methodatlas

Delivery Decision method

Monte Carlo Forecasting

Uses historical flow data for simulations and expresses delivery forecasts as percentiles.

Die Simulation zieht aus beobachtetem Durchsatz oder Cycle Time viele mögliche Zukünfte und verdichtet sie zu Perzentilen. Dadurch wird Vorhersage als Wahrscheinlichkeitsbild lesbar, das sich mit neuen Daten nachschärfen lässt und Diskussionen über Zusagen deutlich erdet.

Core question
What probability distribution for completion duration or delivered count emerges when actual historical throughput values are repeatedly resampled?
Plan Monte Carlo ForecastingMethod guide
Start comparison
Give feedbackIs this method useful to you?
Participants
1-8
Time
60–120 minutes for data check, simulation, and decision review; rerun regularly.

The practice is repeated across comparable work items.

Paper illustration for Monte Carlo Forecasting.

  1. Question

    Forecast question

    Set forecast type, remaining count/time window, start, and required risk level.

  2. Sample

    Throughput data

    Check historical throughput from the same system and cadence; retain zeros and normal variation.

  3. Simulate

    Simulation runs

    Each run samples weekly values with replacement, accumulates until remaining count is reached, and stores duration.

  4. Quantiles

    Probabilities

    Sort durations and report several quantiles as risk levels.

  5. Check

    Robustness view

    Test sensitivity to data window, scope, and WIP change; backtest prior forecasts where possible.

  6. Decide

    Forecast decision

    Choose accepted risk, record assumptions, and reforecast after scope/system change.

Add new evidence at review and version estimates with the prior state.

A good fit when

Wenn Liefertermine unsicher sind und Einzel-Schätzungen zu viel Sicherheit vortäuschen, zeigt Monte Carlo Forecasting die Bandbreite plausibler Enddaten. Historische Flussdaten werden so nutzbar, dass Risiko, Wahrscheinlichkeit und Erwartungswert sauber auseinanderhalten bleiben.

Ideal for

  • Release forecasting
  • Kanban flow
  • Delivery date communication

Not good for

  • Teams without history
  • Very heterogeneous items
  • One-off large projects without comparison data

Outputs

  • Forecast decision

    Reproducible Monte Carlo forecast with input sample, sampling logic, run count, P50/P85/P95, assumptions, sensitivity, and reforecast trigger.

Monte Carlo Forecasting · working structure

Fictional example

Fictional values show the actual method fields.

Monte Carlo Forecasting · working structure
Forecast partInput/logicFictional valueCheckOutcomeDecisionReforecast
QuestionAdd evidence and rationale.Date for quantityWhen will 30 items finish?One count question; start todayOpenP85 internal, P95 externalRemaining count changes
SampleAdd evidence and rationale.Historical weekly throughput5,7,4,6,8,5,0,7,6,4,9,5 per weekSame team/Done/WIP; keep zero12 empirical valuesNo mean divisionPolicy change
SimulationAdd evidence and rationale.Empirical resampling10,000 paths; sample with replacement to ≥30Each draw=historical weekDuration per pathSort quantilesCheck data window
QuantilesAdd evidence and rationale.Sorted durationsP50 6; P85 7; P95 8 weeksNo average dateDistribution, not pointExternal 8 weeksWeekly

  1. 1. Input/logic

    Question
    Date for quantity
    Sample
    Historical weekly throughput
    Simulation
    Empirical resampling
    Quantiles
    Sorted durations

  2. 2. Fictional value

    Question
    When will 30 items finish?
    Sample
    5,7,4,6,8,5,0,7,6,4,9,5 per week
    Simulation
    10,000 paths; sample with replacement to ≥30
    Quantiles
    P50 6; P85 7; P95 8 weeks

  3. 3. Check

    Question
    One count question; start today
    Sample
    Same team/Done/WIP; keep zero
    Simulation
    Each draw=historical week
    Quantiles
    No average date

  4. 4. Outcome

    Question
    Open
    Sample
    12 empirical values
    Simulation
    Duration per path
    Quantiles
    Distribution, not point

  5. 5. Decision

    Question
    P85 internal, P95 external
    Sample
    No mean division
    Simulation
    Sort quantiles
    Quantiles
    External 8 weeks

  6. 6. Reforecast

    Question
    Remaining count changes
    Sample
    Policy change
    Simulation
    Check data window
    Quantiles
    Weekly

All values are fictional.

Reference · original teaching example

First step

Choose one question: when will N items finish or how many by date D; check service boundary, Done, data window, and comparability.

Open the method guide

Good to do first

Paper illustration of a four-column Kanban board with limited ongoing work, a visible blocker and a review loop.
Engineering
Kanban

Supplies the consistent throughput history that the simulation uses as its sample.

Paper illustration for NoEstimates.
Agile
NoEstimates

Supplies flow metrics instead of point estimates as the basis for the simulation.

Where to go next?

Paper illustration of RAID Log with a method-specific labelled workspace.
Delivery
RAID Log

Enters the accepted risk and forecast assumptions into a running log.

Paper illustration for Communication Plan
Delivery
Communication Plan

Communicates the chosen delivery date as a percentile to stakeholders.

Deep dive

Purpose

Helps clarify scope, sequence, and delivery flow in concrete terms. It makes work, boundaries, and order steerable. The result is captured as Forecast Percentiles, Throughput Dataset, and Risk Communication.

How it works

The simulation draws many possible futures from observed throughput or cycle time and condenses them into percentiles. This makes forecasting readable as a probability picture that can be sharpened with new data and clearly grounds discussions about commitments.

In detail

Monte Carlo Forecasting uses historical delivery data to simulate many plausible future paths. Instead of estimating a single duration, throughput or cycle time is drawn randomly from observed data and aggregated across many simulations. The result is a probability distribution, such as a 50, 70, or 85 percent statement. That lets teams communicate risk explicitly and refresh forecasts as new data arrives.

