
| Minimum | -$450,000 | -$200,000 | $0 |
| Maximum | $700,000 | $550,000 | $0 |
| Std Deviation | $433,128 | $331,962 | $0 |
| Outcomes | 3 | 4 | 1 |
The highest average is not the whole story. Launch MVP has the best expected value, $178,200, and a 42% chance of a loss of up to $200,000. Launching at full scale averages $70,000 but loses money 70% of the time.
What could actually happen under each option?
Most spreadsheet decisions are made on the average alone. The Risk Profile report shows every possible payoff for each option and how likely it is, with the expected value, the worst and best case and the standard deviation side by side. The Cumulative Risk Profile gives the same picture as percentiles.
- Compare options on their downside, not just their average: the language risk committees and regulators use.
- Start the profile at the whole model or at any node in the tree.
- Each report is a new sheet with its own table and Excel chart, ready to copy into a board paper.
Which estimates actually move the answer?
One-way sensitivity analysis varies each input across a range and shows how the expected value of every option responds. A tornado chart ranks the inputs by how far they swing the result, a spider chart shows the path of each one, and ModelChoice marks the point where the best choice changes.
- Tornado and spider charts with a results table of every input’s swing.
- Two-way sensitivity picks the two highest-impact inputs and maps which option is best across both at once.
- Spend your time refining the few estimates that change the decision, not all of them.
One-way sensitivity: Product Launch
Expected value of each first choiceHow sure can we be? How far can each estimate move before the answer changes?
Break the Decision finds, for every input, the exact value at which the best strategy flips to another one. A robustness score from 0 to 100 sums up how fragile the recommendation is, and a Monte Carlo test varies all the inputs together to estimate how often it would be wrong. It is the defensible answer to the board’s question: how confident are we?
- Every input ranked by how close it is to flipping the decision, with the strategy it flips to.
- Force to Outcome works the other way round: it finds what would have to be true for a strategy you name to come out best.
- Separate report sheets for the verdict, the vulnerability map, decision boundaries, threats and recommendations.
Performed By: Analyst
Date: Tuesday, October 13, 2026
Model: ‘Product Launch’
| Root Choice | Best EV | Gap vs Optimal |
|---|---|---|
| Launch MVP | $178,200 | — (optimal) |
| Launch Full Scale | $70,000 | $-108,200 |
| Don't Launch | $0 | $-178,200 |
| Parameter | Baseline | Threshold | Change | Flips To | Risk |
|---|---|---|---|---|---|
| Strong Demand (probability) | 0.300 | 0.420 | +40.1% | Launch Full Scale | MEDIUM |
| Pivot Works (probability) | 0.400 | 0.259 | -35.2% | Abandon (was Pivot Product) | MEDIUM |
| Pivot Fails (probability) | 0.600 | 0.741 | +23.5% | Abandon (was Pivot Product) | MEDIUM |
| Weak Demand (probability) | 0.200 | 0.034 | -83.2% | Launch Full Scale | MEDIUM |
| Launch Full Scale (value) | -$500,000 | -$391,800 | -21.6% | Launch Full Scale | MEDIUM |
| Success (value) | $1,100,000 | $842,381 | -23.4% | Launch Full Scale | MEDIUM |
| Success (probability) | 0.700 | 0.460 | -34.3% | Launch Full Scale | MEDIUM |
| Moderate (probability) | 0.300 | 0.540 | +80.1% | Launch Full Scale | MEDIUM |
| Strong Demand (value) | $1,200,000 | $1,560,667 | +30.1% | Launch Full Scale | LOW |
| Favorable (probability) | 0.600 | 0.305 | -49.1% | Launch Full Scale | LOW |
| Unfavorable (probability) | 0.400 | 0.695 | +73.7% | Launch Full Scale | LOW |
| Pivot Works (value) | $300,000 | $205,000 | -31.7% | Abandon (was Pivot Product) | LOW |
| Moderate Demand (probability) | 0.500 | 0.182 | -63.6% | Launch Full Scale | LOW |
| Scale to Full (value) | -$450,000 | -$630,333 | +40.1% | Launch Full Scale | LOW |
| Pivot Product (value) | -$80,000 | -$118,000 | +47.5% | Abandon (was Pivot Product) | LOW |
Launch MVP is the best choice today, but its lead is small: a 40% rise in the chance of strong demand would make a full-scale launch better. Select a row in the report to see what would change the answer.
If the Strong Demand probability rises from 30% to 42.0%, the best first choice becomes Launch Full Scale.
Is a study, pilot or survey worth paying for?
The expected value of perfect information (EVPI) is the most any study could be worth: what knowing the outcome before you decide would add. The expected value of imperfect information (EVII) is what a real test of known accuracy adds. ModelChoice works out both from the tree, so you can set them against the cost of the test before you commission it.
- The report revises the tree’s probabilities for each test result and shows the best decision after each one.
- EVPI caps what any study, expert opinion or extra data could be worth.
