In almost any system, there is an easily overlooked weakness that can bring down the entire operation. On the Death Star in Star Wars, it was a thermal exhaust port. A more earthly example is the failure of small, rural banks causing the Great Depression.
It's important for OEMs to look for the potential exhaust port failures in its buying ecosystem well. If a supply chain relies on a particular screw to ship 50 of its products, and that screw is delayed for 10 days, how many of those days would it take to find out all of the delays were caused by the same problem? How does a recall of single component from one region impact what you have on the shelf across your customer base? How much would it cost in lost receipts?
This is the more slippery cousin to the “single point of failure” – a component to your system that causes a general collapse when it fails. Those are easier to spot in a cursory top-down analysis and address through redundancies or stockpiling. These elusive creatures live in the acceptable risk for each analysis, but because they quietly impact so many aspects of your business, the compounded effect is too great to bear. Finding them takes a bottom-up approach, tracking each part through the system and finding the cost of its failure.
This challenge might look like finding six matching needles in nine haystacks, but data analysis has progressed significantly in recent years, and one of the most important benefits is the ability to create cross-organizational views on purchasing. Procurement teams can determine if there are redundant relationships with any vendor, spot trends in contracts to better anticipate costs and even find high-leverage parts that carry risks disproportionate to their value.
Why didn't we know this already?
High performance storage and data processing have been capable of this kind of analysis for some time, but the data itself has been holding us back. Most companies keep data in several different formats that don't easily integrate, and the work required to bring that data together for each analysis wasn't worth the fleeting knowledge it provided.
Even companies that try to keep data related to a specific domain, such as procurement and purchasing, to one format find that they may need supplemental data that doesn't comply to the standard format, or that one division didn't follow the plan, or an acquired company chose a different path. Further, as companies purchase more complex, finished components, data on the comprised materials increasingly comes from the vendors themselves in whatever format they use. Data wants to be diverse, to paraphrase Stuart Brand. Data unification has emerged to handle data diversity and deliver the benefits of perfectly aligned data without policing formats.
Specifically regarding this challenge, data unification merges information on all inbound products regardless of where they come from and what they go into. It can even trace components of those inbound products. If 14 vendors use the same semiconductor in their design, or 21 vendors buy copper from the same source, you'll know, and have strategies for dealing with the implications.
Data that works together needs people that work together
It isn't difficult to pinpoint the data that matters in this situation. It may take new levels of cooperation among partners, however. Vendors or the IT department will need to provide information they might not typically share. Ultimately, they need to recognize their stake in the health of your supply chain. If they wait for a crisis to share this data, it guarantees delays should problems arise. The data that protects you against a supply chain disaster also immediately confirms your safety when you dodge a close call.
The next step is rationalizing the data, or creating the task-oriented view of data that lives in all of these sources. Any analyst can tell you this is usually the hardest part of the job. Data preparation is a monstrous challenge in any analysis requiring data in multiple formats. It involves manually transforming fields in static databases so it matches the format of another until it's all combined into one file. This manual work has been replaced.
Now companies use automation, guided by machine learning and expert sourcing, to whittle down the manual work. Further, the data preparation tools leave source data in tact, saving the rules for rationalization so the same process can be repeated with the same sources as they grow and change over time, and new sources can be folded in as they pop up. Any time a report is needed, the analyst can use the most current data without restarting all of the work of building a combined dataset.
The combination of machine learning – by which the tools learn how to merge data when used to merge data – and expert sourcing – bringing in subject matter experts for straightforward questions no one else can answer – is by far the most effective route to eliminate the manual work and errors that beset data preparation by people who don't really know what the data means. Data analysts can stay expert in data analysis and get the benefits of battle-earned domain intelligence.
This bottom-up approach to data analysis is a significant change. By connecting sources where they sit, in whatever format they happen to exist, and using domain experts to help connect the dots, analysts can tackle more and new kinds of challenges. Among them, it's possible to proactively address problem of hidden vulnerabilities.
When the data on all incoming products is complete and unified, it's no work at all to perform an annual review of exposure to vulnerabilities related to a single part or ingredient with all of the most current information. If at any point it is prudent to confirm, for example, that a semiconductor is traceable to the vendor and not a counterfeit fake, you can have that information company-wide before the vendor calls.
As much as we root for Luke Skywalkerand the rebel alliance in Star Wars, imagine how the story would have ended if the empire put a grate over that thermal exhaust port. Today, we can find the soft spots in our systems without a crisis finding them for us.