"When will it be done?" is the question teams answer worst. Most answers take an average and divide. You have 30 items left, you finish 1.23 a day, so that is 24 days.

The trouble is that 24 days is close to a coin flip. Half the time you will be late, and nobody who hears the number knows that.

A Monte Carlo forecast gives you a range with odds attached, and you do not need story points or a statistics library. You need one thing: how many items your team finished on each recent day.

The whole idea

Assume the near future looks like the recent past. To simulate one possible future, keep picking a random past day and adding what the team finished that day, until the work is done. Count the days. Do that 10,000 times and you have 10,000 possible finish times.

The code

function simulateWhen(samples, n, trials = 10000) {
  if (!samples.some(x => x > 0) || n <= 0) return null
  const out = new Array(trials)
  for (let t = 0; t < trials; t++) {
    let done = 0
    let days = 0
    while (done < n) {
      done += samples[Math.floor(Math.random() * samples.length)]
      days += 1
    }
    out[t] = days
  }
  return out.sort((a, b) => a - b)
}

The first line matters. If the team finished nothing on every sampled day, the loop would never end. The version I run in production also caps a trial at 3,650 days and uses a seeded random generator, so the same data always gives the same answer.

To read a result, take a percentile of the sorted array:

const percentile = (sorted, p) =>
  sorted[Math.min(sorted.length - 1, Math.ceil((p / 100) * sorted.length) - 1)]

A real run

Here are 30 days of sample throughput, with 37 items finished in total. Notice how lumpy it is, with lots of zeros and a five:

const samples = [0,2,1,0,3,1,0,0,4,1,2,0,1,1,3,0,2,0,1,5,0,1,2,0,1,3,0,0,2,1]

For 30 items, I got this (a run without a seed may differ by a day):

Dividing by the average gave 24.3 days, which sits right next to the 50% line. If you promise that number, you are promising a coin flip. "30 days at 85%" is an answer a stakeholder can plan around, and it is honest about the risk.

The mirror question: how many by this date?

Sometimes the date is fixed and the scope moves. The simulation barely changes: run a fixed number of days and add up what got done.

function simulateHowMany(samples, days, trials = 10000) {
  const out = new Array(trials)
  for (let t = 0; t < trials; t++) {
    let done = 0
    for (let d = 0; d < days; d++) {
      done += samples[Math.floor(Math.random() * samples.length)]
    }
    out[t] = done
  }
  return out.sort((a, b) => a - b)
}

There is one trap, and I think it is the easiest bug to ship here. For "when", more confidence means more days, so you read the 85th percentile. For "how many", more confidence means fewer items, so the 85% answer is the 15th percentile. With my 20-day run:

If you read the 85th percentile by mistake, you will confidently promise far more than is likely.

When this will mislead you

• The team changed. If people left or the process changed, older history describes a different team. Use a recent window, such as 90 days.
• Items vary hugely in size. The simulation counts items and does not look at their size. Split the large ones first.
• The zeros are missing. Days when nothing finished have to stay in the data. Leave them out and every forecast comes out too optimistic.
• Each day is treated as independent. A real team has good and bad weeks in a row, so the true range is often wider. Treat the result as a guide and not a guarantee.

Try it on your own data

If you would rather not write this yourself, RetroLift has a free Monte Carlo forecast tool. You can paste a CSV export with start and done dates (Jira exports work), or type your daily or weekly counts. It runs entirely in your browser, so your data goes nowhere, and it does not need an account.

If you try this on a real project, I would like to hear how far off the 85% line was.

Lean Coffee is a meeting with no agenda prepared in advance, but plenty of structure. The people in the room write the topics, vote on what matters most, and talk through them in short timeboxes. It sounds loose. In practice it is one of the most focused meeting formats there is, because nobody spends time on things the group did not choose.

The format was started in Seattle in 2009 by Jim Benson and Jeremy Lightsmith, who wanted a regular gathering to talk about Lean and agile topics without organising speakers or agendas.

When Lean Coffee works well

• Community of practice sessions, such as a monthly Scrum Master or engineering-lead catch-up.
• Team meetings that have gone stale, where the same person reads the same updates every week.
• Retrospectives for mature teams who already know what they want to discuss.
• Leadership or all-hands Q&A, where the questions should come from the audience.
• Meetups and conferences, as an open session alongside planned talks.

It is less suited to meetings that must reach a specific decision by a deadline, or that need a presentation of new information first.