- The verdict compares the test’s value with its cost, so the go or no-go is on one line.
Performed By: Analyst
Date: Tuesday, October 13, 2026
Test: ‘Market survey’ — Target: ‘Market Response’ — Model: ‘Product Launch’
| Metric | Value |
|---|---|
| Prior EV | 178,200.00 |
| EVPI (upper bound) | 156,540.00 |
| EVII (test value) | 44,663.00 |
| Test Cost | 25,000.00 |
| Net Value (EVII − Cost) | 19,663.00 |
| Signal | P(Signal) | P(Strong Demand|Signal) | P(Moderate Demand|Signal) | P(Weak Demand|Signal) | Best Decision | Optimal EV |
|---|---|---|---|---|---|---|
| Survey positive | 0.5350 | 0.4766 | 0.4673 | 0.0561 | Launch Full Scale | 261,682.24 |
| Survey negative | 0.4650 | 0.0968 | 0.5376 | 0.3656 | Launch MVP | 178,200.00 |
| State \ Signal | Survey positive | Survey negative |
|---|---|---|
| Strong Demand | 0.8500 | 0.1500 |
| Moderate Demand | 0.5000 | 0.5000 |
| Weak Demand | 0.1500 | 0.8500 |
In the example, a market survey before the launch decision costs $25,000. It comes back positive 85% of the time when demand would be strong, half the time when it would be moderate, and 15% of the time when it would be weak.
Try a different survey
When demand would be moderate, the survey comes back positive half the time.
How do you explain the decision to the board?
The Decision Brief puts the analysis on one page: the recommendation and how robust it is, the expected value, the chance of a loss, the estimates that drive the result, the nearest threats and a plain-English summary. It is rebuilt in seconds when an input changes, and Export Brief to PDF writes it as a single page.
- Policy Suggestion ranks every strategy by expected value, with its gap to the best, for when the room has a favourite.
- Policy Tree shows only the chosen path, for a clean slide.
- Strategy Table lists every strategy with its full decision path and expected value, for audit.
Decision Brief
13 Oct 2026
| Threat | Strategy | Flip Condition | EV Gap |
|---|---|---|---|
| MEDIUM | Launch MVP / Scale to Full / Abandon | Pivot Works −35% | $15,200 |
| MEDIUM | Launch Full Scale | Strong Demand +40% | $108,200 |
| LOW | Launch MVP / Stay Small / Pivot Product | Success −36% | $165,000 |
Change the weights
ModelChoice scales the weights to add up to 1.
Performed By: Analyst
Date: Tuesday, October 13, 2026
Model: ‘Supplier Selection’
| Option | Type | Financial | Quality | Delivery | Sustainability | Composite |
|---|---|---|---|---|---|---|
| w=0.30, Max | w=0.30, Max | w=0.20, Max | w=0.20, Max | w=1.00 | ||
| Local Supplier | Raw | 1,200,000.00 | Excellent | OnTime | Good | 0.636 |
| Norm. | 0.602 | 0.617 | 0.425 | 0.925 | ||
| Overseas Supplier | Raw | 930,000.00 | Acceptable | Late | Poor | 0.333 |
| Norm. | 0.477 | 0.433 | 0.250 | 0.050 | ||
| Dual Source | Raw | 930,000.00 | Excellent | OnTime | Fair | 0.573 |
| Norm. | 0.477 | 0.717 | 0.375 | 0.700 |
Local Supplier scores best with these weights, 0.062 ahead of Dual Source.
Try Sustainability at 0% and Quality at 60%: Dual Source comes out on top.
What if the decision is not only about money?
Multi-criteria decision analysis (MCDA) scores each outcome on criteria such as quality, delivery, safety or sustainability as well as the financial payoff. You give each criterion a weight and a scale of named options, and ModelChoice works out a composite score at every outcome and rolls the tree back on it.
- Enter weights directly, or derive them with AHP from pairwise comparisons on Saaty’s 1 to 9 scale, with a consistency ratio that flags contradictory answers.
- Choose weighted sum, weighted geometric mean or WASPAS, depending on whether a strong score should be able to make up for a weak one.
- Utility functions (exponential, logarithmic, square root and linear) give certainty equivalents and risk premiums, and the best choice for a decision maker who is averse to risk.
What if an outcome is a range, not three branches?
Some uncertainties are better described by a continuous distribution than by a few branches: a price, demand, a duration, an interest rate. With ModelRisk, Vose’s Monte Carlo add-in, you can put a distribution such as VosePERT, VoseNormal or VoseLognormal into any payoff or probability in the tree and simulate it thousands of times.
- Type a distribution into a cell, as in any ModelRisk model, or apply one to a range of nodes in the Distribution Editor.
- Worksheet functions such as MC_NodeValue and MC_NodeDecision can be simulation outputs, so you get the distribution of any node’s value.
- The Simulation tab of Model Settings controls how decisions are made in each iteration.
