Lead Time Predictability: Is your Team Predictable?
There are a few criteria you can measure predictability around, but for the context of this blog, we’re going to focus solely on Lead Time. This, however, can also be applied to many other measures such as Throughput, Flow Efficiency, or even Value Demand.
What do we mean when we talk about Predictability?
Something worth stressing is that being highly predictable doesn’t necessarily equate to ‘fast’, or that you are meeting your Service Level Expectations (SLE). It also doesn’t necessarily mean you’re in a high state of ‘flow’. You can be predictably fast, but you can also be predictably slow.
In probability theory and statistics, when looking at lead time distribution charts, you'll be able to see that the Weibull distribution is better suited for knowledge work (and not a bell curve that most people are familiar with). When analysing the distribution on a histogram, there is the concept of fat-tailed distribution versus thin-tailed distribution. We won’t explore this in great detail, but you can surmise them as follows:
- Fat-tailed distribution results in high variance in your lead time, which we have translated as low predictability.
- Thin-tailed distribution results in low variance in your lead time, which we have translated as high predictability.
This is can be visualised on the Lead Time Histogram:
Why is Predictability Important?
Teams that have low predictability potentially have high impact with long delays. Take the following team’s distribution as an example:
It shows that the team is able to complete half its work within 5 days (Median). The majority of its work (85th %ile) is completed within two weeks. However, if the team has a delay, it results in a high impact (i.e. the delays can be quite long). The customer could potentially be waiting for up to almost three months (Tail). It could be argued that this team can be considered fast, but has low predictability.
Take this second team as another example:
This team completed half its work within a two-week sprint (Median -14 days). The majority of its work is completed within two sprints (85th %ile - 23 days). Even if it encounters a delay, the customer would typically wait for about a month (Tail - 30 days). If you compare it with the previous team, you could classify this team as slow but with high predictability.
Let’s now take a scenario of a Program of Work with several teams. All building features, working towards a launch, which requires some coordination with a marketing campaign team. In this instance, you might want your teams to have high predictability, so there is a high level of confidence that you will be able to coordinate your launch activities around a go live date. Conversely, if your teams were fast, but have low predictability, you would have lower confidence in being able to coordinate around that go live date. Depending on your context, speed-to-value might be good to have, but predictability might be of a higher importance.
Another reason why high predictability is desired is the fact that if you were utilising predictive forecasting techniques, such as Monte Carlo simulation, there will be less variance in your forecast, giving you higher confidence. If your teams have low predictability, you’ll find higher variance in your forecast, which would result in lower confidence in your ability to hit the target date.
How Flomatika Calculates Predictability
In this section, we will talk in more detail about how Predictability is calculated in Flomatika.
In probability distribution, there are a few methods to calculate fat or thin-tailed distribution. At Flomatika, we’ve decided to adopt the approach that measures the ratio between Median (50th %tile) and Tail (98th %tile). If it’s greater than 5.6, it’s considered Fat-Tailed (Low Predictability). If it’s less than 5.6, it’s considered Thin-Tailed (High Predictability).
Here are some examples of some calculations:
Median: 96 days
85th %tile: 213 days
Tail: 375 days
The ratio between 96 days (50th %tile - Median) and 375 days (98th %tile - Tail) is less than 5.6, so they have a Thin-Tailed Distribution and are considered to have high predictability.
Median: 5 days
85th %tile: 10 days
Tail: 375 days
The ratio between 10 days and 375 days is greater than 5.6, so this team has a Fat-Tailed Distribution and are therefore considered to have low predictability.
So for the teams that currently have low predictability but want to shift to high predictability, what can they do? The answer is to ‘trim the tail’ on their Lead Time Histogram. Eliminate the items with a long Lead Time.
To assist teams in achieving this, we at Flomatika’s have introduced a ‘Predictability’ line on a number of our charts:
- Lead Time Histogram
- Lead Time Scatterplot
- WIP Age Histogram
- WIP Age Scatterplot
On the Lead Time charts, teams can see which items have classified them as low predictability. (I.E. Items to the right of the predictability line on the histogram. Items above the predictability line on the scatterplot.)
For teams aiming to switch from low to high predictability, they need to eliminate the items that are past the predictability line on the WIP charts. (Again to the right of the predictability line on the histogram and above the predictability line on the scatterplot.)
Some teams might find these concepts new, or it can take some time to digest. At Flomatika we’re looking to simplify these and get teams closer to what is actionable. We're always looking at ways to bring actionable insights to teams. To summarise:
- High predictability in lead time doesn’t equate to fast delivery.
- High predictability teams give you higher confidence on when they will be able to deliver, which also makes probabilistic forecasting techniques more useful for these teams.
- Teams can increase their predictability by ‘trimming the tail’ on their lead time histogram.