Blog

How Autonomy Changes Everything

Vehicle telemetry data promises underwriters the ability to better quantify risk and use real-time alerts and feedback to promote safe driving. However, Usage Based Insurance (UBI) systems are still in their infancy, typically relying on a few simple inputs such as deceleration as a proxy for safe driving while largely ignoring context. Perhaps the driver braked aggressively to avoid an unsecured box that took flight from the bed of a pickup truck, or a child that ran into the road from behind a parked car. Obviously in these cases the driver should be rewarded rather than penalized. As more data are available (weather, road conditions, traffic density, vehicle position in lane, proximity of objects to the vehicle…) better models will be deployed and more accurate risk assessments made. Most of this data is already available on the vehicle data bus, but accessing it requires working with OEMs to create compelling value propositions for all parties.

Meanwhile the industry has moved beyond passive safety systems (airbags, impact protection beams and seatbelt pretensioners) by fitting active systems designed to avoid rather than survive collisions. These Advanced Driver Assistance Systems (ADAS) include adaptive cruise control, lane keeping assist and active braking, the have begun to transition vehicle control from drivers to algorithms. Moving to full autonomy is expected to eliminate about ninety percent of collisions, saving over thirty thousand lives a year in the US alone. Autonomy also renders existing collision insurance obsolete, presenting something of a problem for the $200B auto insurance industry. Autonomy also accelerates the transition to Mobility as a Service (MaaS), where passengers pay for point-to-point transportation rather than acquiring and insuring an underutilized depreciating asset. By bundling depreciation, insurance, maintenance, registration and fuel into a single fee, MaaS creates a compelling opportunity for fleet owners to either self-insure or use the wholesale re-insurance market to manage risk. The net result is that there will not be much left of the retail auto insurance market two decades from now.

Now here’s the fun part: the need for auto insurance will actually grow as the retail market evaporates. To understand why we need to look at how liability will be apportioned in the age of autonomy. This is a complex issue involving many parties: vehicle occupants, owner, OEM, component software and hardware suppliers, intelligent infrastructure suppliers, public sector operators, and even cloud service providers. The lack of a rational, let alone standardized, legal liability framework is as big an impediment to the widespread deployment of autonomous vehicles as any technical hurdle. The risk averse automotive industry has little desire to subject itself to ill-defined and potentially open-ended liability. If that were not enough the risk profile of every vehicle changes over time, sometimes radically overnight years after its manufacture.

To understand how this happens, consider that new vehicles have upwards of 100 million lines of code spread across a hundred, or more, heterogeneous embedded processors. No organization can afford the time or expense to write this code from scratch. So, the software is an amalgam of open source libraries, commercial frameworks, customization and new functions implemented by multiple organizations along the supply chain. Software controls the sensors that provide situational awareness and the ECUs that determine and implement braking, throttle and steering actions. At the industry average of about 15 errors per 1,000 lines of delivered code, this equates to 1.5 million bugs per vehicle. The majority of these bugs may never be discovered, or may come to light years after widespread deployment.

Bugs run the gamut from mildly annoying, such as an occasional false alerts, to the life threatening failure to avoid an obstacle. They also create security vulnerabilities that can be exploited at scale thanks to always-on broadband connectivity. Potential cyber-attacks range from retrieval and resale of personal information or immobilization for ransom to the truly catastrophic. One frightening scenario would be spoofing exhaust manifold temperature readings to cause a fire. Unfortunate if it is one vehicle, but an attractive terrorist target if it is say the majority of vehicles gridlocked over the Brooklyn Bridge during rush hour. Similarly, criminals set on financial gain might instigate a series of attacks against a single brand causing high-speed collisions every few days while demanding ever an increasing ransom to stop. Public reaction to the inevitable media frenzy would likely decimate profitability for the OEM.

