Steve Fenton

A practical right-to-left example

The phrase ‘right-to-left’ is pretty well embedded in software development teams. There are some great works on Kanban and flow that discuss it in detail. Mike Burrows has a great book that is even called ‘Right to Left’. This article doesn’t replace or repeat all that good stuff, it’s just an example of right-to-left based on an anonymised screenshot of a task board that I have in front of me at this moment. I’m going to try and describe what arcs across my brain when I open up this map of the work.

A task board. From the left there are columns named new, development, pull requests, dev test, deployed to test, testing, ready to release

Board description

The board is an accurate description of a team’s process. One of the most important principles for the design of a Kanban board is that it reflects reality, no matter whether that reality is ideal or not. The board only changes if the reality changes first. It’s like a map, rather than a blueprint (we only add that new super-campus onto the map after it has been built). The columns on the board highlight work steps, but also hand-overs and waits. Adding columns to represent waiting states is super-important as this is where cycle-times find an all-you-can-eat buffet of delay. Ideally, I’d like my wait columns to be a different colour to signify that things shouldn’t remain there.

As is often the case, the wait steps on this board aren’t pure waits. They are “a long wait followed by a small task”. For example, if you aren’t aggressively attacking waits, it is common for a PR to wait longer than it takes to review the PR. On this board, the only pure wait step is “Deployed to Test” as the moment work starts, it moves out of the column and into “Testing”. Perhaps the board should be re-designed to separate the wait from the work for the “PR” and “Ready to Release” steps (Awaiting PR and PR, Ready to Release and Releasing). That’s something the team should decide; but in particular I find long wait times for PRs to be particularly troubling as I’ve read Nicole Forsgren’s amazing work (see Accelerate)!

There are swim-lanes on this board that represent software products. These three lanes have independent releasing and can be worked on by different teams. In other organisations these might be separate boards.

Observations

My initial observation on the board is that most of the work is at a wait step. “Ready to Release” is a wait step. “Deployed to Test” is a wait step. “PR” is a wait step. There are signs that there is some left-to-right thinking present as despite all this waiting work, new work is in development.

The board has a shape that I recognise as one that often precedes a problematic release. If we keep adding new work in, or allow the work to get out of the correct order, we end up with a card blocking the release. For example, we have software versions 1, 2, and 3 currently in our pipeline, with two more versions about to land from pull requests, and another version coming from development. If we find a problem with the version about to be tested, we clog up the system and have the task of getting “all in flight cards into a releasable state”, which is far more work and co-ordination than getting a single card released.

Let’s fix it by thinking right-to-left.

What to do next

Focussing on the first software product (the top swim lane), there are two cards ready to release. We need to check that the software version for these two cards does not include any changes from cards on the left. The three cards in the development and pull request columns aren’t yet in a software version, so we put them hold as we don’t want to increase the inter-relationships further. That means anyone who was about to work on those is now available to work on the cards on the right.

If the card awaiting testing has a lower version that the cards we want to deploy, we have to focus on that card. This can only happen when we aren’t thinking right-to-left as we’ve allowed cards from the left to take precedence and overtake this card. We can’t deploy them without having tested the earlier version, so we shouldn’t have let them skip ahead, where they can end up blocked by the earlier card. By thinking right-to-left in the future we avoid piling up work and getting in a mess.

If the cards ready for release are versions 1 and 2, and the card ready to test is version 3… we hit that button and deploy the software. We update the board to show the cards are deployed.

The task board has been updated with the changes

Now we repeat the process, only this time the right-most card is the one waiting to be tested. That’s the next thing to do (and as it moves to the right, you’ll notice that we continue to focus on this card until it is off the board, because it always going to be the right-most card now we are thinking right-to-left). So, we focus on getting the card tested and then we focus on getting the card live.

Each time we do this, we not only finish the card that is closest to completion, we are also clearing the path for any work that follows.

The updated task board shows the potential for work to flow

If we keep following this pattern, we naturally land on single-piece flow as a result of prioritising “finishing” over “starting” work. If you are working on a stack-ranked list of options, this benefits your “most important card” as no other work slows down the cycle time of the important work. This benefit is repeated as each “most important card” gets pulled from the to-do list.

Summary

If you focus on the cards on the right-hand side of the board, you’ll find that you naturally avoid hitting WIP limits, shorten your cycle times (especially by eliminating waits), and never end up with version-wrapping (where a card in version 1 has an issue, fixed in version 5 – but not releasable because the card in version 4 isn’t fixed until version 6… and so on).

Written by Steve Fenton on