Three ways to simulate a tree
With ModelRisk installed=VosePERT(900000, 1200000, 1600000)A fast estimate of the best strategy and its expected value when some inputs are continuous.
Each iteration follows one complete path through the tree and records its payoff, so you see the full spread of outcomes.
Each iteration picks the decision that would have been best for that iteration’s outcomes, giving the EVPI under continuous uncertainty.
Illustrative: a ModelRisk distribution typed into the Strong Demand payoff cell of the Product Launch tree in section 01.
Does it work with what we already have?
ModelChoice is an Excel add-in, not a separate application, so a tree lives next to the data that feeds it. Existing PrecisionTree models come across in one click, trees can be saved as JSON, and everything the ribbon does can be scripted.
Import from PrecisionTree
The Import command converts a copy of a PrecisionTree workbook (.xls or .xlsx) into a native ModelChoice tree. The original is left untouched.
PrecisionTree alternative guide →JSON export and import
Save any tree, with its settings and MCDA set-up, as a JSON file. Keep it in version control, compare versions, or build trees from a database.
Automation API
Headless commands for the main analyses, callable from Python, .NET or VBA through Application.Run, for batch runs and in-house tools.
Worksheet functions
Functions such as MC_NodeValue, MC_NodeDecision and MC_NodeProbability return a node’s value, decision or probability in any cell.
Audit trail in the workbook
Key values are named ranges (MC_V_…, MC_P_…), every formula can be inspected, and sheet protection locks a finished model.
AI assistants
The open modelchoice-mcp server lets assistants such as Claude build and edit trees, import PrecisionTree models and run ModelChoice reports through conversation.
modelchoice-mcp on PyPI →Seven languages
Ribbon, dialogs, messages and help in English, German, Spanish, French, Japanese, Portuguese and Russian.
59 example models
Ready-to-run trees for launches, investments, R&D, expansions, operations and risk decisions, each with a description of the problem it solves.
Explore ModelChoice for your industry
Where decision trees fit in your industry, the decisions they settle and the products that work alongside ModelChoice, in your industry’s own terms.
Everything in ModelChoice
The full list. Everything is included in every ModelChoice licence; Monte Carlo simulation also needs ModelRisk. Download the ModelChoice brochure (PDF, 2.8 MB) ↓Download the one-page overview (PDF, 0.6 MB) ↓
Common questions
What is ModelChoice used for?
ModelChoice is a decision analysis add-in for Microsoft Excel. Teams use it to compare the options for a big, uncertain decision: whether to go ahead now or test first, which project or product to fund, how to structure a deal, whether a study is worth paying for. It finds the option with the best expected value, shows the risk that comes with each option, tests what would change the answer and works out whether more information is worth its cost.
How is a decision tree different from a spreadsheet NPV?
A spreadsheet NPV gives one number for one scenario. A decision tree lays out the choices you can make, the events that could follow with their probabilities, and the later decisions you would take once you know more. That captures the value of being able to change course, such as scaling up only after a small launch goes well, which a single NPV leaves out.
When is a decision tree better than a Monte Carlo simulation?
Use a decision tree when the question is which way to go, especially when there are choices to make along the way: test first or go ahead, continue or stop after a result. Use a Monte Carlo simulation when the question is how wide the range of outcomes is, such as the spread of a project’s cost. When a decision needs both, ModelChoice can run a ModelRisk simulation through the tree.
How much does ModelChoice cost?
ModelChoice is sold as an annual licence, for a single computer or as a network licence, in one edition with every feature on this page included and a fully functional 15-day trial. Current prices and volume discounts are on the pricing page. Our comparison of seven decision analysis tools lists the prices other vendors publish.
Do I need ModelRisk to use ModelChoice?
No. ModelChoice works on its own in Excel. You need ModelRisk only if you want continuous distributions in the tree and a Monte Carlo simulation through it.
Can I import my PrecisionTree models?
Yes. The Import command converts a copy of a PrecisionTree workbook (.xls or .xlsx) into a native ModelChoice tree and leaves the original untouched. Our PrecisionTree alternative guide compares the two products.
Where is my decision tree stored?
In your Excel workbook. ModelChoice draws the tree as Excel shapes next to the cells that hold its values, so it is saved, shared and version-controlled with the workbook, and colleagues without ModelChoice can still open it. You can also export any tree to JSON.
What do I need to run ModelChoice?
Microsoft Excel 2016, 2019, 2021 or Microsoft 365, 32-bit or 64-bit, on Windows 10 or 11, with the .NET 9 runtime. ModelChoice does not run on macOS. Full details are on the system requirements page.
Is there a free trial?
Yes. The 15-day trial is the complete product, with every analysis and all 59 example models, including the Product Launch model used on this page.
Build your first decision tree free for 15 days
The full product: draw a tree with your own numbers and see its risk profile, robustness score and Decision Brief. See prices and discounts
Download the ModelChoice brochure (PDF, 2.8 MB) ↓Download the one-page overview (PDF, 0.6 MB) ↓