The average light vehicle is 11.6 years old, an eternity in software terms. Thanks to Moore’s law, systems thought secure when originally developed become increasingly vulnerable to faster hacking tools and known exploits in shared libraries. Microsoft faced the same issue a decade ago with Windows. Its’ solution was to push patches to every PC every month, with zero-day exploits and other critical bugs patched immediately. This same push update approach is essential for each ECU in every connected vehicle for the operating life of the vehicle. To understand why this is so important, consider that historically nearly all vehicle collisions were caused by driver error – relatively high frequency, uncorrelated low impact events. Future risks will impact thousands or potentially millions of vehicles –low frequency, highly correlated with very high impact.

This new threat profile points to the future of auto insurance: OEMs and their supply chain partners will insure against extremely unlikely, but potentially catastrophic events. As with commercial cybersecurity insurance, organizations (and their supply chain) will need to demonstrate sound security practices and be subject to recurring security audits. By necessity, the currently in vogue approach of security-by-obscurity will be replaced by rapid iterations of testing, hardening and Over The Air (OTA) patching. Without this many vehicles will simply be too dangerous to be operated on public roads. Insurance underwriters, and eventually legal regulation, will determine vehicle end-of-life rather than mechanical failure.

Microsoft commits to ten years of support for each new version of Windows. Vehicles might require a fifteen or twenty year support commitment, during which sustaining engineering teams are funded. Customer perceptions need to change from the appliance model to the cellphone model, where an expensive product is useless without recurring network service payments. MaaS and all-inclusive lease programs (such as BOOK by Cadillac) are making it easier to fund long term software support. Perhaps the outright purchase of a vehicle will soon be a thing of the past. One thing is for sure, this will not be your father’s auto insurance market.

[Note: this post builds on ideas introduced in two earlier posts RIP Automotive Conventional Wisdom and Vehicle Data: What’s The Big Deal? they are probably worth a read.]

Apple’s market capitalization briefly exceeded $750B last year, despite falling from that high it is the world’s most valuable public company. ARM, the UK-based leader in low power processors for mobile  and embedded devices, was just acquired for $32B. What these companies have in common is that neither of them makes anything. Apple uses a variety of contract manufacturers, most notably Hon Hai, to build its iPhone, iPad, and Macs. ARM goes one step further and licenses its chip designs to semi-conductor fabricators for a royalty stream.  Both are adept at developing valuable intellectual property (IP) and have optimized their businesses around IP monetization. For either company owning manufacturing plant would increase capital intensity, reduce margins and erode shareholder value.

So what does this have to do with the capital intensive, manufacture-at-scale global automotive industry?  Everything as it turns out.  By way of analogy consider the recent history of the smartphone industry.  A decade ago Nokia, RIM and Palm dominated.  Then in 2006 Steve Jobs introduced the iPhone, a radically better device that drove Apple’s meteoric rise while mortally wounding the slow to react incumbents.  What happened next is the really interesting part: Google acquired Andy Rubin’s Android startup, refocusing it as a smartphone operating system. Rather than use the manufacturing abilities of recently acquired Motorola, Google decided to build reference phone designs and open source Android. One key requirement was that Android phones were required to connect to the Google store and provide Google with telemetry. Within a few years Android taken over 80% market share and Apple had become more of a high-end niche player (the Tesla of phones, but with better margins).

So now really back to cars: it costs several billion dollars to develop a world-class vehicle platform. Leveraging the platform across as many models as possible without compromising unique brand attributes maximizes profit per vehicle (see 1970-80s Detroit badge engineering for what not to do).  Volkswagen are the masters with just four platforms supporting tens of vehicles across brands as diverse as Skoda, Audi, Bentley and Porsche.

