Presentation: More Reliable Delivery with Monte Carlo & Mapping

Conal: I’m a product manager at Handshake, a B2B commerce platform that facilitates sales between manufacturers/distributors and their customers. I’m trying to make sure that we’re delivering as much value to our customers as we can. Like everywhere, our resources are limited, so mostly I’m focused on what outcomes can we influence the most, and what solutions can we design to help our customers.

Conal: I’ve gotten a lot out of the New York City lean/agile community, especially from people like Jonathan Hansen. I’m excited to have the opportunity to give back, and help other practitioners improve the way they work.

I’ve also rarely talked to someone that’s happy with their planning process, but everyone seems to try the same things when there is a more useful, less painful approach that relies on sound statistical methodologies and can be fairly automated. This seems like it’s the perfect topic for the QCon audience.

Conal: I’m aiming this to experienced developers who have tried a lot of different planning strategies, and aren’t quite thrilled with anything. They’re more senior, and should be expected to influence the processes at their company.

Conal: A Monte Carlo Simulation is essentially just looking at past data to predict future results.

There's a couple of different statistical techniques that go into it. Essentially, you're going to get a range of all the cycle times (or the takt times for stories you've delivered before). You're going to then extrapolate those to a very large sample size.

We're talking about usually 10,000 samples of cycle times (or takt times). Then you're going to input the size of your project to figure out roughly how long it's going to take you to complete the stories based on a historical random sample.

The way this works is that by looking at the past data and then extracting it, we're doing something called bootstrapping. Where we're taking something that might have very high variability and by extrapolating that into a larger sample size, we're going to key in on what the actual true values are. Then we're going to take that and distribute that into our project to make sure that we have a good understanding about when the work's going to be delivered later on.

It's a little bit different from traditional project management. With Monte Carlo Simulation what comes out of it is a range of probabilistic outcomes. So rather than saying that we're going to be done with this on July 4th. We're now going to say we have a 50 percent chance of finishing this by June 30th, and we have a 75 percent chance by July 4th.

Conal: I have two main goals: first, I want people to buy in to the fact that this is the right way to build a backlog and forecast your projects; second, I want to provide enough information and references to enable them to implement Monte Carlo simulations with story mapping effectively. This is a deep topic, but if I can pique interest enough for attendees to try to experiment with this forecasting on their own, I’ll call it a success.

Conal: As a product manager, I’m always striving for real continuous improvement in how we deliver value to our customers. We have to be willing to experiment, measure, and learn. With Agile being so well-established, it’s easy to fall into the same patterns that are common in many organizations, even if they’re not productive. To differentiate your product, company, and culture, you have to be willing to try new things and find what works specifically for your team.