Forecasting Without Story Points: A Monte Carlo Playbook

Build probabilistic delivery forecasts from historical throughput—plus caveats, confidence intervals, and an executive-ready chart.

Most teams still try to “estimate” delivery with story points. You don’t need them. If you have a list of items finished per day/week (throughput), you can simulate thousands of futures in seconds and answer the only questions that matter: “When will this many items be done?” or “How many will we finish by this date?” The output is a range with confidence levels (P50/P70/P85/P95) you can share with leadership—no cosplay math, just evidence.

---

The 80/20 of Monte Carlo (what actually moves the needle)

• Use item count, not story points. Count how many work items you finish per week. That’s your input distribution.
• Simulate 5k–50k runs. Randomly sample past weeks, add them up, and observe the distribution of possible outcomes.
• Report percentiles, not single dates. Share P50, P85, and P95; anchor decisions on risk appetite.
• Update weekly. As new data lands, rerun the simulation; compare actuals vs forecast percentiles.
• Keep classes of work separate. Different workflows → different throughput distributions → separate forecasts.

If you do only that, you’ll outperform point-based plans most of the time. For background and tools, see Troy Magennis’s resources at https://www.focusedobjective.com/ and the ActionableAgile docs: https://55degrees.atlassian.net/wiki/spaces/ActionableAgile/pages/2407399482.

---

What you need (and how to get it fast)

1. A clean Done date for each completed item (e.g., “Ready for Production” or “Released”).
2. A time bucket (usually 1 week). Tally items finished each week for the last 12–26 weeks.
3. A single question:
   • When will N items be done? (delivery date range), or
   • How many items by date D? (scope range)

If your system’s flow is reasonably stable, throughput is enough. You don’t need normalized points or task-hours. ProKanban’s take on trusting Monte Carlo—with stability checks—is a solid primer: https://www.prokanban.org/blog/trustingmcs.

---

Step-by-step: Throughput-based Monte Carlo (no points required)

Step 1 — Build your throughput history
Export your Done items, group by week, and list weekly counts (e.g., 6, 9, 4, 10, 7, …). Visualize a quick histogram to see spread and outliers. Businessmap’s overview is handy if you’re new to throughput histograms: https://businessmap.io/kanban-resources/kanban-analytics/throughput-histogram.

Step 2 — Choose the simulation type

• “When will N items be done?”
  Simulate future weeks by randomly sampling past weekly counts with replacement, summing until the total ≥ N. Record weeks taken; repeat 10k times.
• “How many by date D?”
  Count how many weeks until D, sample that many historical weeks, sum items; repeat 10k times.

Step 3 — Read the results
You’ll get a distribution of completion dates (or item counts). Pull P50, P70, P85, P95. Present them as ranges, not promises. Scrum.org’s “Getting to 85th percentile” article is a useful mental model for explaining percentile thinking to stakeholders: https://www.scrum.org/resources/blog/getting-85-agile-metrics-actionableagile-part-1.

Step 4 — Turn it into decisions

• Capacity choices (pull less/more).
• Release planning (ship at P85; monitor drift).
• Trade-offs (scope, staffing, sequencing) when the forecast misses a business date.

Tools (pick one):

• ActionableAgile (Jira, Azure Boards): https://55degrees.atlassian.net/wiki/spaces/ActionableAgile/pages/2407399482.
• Focused Objective spreadsheets (quick start, no tooling): https://www.focusedobjective.com/courses/using-the-monte-carlo-forecasting-spreadsheets.

---

Executive-ready chart (copy this pattern)

Title: Delivery forecast for 50 backlog items (weekly throughput, last 20 weeks)
Subtitle: “Date ranges at different confidence levels; assumes current system behavior continues.”

| Percentile | Forecasted completion date |
|-----------:|----------------------------|
|     P50    |  2025-11-21                |
|     P70    |  2025-12-02                |
|     P85    |  2025-12-12                |
|     P95    |  2025-12-23                |

Recommended visual: a cumulative probability fan (CDF line or percentile bars) with vertical markers at P50/P70/P85/P95 and a thin marker for the business target date. Keep it minimal: one chart, four markers, short footnote:

Method: 10,000 Monte Carlo simulations from historical weekly throughput (last 20 weeks). Assumptions: team, policies, and arrival quality remain comparable to history.

