Reading time: 5 minutes
UPDATE: If you read the early-access preview of this article, you’ll definitely want to download the newest infographic and read this updated blog.
In 2015 during the early days of CX3 Marketing, we created an infographic that explored the state of automotive recalls. Starting with raw data that the US government maintains on recalls at the Office of Defect Investigation (ODI), we extracted some intriguing highlights with some Python scripts and home-grown machine learning techniques.
It’s now five years later so we felt it was the perfect time to revisit the data. We published an initial update under our new Third Law brand and within minutes our friend Roger Lanctot from Strategy Analytics called us up. He pointed out that Sibros just published their own software recall stats that differed substantially from ours and asked, “Why aren’t we getting the same results?”.
Revisiting the data
We reviewed our data and discovered a few recall campaigns that had been mis-categorized by the automated algorithm. We fixed those and re-checked all of the analysis by hand. In doing that, we uncovered a more fundamental question, which is what should we consider as a software problem? For example, do any or all of these count?
- Software that corrects novel problems discovered after deployment
- Software that works around a mechanical problem
- Software that leads to mechanical failure
- Software that is installed as part of a system upgrade
- Software that is being brought up to the latest release
Whether or not you analyze a particular recall as software-related is somewhat subjective. Based on the number of software related recalls Sibros is reporting, we believe that they are using the broadest definition of a software recall, which might be categorized as “any recall that involves a software update of any type”.
Software Fixable
Our approach is different and that is to look at software recalls as a proxy for over-the-air (OTA) viability. In other words, we’re looking for software fixable problems.
That means that we won’t consider it a “software recall” if the car needs to be brought in to the dealer for an inspection and/or part replacement, even if an ECU update might be part of this. Likewise, even a software problem can have damaged physical parts – like an improper engine calibration frying the transmission, or a sensor mis-read overheating the catalytic converter – since the recall requires mechanical intervention, it’s not software fixable.
Why not consider software failures of this last category as software-related? It depends on what you’re trying to determine by tallying up “software-related” recalls. Software is an integral part of the automotive system, and these type of failures are not software bugs per se; they could not be detected by static code analysis or any software-only discipline. A failure of this type is more indicative of a failure to thoroughly test under all conditions. And most fundamentally, these type of problems won’t result in a recall that could be avoided by an OTA reflash.
Our approach is different and that is to look at software recalls as a proxy for over-the-air (OTA) viability. In other words, we’re looking for software fixable problems.
Software fixing the car
Pouring through thousands of recall records also made us realize something else that we were missing, particularly with our view that OTA could change the recall landscape – identifying software fixes that correct hardware problems. While perhaps most famously first done publicly by the folks at Tesla who altered their cars’ undercarriage clearance via a software update, it’s clear that other manufacturers are following suit with software fixes to hardware problems. While most OEM software fixes currently have to be brought into the dealer, they’re being addressed by just an ECU reflash rather than by more intrusive repairs.
We’ve added a new category “Software corrected” to our infographic to count these separately, since it seems entirely unfair to label recalls of this sort as software issues when they’re actually software coming to the rescue. Examples of software corrected recalls from 2019:
- Volvo clutch assembly internal failure
- GM brake vacuum pump losing pressure over time
- Volkswagen transmission hairline cracks
Have software recalls stabilized?
As can be expected, some data points have remained the same, while others have changed. We’ve also added a new section:
- Takata is still taking the cake for dominating the total number of recalls, even seven years after their defective airbags were first recognized. Needless to say, they’ve since filed for bankruptcy.
- For a while it looked like we were on trend for software recalls to continue exponentially. If we only look at the software defects and not software corrected issues, we’re at 5.8 million this year, which might be good considering that we’ve been on track for much more. (For example, 2016 was a record-breaking year with 9.6M software defect recalls.)
- We’ve included a new stat we’re calling the Recall Ratio, which measures an automaker’s recalls against their sales over the same period. This allows us to compare relative “quality” of larger and smaller players, and it shows some surprising results under both the best and worst columns.
Massive cost may be unavoidable
You’ll see in the infographic that automotive recalls cost the industry over $15 billion USD last year. One may be tempted to point the finger at quality issues in the auto industry. However, automaker quality is overall quite good. We believe – especially after reading the endless recall reports – that this large cost is a nearly unavoidable result of building hugely complex products with broad and diverse supply chains all the while ensuring those products meet many regulations, customer demands, and harsh environments for over ten years.
The fundamental reason for Tesla’s performance
Now that Tesla is building scale volumes beyond the boutique, it’s very instructive to look at their performance as the vanguard of the EV disruptors.
Tesla is more than holding its own in the build quality department. It’s clear they learned hard lessons from the Model 3. They’re also able to avoid a decent number of recalls due to OTA; something that we’ve hinted at here being a potential future strategy for other OEMs. Perhaps some measure of quality is due to their reimagined approach to car production, where a blank slate approach gives some advantages in supply chain and integration that traditional OEMs can’t utilize easily.
Since manufacture, assembly, and wear are 50 percent of the reasons behind your average recall, eliminating a large number of parts gives you a tremendous advantage.
However, we believe that the most fundamental reason for Tesla’s excellent recall performance is the dramatic reduction of mechanical complexity going from ICE to EV. Since manufacture, assembly, and wear are 50 percent of the reasons behind your average automotive recall, eliminating a large number of parts gives you a tremendous advantage. Simply put, fewer parts means fewer problems, and much fewer parts means much fewer problems.
What does the future hold?
It’s hard to make predictions so most people and companies shy away from it. But this doesn’t scare us. Here’s our thoughts on what may await us in 2020 and beyond:
- Although software-fixable recalls haven’t dominated recalls as much as we were expecting, they still make up a significant chunk. The increasing penetration of OTA solutions across all automakers should keep these numbers in check, since new firmware downloads can erase software-driven problems and obviate the need for a recall. Despite every vehicle’s escalating reliance on software, we expect next year’s software-related recalls to be around the same – hovering around ten million.
- As more OEMs convert progressively more of their fleets to EVs, we expect automotive recalls to continue to come down since they’ll be leveraging the simplicity of an EV engine/design/architecture. While EV fleet proportions will positively impact recall reduction, it pays to remember that recalls are issued for all vehicles on the road, so this “calming” effect will take a while to start showing up in the bigger manufacturers who have significant fleets of already deployed vehicles that may experience recall issues.
- A flotilla of new EV challengers will be bringing their new models on line soon. While Tesla has been making great strides on keeping their recalls down, expect that these new OEMs will initially have very poor recall stats as they relearn the lessons Tesla (and every OEM) has experienced. EV OEMs (such as Lucid and Rivian) that have significantly poached staff from existing OEMs will fare better than the purely Silicon Valley or Chinese VC-driven efforts to keep down their recall spike as they enter the market. But every new entrant will need to piece together for themselves how to build quality in production volumes.
About the author