The Lean Coffee format, step by step

To run Lean Coffee, set up a board with To discuss, Discussing and Discussed columns, have everyone write topics, dot vote, then discuss the top topic in a timebox of about seven or eight minutes. When the timer ends, a thumbs vote decides whether to continue or move to the next topic.

1. Set up the board

Draw three columns: To discuss, Discussing and Discussed. On a physical wall this is three strips of tape. Online, it is three columns on a board.

2. Brainstorm topics (5 minutes)

Everyone writes topics they would like to talk about, one per card. A topic can be a question ("How do we handle urgent requests mid-sprint?"), a problem, or an idea. Each person briefly reads out their cards so everyone understands them.

3. Vote (2–3 minutes)

Each person gets two or three votes, usually as dots. They can put more than one vote on a topic they care about. Sort the To discuss column so the most-voted topic is at the top. Group near-identical topics first.

4. Discuss in timeboxes

Move the top topic to Discussing and start a timer. Seven or eight minutes is a common first timebox. The person who wrote the topic usually starts by saying what they want from the conversation.

5. Check whether to continue

When the timer goes off, everyone votes with their thumb:

• Thumb up – keep going on this topic.
• Thumb sideways – either way is fine.
• Thumb down – let us move on.

If most thumbs are up, add a shorter timebox (often half the first one, so four minutes). If not, move the card to Discussed, note any takeaways, and pull the next topic.

6. Close (5 minutes)

Capture any actions or decisions, with owners. Topics that were not reached can carry over to the next session or be handled elsewhere. Finish with a quick round of "what was most useful today?".

A sample 45-minute run sheet

In 30 minutes of discussion you will typically cover three or four topics. That is the point: the group's top three or four, not everything.

Facilitation tips

• Keep the timer visible. The timebox is what gives Lean Coffee its energy. Let everyone see it.
• Be strict about the thumbs vote. It feels abrupt at first. It is also the thing that stops one topic eating the whole hour.
• Let the topic owner steer. Ask them up front: "What would a useful outcome be?". It might be a decision, advice, or just hearing different views.
• Write takeaways on the card. When a topic moves to Discussed, add a line or two. This becomes your meeting notes for free.
• Keep groups small. Five to eight people is ideal. With 20, split into parallel tables or breakout rooms and share highlights at the end.
• Do not let the facilitator dominate. If you are facilitating, you can add topics, but be careful about voting your own topics up.

Running Lean Coffee online

Remote Lean Coffee works well and is often easier than the physical version, because voting and sorting are instant.

• Use a shared board with the three columns, so topic writing happens in parallel.
• Use a built-in timer rather than someone's phone, so everyone sees the same countdown.
• Run the thumbs vote with reactions or a quick poll. Cameras on helps here.
• For distributed teams, say Adelaide and Singapore, consider collecting topics a day ahead so people in quieter time slots still get a say.

MeetingLift has a Lean Coffee mode where people propose topics, vote and timebox the discussion, alongside ready-made agendas with a facilitator timer, icebreakers, a meeting cost clock and follow-up. It is free and made by AIYUN, and nobody needs to sign up.

Common problems and fixes

Lean Coffee as a retrospective

For teams who are tired of fixed retro formats, Lean Coffee makes a good change. Use a prompt like "What should we talk about from this sprint?", keep the usual Prime Directive opening, and make sure you finish with one to three actions. If you also want themed boards and action tracking, RetroLift has a Lean Coffee template that runs a similar flow.

Frequently asked questions

How many topics fit in a Lean Coffee?

In 30 minutes of discussion, a group typically covers three or four topics, using timeboxes of about eight minutes with shorter extensions. That is by design: you discuss the group's top-voted topics, not everything on the board. Topics you do not reach can carry over to the next session.

How long should a Lean Coffee session be?

Between 30 and 90 minutes. Under 30 you will cover only one or two topics; over 90, energy drops.

What timebox should you use for Lean Coffee?

Start with five to eight minutes per topic and extensions of about half that. Adjust once you see how your group works.

Does Lean Coffee need coffee?

No. The name comes from the original group meeting in a café. Tea, water or nothing works fine, and the format runs just as well online, for example in the Lean Coffee mode of MeetingLift, a free tool by AIYUN.

Disclosure: I build MeetingLift, RetroLift and the other free Lift Apps tools mentioned here.