Now the fun part: let’s say you and I run successful a cloud services company with over $100B in cash.  We might look at the rise of Android and apply the lessons learned to automobiles.  Our goal would be to gain an unassailable position as the de facto standard for connected and autonomous vehicles in six (relatively) simple steps:

1. Invest several billion dollars into developing a modular vehicle architecture that accepts multiple powertrains and is flexible enough to underpin family sedans, sports coupes and crossover SUVs.  We might arm our recruiters with signing bonuses and stock options and point them towards VW’s platform engineers to speed things along.
2. Pay careful attention to creating stable interface points: both physical (such as firewall position and suspension mounting points) and virtual (such as cloud and vehicle software APIs).  We would ignore legacy technologies, implementing truly secure high speed networking (STP with Gb not Kb throughput), a 48v electrical system, push firmware upgrades to every device, etc.
3. Leverage our firm’s extensive cloud services infrastructure, machine learning capabilities and monetization model for third party developers. We understand that the ecosystem is key and will pull out all the stops to make it compelling for both developers and consumers.
4. Build a series of reference vehicles and certify them for sale in key markets initially and worldwide ultimately. We use more petty cash to ensure that we had a seat at the table for vehicle regulation and standards discussions.  This will help us to avoid surprises and ensure that our designs remained legally compliant in all markets.
5. Open source the design and allow anyone to manufacture either a complete vehicle or subsystem royalty free.  The only requirement is that every vehicle must connect to our cloud services and share telemetry data with us.  We might get the ball rolling by working with a few of the larger Chinese manufacturers anxious to make the jump to world-class product much faster than their Korean and Japanese neighbors.
6. We would setup a component certification function to ensure interface compliance and safety. Testing costs would be borne by the component manufacturer and once certified we would help create a market for their component (think B2B app store). The objective is to remove the big company politics and purchasing department inertia to unleash innovation: from novel powertrain systems to HMI advances. The best ideas and most compelling value will quickly bubble to the surface. We might acquire some of the best ideas to make them available to the whole ecosystem.

Our six stage plan represents a double digit billion investment, perhaps 10-15% of the cash on our balance sheet.  However, we would have fundamentally changed the economics of being a vehicle manufacturer.  Unburdened of R&D, legal and regulatory compliance, qualification and purchasing negotiations, the minimum efficient scale to deliver world class product falls through the floor.  We are aligned with consumer purchasing preferences moving from horsepower to infotainment; something likely to accelerate with the rollout of autonomous vehicles.  As an added benefit consumers can keep their cars current with lifetime free software updates from us and the ability to upgrade modular hardware components every few years. Like watches there will likely always be high-end niche manufacturers, but the potential for this approach to gain traction should worry mainstream manufacturers and suppliers.

Another point to consider is that by focusing on IP creation rather than manufacturing plant we can employ a bottom-up rather a top-down market entry strategy: i.e. we could start with low cost vehicles in highly populated, but less litigious and regulated markets such as China and India.  We can them move upmarket with our manufacturing partners once we collectively gain experience.

P.S. If you happen to run a successful a cloud services company with over $100B in cash, or know someone who does, I have a great idea for you.

Since the 1980s Formula One pit crews have been using telemetry data transmitted from probes connected to the engine, chassis, brakes and driver to optimize vehicle setup, driver performance and race strategy.  It is an essential contributor to winning.  Similarly, telemetry for road vehicles has always been valuable.  Development vehicles are heavily instrumented to help dial in settings for suspension, engine management and a host of other systems.  For over twenty years automotive OEMs have been aggregating and analyzing telemetry data every time your car is hooked up to a diagnostic computer at a franchised dealer.  It helps trace faults, identifies and applies needed software updates, and ultimately leads to better built, longer lasting vehicles.

With the advent of cheap reliable cellular communications in the 1990s connected vehicles went mainstream.  GM launched OnStar in 1996 and moved to 4G/LTE embedded radios in 2014, closely followed by other OEMs.  New vehicles offer rich infotainment experiences and can provide constant telemetry to be aggregated analyzed and acted on (1ms latency and up to 10Gbps with 5G).

