CES 2024: AI and SDVs

Another CES is behind us. The automotive and mobility extravaganza of years past was replaced this year by an AI extravaganza. For the automakers and their suppliers that participated at CES this year was about AI and Software-Defined Vehicles (SDVs). There were lots of demos involving AI and SDVs, and even more marketing statements full of hyperbole about AI’s potential to benefit the customer experience. Let’s give it 12-18 months to see what turns from vision to reality.

I spent a few days at CES this year. It was my first visit since 2020. The crowds returned (and along with them the long lines to every service). From the time you arrived at the Las Vegas airport, you could tell that this CES was going to be about AI. But it was also the CES where Software-Defined Vehicles were no longer an abstract notion as in years past. I left Las Vegas feeling positive about the progress relating to SDVs and concerned about the AI hype. It’s always good to see corporations present their technology visions. But it’s a problem whenever a vision is presented as reality.

As I write in my new book, The Flagship Experience: How AI and Software-Defined Vehicles Will Revolutionize the Automotive Customer Experience, Software-Defined Vehicles are vehicles that are based on alternative powertrains, primarily battery electric, incorporate a level of driving automation that provides them with enhanced safety characteristics, and whose novel architecture enables their easy configuration and continuous updating. They require new architectures, new tooling to develop and test them, powerful computing platforms to process the data they generate and provide the automations that characterize them, and complex cloud infrastructures to them properly configured and updated. This year the term SDV could be seen in many more automotive booths and suites. The announcements and demonstrations by Hyundai, Honda, Afeela, Mercedes, BMW, Bosch, Continental, Valeo, but also AWS, KPIT, and Sonatus led me to believe that certain incumbent automakers and their suppliers are making good progress toward introducing SDVs that are based on zonal architectures before the end of the decade. I consider zonal architectures a prerequisite for taking full advantage of what software-defined vehicles claim to offer. While these days we read about the slowing demand for EVs, and many OEMs are reconsidering their investment plans around such vehicles, the capabilities of zonal SDVs, combined with continued cost improvements, will make such vehicles strong value creators for automakers and users alike (be it consumers or business fleet operators). I felt very encouraged by what I saw and heard.

The Flagship Experience is based on the premise that zonal SDVs combined with AI will offer automakers two important value-creation opportunities. First, the opportunity to better understand their customers’ mobility behavior and offer them personalized and often monetizable customer experiences. Second, the opportunity to monetize the SDV over all its owners rather than just the initial one, as is the case with vehicles today. This year automakers and suppliers were eager to show how AI can personalize the automotive customer experience. A good case in point: BMW and Valeo demonstrated a valet parking assistant that takes advantage of the SDV’s ADAS capabilities (which include AI), and teleoperation (which also incorporates AI), to remotely park a vehicle without the driver being in it. Less convincing but still interesting: Mercedes demonstrated how AI can be used to tailor the in-cabin music experience to the driver’s style.

Problematic case(s): Many other companies from the automotive and mobility ecosystems demonstrated generative AI-based assistants that can address various aspects of the mobility customer experience, e.g., charging, parking, navigation, etc. For example, companies such as VW, Afeela, Hyundai/42dot, TomTom, Mapbox, and others demonstrated conversational AI applications that are based on ChatGPT for intelligent navigation (for example, to find the nearest charging station or restaurant serving a particular type of food). With my investor and technologist hat on, I had a problem with several of these and the marketing statements that accompanied several of these demonstrations.

Like many companies, over the past year, our firm has been experimenting with generative AI tools and models on behalf of our corporate customers and to address our internal needs. Through these efforts we have come to understand how far chatbots based on publicly available LLMs can take us towards accomplishing specific tasks, the advantages we gain by fine-tuning LLMs and by creating proprietary Small Language Models (SLMs), and the work we need to perform to create intelligent agents that reason and automatically learn. The bottom line is that while it is easy to build an AI application demo that presents a vision by hardwiring a few software components with some data, it is mighty hard to transform the vision into reality and build a scalable system that can work across different cases and simultaneously be used by many users. Even if the technology challenges can be overcome the costs to operate and maintain such a system can be prohibitive in the absence of a solid business model. What bothered me with the AI demos I saw at CES is that neither the technology-related nor the business-related parts of the requisite analysis appear to have been performed by the companies that are presenting these demos. For example, what tools and knowledge bases need to be combined with ChatGPT so that restaurant recommendations will match the user’s preferences, vehicle location, and stated constraints? What type of cloud-based and communication infrastructures needs to be created to provide such a service to all users consistently and promptly? How much will all this cost the provider and which users will be willing to pay for such a service consistently to make it viable? Automakers tried several times to offer digital services to the owners of their connected vehicles without understanding their needs. Few of the services that were introduced in this way ended up being used repeatedly and even fewer had an enduring business model.

There exists a cross-industry excitement about AI. CES 2024 was only the most recent manifestation of this. The automotive industry is correct to see a potentially big opportunity that can be captured by applying various AI technologies, including those associated with generative AI, to both the next-generation Software-Defined Vehicles and the customer experience automakers want to offer around these vehicles. However, to avoid the problems with the previous AI eras when companies overpromised but underdelivered, and ensure that what is delivered provides value to both the provider and the customer while making solid financial sense, over the next 12-18 months it will be important to clearly understand their customer characteristics and needs, the true capabilities and scalability of the AI technologies used, the costs of the envisioned solutions, and the business model that will bring everything together and make it a winning proposition for provider and customer.