You can generate this directly in ActionableAgile (toggle percentile boxes on the Monte Carlo “How many/When” charts) or replicate in a spreadsheet by charting the percentile table. The AA docs show the exact chart areas and percentile boxes you can enable. 55degrees.atlassian.net

---

Practical spreadsheet recipe (Google Sheets / Excel)

You only need a column of weekly throughput numbers.

1. Put historical weekly counts in Throughput!A2:A200.
2. Decide N (items) or W (weeks).
3. When will N items be done?
   • For each simulation row (e.g., 1..10,000), keep sampling random rows from A2:A200 and summing until ≥ N.
   • In Sheets, use INDEX(A:A, RANDBETWEEN(2, lastRow)) inside a small script or iterative range; in Excel, use Office Scripts or Power Query to do the loop cleanly.
4. How many by date D?
   • Compute W = weeks until D.
   • For each simulation row, sum W random samples from A2:A200.
5. Compute percentiles with PERCENTILE.INC(sim_results, {0.5, 0.7, 0.85, 0.95}).
6. Chart the percentile table (bars or a simple line); mark your target date/scope.

If you prefer a ready-made approach, Troy Magennis’s spreadsheets are plug-and-play and widely used in industry. focusedobjective.com+1

---

Caveats & failure modes (read before you brief execs)

• Unstable system = unreliable forecast. Large, recent policy shifts (WIP limits, definition of Done, staffing, demand mix) will distort historical throughput. Run a quick control chart or just eyeball the weekly counts for step changes before simulating. ProKanban’s “What Monte Carlo is (and isn’t)” shows how to think about throughput stability. ProKanban
• Mixed classes of work. Don’t blend “tiny bugs” with “large cross-team features.” Forecast them separately, then combine ranges if needed.
• Batching & holidays. Release trains or seasonal slowdowns create lumpy throughput; either exclude anomalous weeks or annotate your forecast.
• Data quality. If “Done” isn’t reliable, fix that first. Garbage-in will give you impressive-looking garbage-out.
• Communication risk. A single date is easier to quote but kills trust later. Teach percentiles early (P50 is a coin flip; P85 means “5 out of 6 times, we beat this”).

Industrial Logic’s case study mirrors real-world adoption: once you show calibrated forecasts, the conversation shifts from wishing to trade-offs. https://www.industriallogic.com/blog/reckoning-with-reality-with-probabilistic-forecasting/.

---

Make it stick (cadence, governance, and culture)

• Weekly refresh. Re-run the simulation every week; track actuals vs percentile to calibrate.
• Decision thresholds. Pre-agree on triggers (e.g., if P85 slips past the business date by >2 weeks, we either reduce scope by 15% or add capacity).
• Evidence-based leadership. Tie forecasts to outcomes you care about—market impact, cycle time goals, and investment decisions. If your org uses Evidence-Based Management, hook forecasts into the way you inspect and adapt. Learn more: https://igorlakic.com/agile-coaching/evidence-based-management/.
• Upskill your product delivery muscle. Forecasts are one side of the coin; solid Scrum/flow practices are the other. If you want help getting both right, start here: https://igorlakic.com/agile-coaching/scrum/.

For broader coaching and tailored enablement, see: https://igorlakic.com/agile-coaching/ or reach out directly: https://igorlakic.com/contact/.

---

30-minute quick start

1. Export 6–12 months of Done items.
2. Build a weekly throughput list.
3. Load the data into ActionableAgile or a Monte Carlo spreadsheet.
4. Run 10k simulations for the specific question.
5. Share P50/P70/P85/P95 and the one-page chart.
6. Set a weekly refresh reminder and compare forecast vs reality.

That’s it. No story points. No debates. Just a clear, probabilistic read on delivery risk executives can act on.