This has not gone unnoticed by the leading technology companies, all of which extended their ecosystems to smartphones over the past decade adding tens of billions to their market capitalizations in the process. So far two have tipped their hand (a little): Apple released CarPlay in 2014 and Google followed with Android Auto in 2015. They both extend the phone experience familiar to over a billion users to the vehicle, displaying a simplified interface on the infotainment screen.  This is just the beginning: at Google’s recent IO conference Android Auto was demonstrated driving a 17″ center display and a full LCD instrument cluster.  This required integration with powertrain management, climate control, and other vehicle systems historically walled off by oems.  The real prize is not device or system sales, it is to collect as much useful data as possible with the lowest latency and to monetize it across three categories:

Vehicle Telemetry

Modern vehicles have over 100 compute nodes and hundreds of probes sitting on multiple networks.  Aggregating and analyzing this data can identify when component is nearing failure so that a replacement can be waiting at the dealer when you bring your car in.  It could eliminate OEM emissions certification by reporting the vehicle’s actual environmental impact (interesting to tax authorities).  Aggregating data across OEMs or suppliers would enable a quantitative ranking of attributes that Consumer Reports and industry analysts can only approximate.  Every interaction between a vehicle and its occupants generates data that can be used to tune the Human Machine Interface (HMI) in much that same way that firms like Facebook and Amazon are continually testing refinements to their user interfaces.

Situational Awareness

New cars know a lot about their immediate surroundings: external temperature and precipitation, rear and increasingly 360° video, speed and traffic density (from park assist, adaptive cruise control and lane change assist sensors), road conditions (from accelerometers and abs).  Aggregating this data with low latency provides lane level insight into what is happening on the roads.  Diversions, debris, animals (live, dead and in-between), potholes, accidents, oil patches can all be identified and drivers alerted–easing congestion and improving safety.

Machine learning from a massive number of data points will provide accurate predictive traffic models of tremendous use to roadway planners in the long term, and as inputs to traffic signal phase and timing in the short term.  Aggregating navigation destinations from a critical mass of vehicles, in conjunction with the predictive traffic model, will enable routing based on very accurate lane level traffic forecasts and provide drivers with estimated journey time at different departure times.

Determining accident fault becomes easier when you have telemetry from the crashed vehicles, adjacent vehicles and video streams of the incident from multiple angles.

Occupant Information

Every time you interact with a Google, Facebook, Amazon or other sophisticated cloud-based service you are revealing something about your preferences, interests and behavior–enriching their profile of you.  This enables you to receive better service, quickly find things that are important to you and avoid much of the chaff.  It also enables more granular marketing, creating higher conversion and ad rates. Studies have demonstrated that smartphone users willingly sacrifice privacy if offered something of value in return; vehicle occupant data is no different.  Using a combination of vehicle, smartphone and key fob data it is trivial to identify vehicle occupants with a very high degree of certainty making privacy and anonymity notional without taking extreme measures.

You may not be able to see the drool, but this is causing much salivating in Mountain View.  This is a winner takes pretty much everything game: think Google (>70% market share) vs Bing (<15%), Android (>80%) vs Apple (<15%), Facebook (1.1B active users) vs Google Plus+ (120M), you get the idea.  If you are player number four or five it is probably best to pack up and go home.  Traditionally, the automotive industry does not work this way.  For example, Porsche has about 0.25% global market share yet generates similar profits to OEMs with more than an order of magnitude more sales.  The difference is that winning technology platforms create a vitreous cycle that builds and reinforces both the user base and ecosystem.  Users are invested in their ecosystem and want their favorite apps.  Developers go where the users are: don’t hold your breath for that Android/iOS  app to appear on Windows Phone.  The best developers are unlikely to build applications for OEM platforms regardless of how elegant the API is if most of the users are on CarPlay and Android Auto, it just does not make economic sense.  Without broad app support and a critical mass of users you don’t have a valuable ecosystem and you are not going to win.

