The kitten in the black box of complexity.

At the company I'm a co-founder of, DigiFabster, we listen to our customers. Plus we want to offer maximum value at a minimal price. When you cross those two principles you get a lot of flexibility, but also a lot of complexity, because we share all executed customer requests with all other customers.?

It’s like the old story of the guy who wanted to be awarded with one grain of rice on the first left top field of a chess board, two grains on the second left top field, 4 on the third, etc. The awarder, a king who had played hooky during geometry lessons in school, gladly agreed. That was a mistake. Given that a chess board is 8x8, the number of grains on the last field would be 2 to the power of 64, weighing in total 1,199,038,364,791 metric tons. To put that in perspective, and to show another nice picture:

Now look at the pricing parameters in a “simple” technology like MJF in Digifabster: At the request of our users we kept adding more and more parameters until now there are at least 38, not counting discounts, premium pricing for fast delivery, or pricing depending on layer thickness and infill. If we just consider not using (entering 0) OR using (entering >0) this or that parameter, we get a total of 2 to the power of? 38 (274,877,906,944) different pricing strategies to choose from. The next customer request will bring that to 2 x 274,877,906,944=549,755,813,888.?

The chances of a customer finding the right strategy for his or her shop by trial and error are proportionally small. We realized that, so we introduced the price matching tool. It works with a limited set of parameters (17), those we see most used for the technology, and tries to fit a data set, consisting of 10 models and their geometry, with prices for 1 piece and 10 pieces, with those 17 parameters. So from a possible 2^17 (i.e. 131,072) strategies, it chooses the one best fitting the data.

This is an oversimplification, because the values entered in each field are not 0 OR 1, but 0 to virtually infinity (∞^17),? however, exploring that nuance here will not improve readability. The point is that the price matching tool does an unimaginable amount of trial-and-error work within seconds and comes up with a solution. That solution is not perfect, and it’s not meant to be. It’s just a starting point for further tweaking.?

Digifabster has a price-tweaking tool. In that tool, the user can see what the set of parameters, generated by the price-matching tool, does for every model he loads into it. It’s not perfect, and there are plans to develop it further. One of the known limitations is that you only see the changes, made in the price tweaker, take its effect on the price of one model at a time. For now, we try to mitigate this by offering Assisted Setups. This means the (prospective) customer can upload the same set of models and prices as required for the price matching tool into a form on the Material page in question. One of our application engineers will load those data in the same price-matching tool and will share the results in a spreadsheet that allows the user to see the effects of changes in input on the prices of the 10 models at one glance. Given that it’s a spreadsheet, the user can change the type of diagram in a way that fits his or her preferred type of visualization.

Given that the spreadsheet can only reflect 131,072 out of a possible 274,877,906,944 pricing strategies, it is certainly possible to tweak further, by going back to the price tweaking tool.?

Another limitation of the price tweaking tool is its layout. It’s just a web page, and web pages use frameworks, and those frameworks limit how data can be easily presented.?

Cuddly kittens are one thing, but this is tabulated data, interacting in real-time with both the user on the client side and the calculation engine on the server side. In two words, it’s an ugly worm, not a kitten, and it demands scrolling, not cuddling:

The good news is that for the CNC side of the house, we’ve finally broken out of the limitations of the framework, which makes for a much easier understanding, even if the calculation process itself is many times more complicated (imagine a really big number here to quantity the possible pricing strategies in CNC). The CNC tweaker will look like this, more like a crouching tiger:

Only a small part of the calculation is shown on the left, but the effect of changes in any of the fields is immediately visible on the right. (Thanks Alexander Geim !)

Once we’ve got the new CNC tweaker tested online and got our first feedback, we’ll upgrade the AM tweaker in the same way.

Next up: calculating and visualizing for up to 100 models at a time. So still a bit of work to do, but that's what keeps us young and playful.