Every project manager, product manager, and producer gets asked the same daunting question, usually by someone senior and usually weeks later than would have been useful:
"Are we going to hit the date?"
In my twenty-five years of shipping games and helping product teams run their tools, I've heard a lot of stressed-out answers to that question, and most were guesses based on countless hours of up-front estimating and a burndown chart. And they were wrong.
Burndowns are useful because they're simple enough for a stakeholder to look at and get the gist. They're usually based on estimates that, at best, are just a guess and, at worst, a team trying to build in its own buffer. A burndown that combines both scope increases and completed work on a single line can be very misleading. With a steady team, the line will probably end up in the right place. Most teams, however, don't complete work at a steady day-to-day rate. The burndown line gives you a date with zero confidence attached. In reality, the chance of hitting that date is probably less than 50% (particularly if a large proportion of the effort is at the end). The first 80% is easy. The last 20% is hard.
I set up a demo board in Favro to show what that means in practice (see the attached video). It's a combat team working toward a vertical slice milestone due December 7, with 35 cards left. Favro's burndown predicts about five cards still open on December 7. So the obvious answer to “Are we going to hit the date?” is “No, but if we cut scope by 5 cards, then yes, we’ll hit the milestone.” That would be naive and wrong.
A Monte Carlo simulation is of course named after the casino in Monaco. It was developed at Los Alamos in the 1940s and is used to predict highly complex and uncertain outcomes, like the weather and financial markets. A burndown chart draws one line forward, and a control chart shows only what's already happened; both rely on averages, which hide massive tail-end risks. A Monte Carlo forecast, by contrast, plays out thousands of possible futures and counts how many come in on time.
Most PMs and Producers either haven’t heard of Monte Carlo probabilistic forecasting or they haven’t tried it for three main reasons:
Luckily, Favro automatically captures that flow data when your teams use Favro boards.
On the demo board, the AI workflow I created uses Monte Carlo forecasting to replay the team's own Favro history. Over the last ten working days, the team finished anywhere from zero to four cards a day. The forecast builds a possible future by randomly drawing from those real days, one per working day, until the remaining cards are done. It does that 10,000 times and counts how many futures finish by December 7. It also runs a second version where new cards keep arriving at the current pace, because they do on real projects.
The answer comes back as 80% likely if scope holds, and 37% if new work keeps arriving. The team's pace isn’t the problem. The risk is scope.
I tested the burndown's advice before touching the board. I cut five cards, and the forecast says 96% if nothing else gets added, and 67% if new work keeps coming at the current pace. However, one time in three, cutting only five cards still misses the date.
That’s because the burndown's projection assumes every day ahead is an average day. Real days aren't average days. This team's days ranged from adding one card per day to finishing four. In a third of the simulated futures, the team misses December 7th.
A one-in-three chance isn't very good, especially when a milestone or release is at risk.
Then I cut nine cards, and the odds go to >99% if no additional work is added and 86% if new work keeps arriving. That's a probability you can commit to. The burndown told me to cut to the bare minimum scope. The AI workflow forecast told me how much to cut to be certain.
The AI workflow produces a report that includes a burn-up, which still won’t give you the odds alone, but is far better than a single-line burndown. The burn-up separates scope from completion: one line rises as work gets finished, a second line above it tracks total scope. The simulation's likely range is shaded around the finished line. You hit the milestone where the lines meet before the date. Scope changes are made explicit, so a conversation about scope doesn't get mistaken for a conversation about velocity.
The forecast counts cards and ignores estimates (points, days, hours, etc.). That sounds crude until you notice that the size differences between cards are already baked into the history: a day when the team closed one big card shows up as a 1, and a day full of small fixes shows up as a 3. It works the same way a store manager can staff checkout by counting customers per hour without counting the items in each cart.
If the team goes for the quick wins first and saves the big one for last, then the forecast will be very optimistic. The fix is to split anything that's obviously bigger than a normal card before you trust the numbers. Even a half-baked attempt at “right sizing” cards will give you far better forecast results than estimates and burndowns.
I've turned all of that into a free AI workflow skill for Favro. You just ask in plain English, "Are we going to hit our Dec 7 milestone on the Combat Pod board?" and it reads the cards via Favro's MCP connector, works out when each one finishes, runs the simulation, and produces a forecast. The forecast includes the odds, a burn-up with the likely range, the 50%, 85%, and 95% dates for both scope scenarios, and a short list of what could change the answer (e.g. blocked cards, a short history, holidays etc).
After one run, you can ask what-if questions before you make any changes (e.g., "What if we cut 5 cards?"), and it will answer without altering anything in Favro. Once you're happy with the answer, it will ask before posting the forecast as a comment on your milestone card with the burn-up attached, so the team can see the forecast where they’re already working. Run it every week, and the milestone card becomes the milestone's forecast history.
The skill has its limits. It looks at one board, and most milestones require multiple teams to land together. Each team has its own burndown with its own date, and nothing on any one of those individual burndowns tells you the probability that they'll all hit the date. If four teams are each 90% likely to hit their date, then the milestone is only about 66% likely, as every one of them has to hit it. A forecast across boards answers "Are we going to hit the date?" and gives you one number for any-size team. That's what I'm building next as a secure, hosted, continuously running agent.
So now when you get the dreaded 'date' question, you'll have an answer you can be confident in. And if the answer is bad news, you'll have the evidence to make the right decision going forward. That's a far better conversation to have than one based on a gut feeling and a burndown.
Download the milestone forecast skill, free, with a short guide to using it with your AI assistant of choice.
Check all of this out in practice in this video.