Let’s Scare the Neighbors

Cars have been around since the late nineteenth century, starting as a hobby for engineering enthusiasts intent on frightening their neighbors.  By 1910 there were thousands of manufacturers producing steam, electric, gas and diesel vehicles of wildly different design, quality and price.  They remained toys for the wealthy until Henry Ford combined Frederick Winslow Taylor’s industrial efficiency ideas with a moving production line.  Together these enabled Ford to consistently lower cost throughout the twenty-year life of the Model T, passing savings on to the increasing number of consumers that could afford to buy it.  By the mid-1920s Ford had defined conventional wisdom for automotive manufacturing: standardized models with limited options, powered by gasoline engines, produced at scale, minimized and deskilled manual labor, with vertical integration and tight contractual relationships with a just few suppliers.  It worked spectacularly well and within a decade over 1,700 of Ford’s competitors closed their doors.

Middle Age Spread

Following a period of rapid growth and consolidation the manufacturing powerhouses that survived enjoyed substantial leverage over their suppliers and dealers.  As profits rolled in, markets were segmented with new models and brands, international subsidiaries created, and fiefdoms built.  Despite both valiant and reckless attempts to build new domestic automobile manufacturers, the financial hurdles to efficient scale doomed them to failure. Or at least that was true until Tesla. In 2015 Tesla sold more luxury cars in the United States than any other manufacturer. Tesla is disruptive, and it is no co-incidence that it is the only automotive manufacturer run by software guys. Much has been written about Tesla, but stepping back from the hype, there are three key decisions that Tesla made that set it up for success:

1. Legacy, What Legacy?

Tesla started with an electric-only vehicle that simplified platform and power train engineering, reducing both development and regulatory compliance costs. Just as importantly eliminating the engine, transmission, prop shaft, differential, exhaust and a host of other legacy technology created an opportunity for significantly more efficient packaging than the competition.

2. Apple Store for Cars

Having studied Ron Johnson’s success with Apple Stores, and realizing that electric vehicles had minimal servicing needs, Tesla decided on Internet sales supported by company-owned stores. This enabled Tesla to tightly control messaging and the customer experience. Both proved essential to building the goodwill necessary to overcome multiple product delays and deficiencies on the way to achieving record scores in Consumer Reports customer satisfaction surveys. As a bonus Tesla Stores provide convenient charging locations helping to overcome range anxiety. Initially eBay executives were shocked that people were buying expensive cars on the auction site; so it should be little surprise that a decade later consumers are comfortable using online configurators to spec and purchase new cars.

3. Scrum, Open Source & Off the Shelf

Perhaps most importantly Tesla employs modern software and hardware development best practices. Control systems leverage open source Linux software and cheap off-the-shelf components hardened just enough for acceptable duty cycles. Building on a community supported base enables Tesla to benefit from the long term evolution of a broad and robust software platform without having to underwrite its development, or be held captive to a Tier 1 supplier’s proprietary system. Screens, GPUs, CPUs and RAM can upgraded during the 12+ year life of the vehicle, enhancing both customer satisfaction and residual values while generating high-margin revenue. Agile software development sprints, open beta programs, push update capabilities, and cloud-based telemetry tracking enable continuous evolution; Tesla can fix bugs and enhance the user experience while owners sleep. Meaningful new features are sold to existing customers, immediately amortizing software development across the entire user base, not just future vehicle sales.

Lean, Scrappy & Innovative

As a company run by Silicon Valley software guys Tesla exhibits the scrappiness and innovation that characterized Ford a hundred years ago. As a startup Tesla did not have the industry or financial credibility to forge meaningful relationships with leading automotive suppliers, pushing Tesla to innovate. Yes, Tesla is facing scale-up issues at its Fremont plant and Elon Musk is perceived by many as too aggressive with the scope of his vision, financial projections, time to market, and releasing new features before they are fully baked. All true, but in a few years Tesla has overturned one hundred years of conventional wisdom, becoming a large scale manufacturer with sales growth, order backlog and customer loyalty envied by the rest of the industry.

Many competitors are trying to emulate aspects of Tesla.  Unfortunately nearly all of their profit is driven by what Musk would call legacy business, placing them in the classic innovators dilemma. Tesla is the most visible sign of much more substantial technology-driven dislocation underway in the global automotive industry. In the words of BTO “You ain’t seen nothing yet”.