By compilation I refer to the art of crafting compilers, that is, computer programs that translate source code into another language. Few people outside the ranks of programmers know what a compiler does, and few people within the ranks of programmers know how a compiler is designed. At first, compilation concerns appear distant (to say the least) to supply chain concerns. Yet, nowadays, at Lokad, it’s compilation stuff that keeps saving the day; one supply chain project after another.
Shameless plug: software engineers with compilation skills don’t grow on trees, and we are hiring. Want to work on stuff that matters? Well, the next time your plane is late because a part was missing, or the next time the drug you seek is out of stock, just remember that you could have made a difference by joining Lokad :-)
Supply chains are complex, maddeningly complex. Globalization has multiplied sourcing opportunities, but delays are longer and more erratic than ever. Sales channel are being multiplied too: there are physical stores, online stores, marketplaces, resellers, wholesalers, … And now, thanks to Amazon, everyone, everywhere expects everything to be ordered and received overnight. Supply chain expectations are higher than ever.
Approaching supply chain problems with anything less than the full expressiveness of a programming language does not work. Just like Lego programming is not going happen, supply chain challenges won’t fit into checkboxes and dropdowns. This does not prevent software vendors from trying, mind you. Solutions that include more than 1000 tables, each table hovering at around 100 fields on average, are all too common. And while the client company is only using about 1% of the solution’s feature area, they still have to cope with its pervasive complexity.
Compilation saves the day because it provides a huge body of knowledge and know-how when it comes to crafting high-quality abstractions intended as power tools for solving statistical and combinatorial problems (and much more actually). And most supply chain challenges happen to be precisely statistical and combinatorial. For example, at Lokad, by introducing an algebra of distributions, we managed to “crack down” on complicated lead time problems which were resisting our more direct approaches through packaged software features.
What makes language features different from, say, the usual app features (wysiwyg), is that language features are much less sensitive to the specificities of a given challenge than their app features counterparts. For example, let’s consider a situation where your stock-out detection logic backfires in the specific case of ultra-seasonal products. If the feature is delivered through a language construct, then you can always narrow down the data scope until the feature works exactly where it’s intended to do so; possibly dynamically adjusting the scope through an ad-hoc numerical analysis. In contrast, with an app feature, you’re stuck with the filtering options that have been built into this feature. App features are a good fit only if your problems are narrow and well-defined, which is actually very unlike supply chain optimization.
In supply chain, programmability shines because:
- Problems are both highly numerical and very structured
- Supply chains are modular and this modularity needs to be leveraged
- The number of variables is significant but not overwhelming
- Fitting the precise shape of the problems is critical
It is slightly amusing to see how many software vendors tend to gradually re-invent programmability. As the user interface grows in depth and complexity, with the possibility to add filters, options, pre-process or post-process-hooks, templated alerts, KPI monitors, the user interface gradually becomes a programmable thing, and reaches the point where only a programmer can actually make sense of it (precisely thanks to his or her pre-existing programming skills). Programmable yes, but in a highly convoluted way.
Compilation is the art of amplifying engineering skills: one has to craft abstractions and language constructs that streamline thinking the resolution of problems. As Brian Kernighan famously wrote: Everyone knows that debugging is twice as hard as writing a program in the first place. So if you’re as clever as you can be when you write it, how will you ever debug it? The same logic applies to supply chain optimization, because it’s essentially the same thing as writing code. Well, at Lokad, it literally is the same thing.
Conventional IT wisdom states that one should automate the easy parts first, leaving human experts to cope with the more complex elements. Yet, in supply chain, this approach backfires badly every single time. The most complex parts of supply chain are nearly always the most costly ones, the ones that urgently need attention. The easy parts can take care of themselves through min/max inventory or Kanban. Just like you wouldn’t build software for autonomous cars by refining software for automatic train operations, you can’t tackle difficult supply chain problems by refining software initially designed to resolve simple challenges.
Naturally, compilation alone isn’t sufficient to cope with supply chain challenges. Machine learning, big data processing and a sizable amount of human skills are worth mentioning as well. However, in all cases, having carefully crafting high-quality abstractions helps considerably. Machine learning is vastly simpler when input data is well-prepared. Big data processing is also much more straightforward when computations lend themselves easily to a high degree of parallelization.