Facilitation

Clarify system boundaries, data window, and forecast question up front. Communicate probabilities instead of guarantees and show assumptions transparently. Refresh forecasts regularly and use deviations to improve flow policies and slicing.

The method guide walks execution through 6 phases, timing, 6 pitfalls, and clear stop criteria.

Open method guide

Your perspective

Does this method help with your work?

Your feedback helps shape the next improvements to methodatlas.

Sources and topics

Topics
ForecastingFlowDelivery
Sources
Monte Carlo Forecastingprokanban.orgProKanban.org; a pocket guide to Monte Carlo forecasting using throughput simulations instead of averages.
In playbooks
Improve Estimation and ForecastingMove from intuition to shared sizing, uncertainty, and delivery forecast.
Artifact templates
Monte Carlo Forecasting: WorksheetSet subject, data, roles, flow, and review.
markdown

monte-carlo-forecasting-working-template.md

Set subject, data, roles, flow, and review.

Monte Carlo Forecasting: Worksheet

Question

To be confirmed

Desired outcome

To be confirmed

Scope

Practice run Time: 60–120 minutes for data check, simulation, and decision review; rerun regularly.

Complete this worksheet on paper or in your own document during the work.

Preparation for this scope

Method setup

  • Materials: Remaining count or target date, consistent historical throughput sample, system/WIP context, many-run simulation, and quantile report
  • Roles: Service owner · flow/data expert · delivery team · decision makers
  • Advance information: Choose one question: when will N items finish or how many by date D; check service boundary, Done, data window, and comparability.
  • Overall time needed: 60–120 minutes for data check, simulation, and decision review; rerun regularly.
  • Setup: Prepare a workspace with Question · Sample · Simulate · Quantiles · Check · Decide; separate example from live data.

Prepare for the work steps

Question: Forecast question

  • Choose one question: when will N items finish or how many by date D; check service boundary, Done, data window, and comparability.
  • Remaining count or target date, consistent historical throughput sample, system/WIP context, many-run simulation, and quantile report

Sample: Throughput data

  • Forecast question

Simulate: Simulation runs

  • Throughput data

Quantiles: Probabilities

  • Simulation runs

Check: Robustness view

  • Probabilities

Decide: Forecast decision

  • Robustness view

Monte Carlo Forecasting · working structure

Forecast partInput/logicFictional valueCheckOutcomeDecisionReforecast
Question
Sample
Simulation
Quantiles
  • Question: Add evidence and rationale.
  • Sample: Add evidence and rationale.
  • Simulation: Add evidence and rationale.
  • Quantiles: Add evidence and rationale.

The blank worksheet contains prompts only.

Reference · original teaching example

Question: Forecast question

Expected artifact: Forecast question

Set forecast type, remaining count/time window, start, and required risk level.

Entry:

...

  • Is exactly one quantity or date question expressed in count units?

Sample: Throughput data

Expected artifact: Throughput data

Check historical throughput from the same system and cadence; retain zeros and normal variation.

Entry:

...

  • Are period, team, WIP policy, and Done comparable enough?

Simulate: Simulation runs

Expected artifact: Simulation runs

Each run samples weekly values with replacement, accumulates until remaining count is reached, and stores duration.

Entry:

...

  • Does each draw correspond to one observed throughput period?

Quantiles: Probabilities

Expected artifact: Probabilities

Sort durations and report several quantiles as risk levels.

Entry:

...

  • Are P50/P85/P95 explained as by-with-likelihood rather than guarantees?

Check: Robustness view

Expected artifact: Robustness view

Test sensitivity to data window, scope, and WIP change; backtest prior forecasts where possible.

Entry:

...

  • Does the decision survive plausible data windows?

Decide: Forecast decision

Expected artifact: Forecast decision

Choose accepted risk, record assumptions, and reforecast after scope/system change.

Entry:

...

  • Is a concrete action tied to the quantile?

Open questions and next steps

...

Method guide: https://methodatlas.meierhoff-systems.de/en/methods/monte-carlo-forecasting/run-sheet

Alternatives and neighbors

Fitting alternatives with decision guidance, followed by other related methods.

All methods
Paper illustration for PERT Estimation.
Decision Making
PERT Estimation

Instead of Monte Carlo Forecasting, choose PERT Estimation when you want to derive expected values from ranges and make uncertainty explicit.

Paper illustration for Story Points.
Agile
Story Points

Instead of Monte Carlo Forecasting, choose Story Points when you first want to understand relative complexity as a team and do not yet have flow data.

Value Stream Mapping workspace showing the question, observations, and next decision.
Delivery
Value Stream Mapping

Maps a service's path all the way to delivery and exposes wait times, waste, and bottlenecks.

Paper illustration of Cost of Delay with its method-specific working model.
Delivery
Cost of Delay

Moves scope, order, and delivery flow toward a concrete result through "adopt the initiative list", "clarify value types for each initiative, such as revenue, efficiency, or risk", and "derive a sorted sequence".

Paper illustration of Shape Up with its method-specific working model.
Delivery
Shape Up

Structures product work around shaped pitches, bets, fixed cycles, hill charts, and scope hammering.

Paper illustration for Theory of Constraints.
Operations
Theory of Constraints

Finds a system's bottleneck, aligns work around it, and cyclically checks whether flow improves.

Constraints and origin
HighWorkshop + async30-90 min Setup, danach laufend

Profile

Method form
Decision method
Participants
1-8
Time
30-90 min Setup, danach laufend
Maturity
Established
Origin
Operations Research Community
Paper illustration of RAID Log with a method-specific labelled workspace.
Next up: RAID Log