Can you do simple maths?

Have you ever seen an assembly line in a factory hall with your own eyes where parts for a vehicle are being built? I have. My first summer jobs were directly near the assembly line.

Now imagine the factory hall and each station on the assembly line completely cluttered with tools and materials so that nothing can be found. Or imagine there is never enough material or the right tools available, so the worker on the line is constantly waiting. Or imagine managers with no knowledge of assembly or manufacturing constantly interrupting the worker and assigning random tasks with no real purpose.

It’s difficult to picture because it’s frankly nonsensical. And yet some organizations do exactly this when it comes to data and AI topics. Not out of bad intent. They do it because they do not know any better.

What is missing is a basic understanding of lean principles.

There is a lot to learn, so let us start there.

Take the level above the teams, the level that decides which initiatives and products are taken on in the first place. Sometimes it is even several levels above a team making these decisions. Sometimes it is multiple independent organizational units that end up bombarding teams with too many things at the same time.

Now imagine these organizational units actually measured the following. How many initiatives, projects, products, or tasks are we placing simultaneously into backlogs or directly into teams on an ad hoc basis? How many of those are completed within a certain period of time, meaning running stably in production?

I once measured exactly that. Over a period of one and a half years, I measured how many epics a program consisting of seven teams planned for each three month period and how many of those were completed at the end of those three months, meaning fully rolled out to production. They consistently planned more than 50 epics and consistently completed exactly 21 epics. Each team, on average, rolled out three product increments within three months. I created a wiki page and visualized the data and measurements. Only then did the leadership team begin to understand. Why plan more than twice as much when delivery has been stable for one and a half years?

That program was exemplary. We all know there are many data and AI organizations that can only dream of that level of delivery.

There is no shortcut. You have to start somewhere. My advice is to start by doing less at the same time. Kanban is pure mathematics.

Little’s Law states:

Lead Time = Work In Progress / Throughput

A simple example. Using Little’s Law (L=λ · W) (so (W=L/λ)):

• Example 1: If you have 23 items in progress and you finish 3 items per year, the average lead time is approx 7.7 years per item.
• Example 2: If you have 3 items in progress and you finish 10 items per year, the average lead time is 0.3 years, i.e. 3.6 months per item.

The math is straightforward. Decide that work should be finished one item after another. Reduce work in progress. Protect teams from unnecessary interruptions so they can focus and complete what they are already doing.