Berkshire & Markel, Tesla, ServiceNOW, Bergman & Beving

Content Published Last Week

1. Tesla: Artificial Intelligence in Autonomous Driving

2. Tesla: Manufacturing Process Advantage

3. ServiceNow: Customer Maturity, New Products & Value Realization

4. IP Research: Bergman & Beving

Sponsor

close

Zoom InZoom OutReset

Visible Alpha built a platform to analyze consensus data for financial and operating metrics on over 6,400 publicly traded companies. Rather than digging through models one by one, Visible Alpha creates consensus data for every line item included in sell-side models so institutional investors can better understand expectations on metrics beyond just revenue and earnings. Try Visible Alpha for free here.

Berkshire 1968-88 & Markel: Power of Float

As always, visiting Omaha for Berkshire’s AGM was good fun. It was great to see many of our Partners and subscribers throughout the weekend.

We also attended Markel’s Brunch on Sunday. Markel often gets pitched as the mini-Berkshire given it has the three compounding engines: insurance, privately-owned businesses, and public investments.

Although Gaynor himself proclaims he is no Buffett, maybe he doesn't have to be for Markel to compound ~15% for decades?

Over the last few weeks, we skimmed Berkshire’s 1968 - 88 letters and annual reports. One fascinating takeaway is that BRK’s insurance division had significant underwriting losses during the period, yet its BVPS compounded at 23%+ over the 21 years.

The investment gains from investing the float compounds equity value even in highly challenging times.

As the table below shows, BRK's insurance underwriting losses in 1975 and 1976 alone wiped out all historical cumulative underwriting profit of the previous 8 years.

Even worse, the stream of yellow highlighted cells in the table shows how inflation and increased competition in the late 70’s / early 80’s crushed underwriting performance. This not only highlights how difficult insurance is, but it proves the resilience of intelligently allocating insurance float over long durations. Please note the premiums and underwriting losses in the table below excludes structured settlements and portfolio reinsurance to get a clean underwriting result for direct insurance and reinsurance.

close

Zoom InZoom OutReset

On insurance, one Buffett quote comes to mind:

Surprises in insurance are far from symmetrical. You are lucky if you get one that is pleasant for every ten that go the other way - Warren Buffett, 2006 BRK Letter

Such poor results above weren't due to Berkshire's lack of underwriting skills, the whole industry was crushed:

close

Zoom InZoom OutReset

Even with nearly a decade of underwriting losses, BRK compounded BVPS at 23% from 1968-88. More impressively, from 1980-88, the period with greatest underwriting losses, BRK compounded BVPS at 27% annually. This highlights the power of the free leverage from investing the insurance float. The result is even more amazing given BRK held its declining Textile Mill during this period.

Like Berkshire, Markel has an aligned and high-integrity management team that benefits from the 'free leverage' of an insurance float.

Buffett solved both the cost of capital side via free float and the investment side by his capital allocation genius. Markel may have the cost of capital side solved, the question is how does the investment side perform?

But even without a capital allocator like Buffett at the helm, Markel's cheap float can lead to an attractive risk-reward of 12-15% compounded over a long duration. Such a return doesn't necessarily require Markel Ventures and its investment portfolio to achieve anywhere near Buffett's investment returns.

Over the next 6 months, we will be unpacking BRK's early returns and comparing it to Markel, Boston Omaha and others intelligently allocating insurance float.

Tesla Full Autonomy

Last quarter, Musk made an interesting comment on TSLA’s Q1 call:

Tesla is in a uniquely strong strategic position because we’re the only ones making cars that technically we could sell for zero profit for now, and then yield actually tremendous economics in the future through autonomy. - Elon Musk, TSLA Q1 2023

This is a bold statement.

The first of our two TSLA interviews published this week was exploring just this topic: how realistic is it that any manufacturer can reach Level 5 Autonomy within the next 5 years?

The executive interviewed has over a decade of experience at a top ride-sharing company and Toyota's AI division building out Level 5 autonomy.

In short, the executive's answer, at a minimum, is that Level 5 autonomy a decade away.

This is mainly because it requires such a huge amount of real-world training data to cover all edge cases when behind the wheel:

In autonomous vehicles, AI is responsible for processing information from various sensors to understand the environment and make decisions. These sensors include cameras, LiDARs, radars, IMU, GPS, and others. The challenge for AI is to take all that sensor information and decide what to do next, such as accelerating or slowing down and turning the steering wheel left or right. Essentially, the AI for a self-driving car is a contextual engine that takes context, in the form of sensor data, and outputs velocity and steering angle. It's a large multi-modal model with a simple output. However, this is an oversimplification, as the AI must also make these decisions in a very short amount of time. The faster the vehicle is driving, the less latency can be tolerated in the system. If it takes multiple seconds to decide whether to accelerate or slow down. while driving at 50 miles per hour, accidents can occur. Latency is critical. - Former Head of Level 5 at Lyft

OEM’s will define themselves by the data they collect and keep. And given the scale of data collected, all data from the vehicle cannot be kept.

What I'm saying is that data matters, and the way you decide what data to keep and train on is where some of the magic lies. You can't keep or train on everything, so you have to recognize, perhaps using other AI, what data might be useful. The edge case data that you collect yourself or generate through simulations is a critical piece of the puzzle. -Former Head of Level 5 at Lyft

TSLA seems to be the clear leader in terms of real data collected:

One advantage of Tesla is their shadow mode. I believe that even if you don't have full self-driving mode enabled or purchased, some of it is running all the time, predicting what it should be doing. When it notices a deviation between its prediction and what the driver actually does, that is considered interesting, and some of the data is saved for training. With four million cars, you gain a huge advantage in understanding real-world data and edge cases. There are different types of accidents, from fender benders happening every 10,000 or 50,000 miles to fatalities occurring every 100 million miles. This indicates the size of a fleet needed for companies like Waymo or Cruise to observe these unusual scenarios. Tesla has a significant advantage, in this sense, with their enormous fleet. - Former Head of Level 5 at Lyft

The jury is out on whether companies like Cruise or Waymo can replicate such real-world data via simulations.

But even with TSLA’s potential advantage, if Level 5 Autonomy is over a decade away, how does the company perform until then with the onslaught of competition?

Part of TSLA’s bull case is that it can price vehicles lower than legacy OEM’s and still earn higher gross margins due to its unique procurement and production process. Our second interview explores this point and shares insights into how TSLA’s builds, organises, and designs its engineering processes:

In car manufacturing, you see a lot of up and down; you see a lot of turning and movement that adds no value. It's just moving. Elon scrutinized the details of those movements and directed his trusted people for any new factory layouts to ensure that nonvalue-add works were minimized. This was one of the key factors. He always insisted at Tesla that the strategy was to have a dense factory with a reduced footprint because footprint is a direct factor to capex. If you can save one square foot, you will save on HVAC, lighting, and everything in the factory. One square foot means a lot. The others, honestly, I think the other companies have a best practice based on their previous projects. None of those companies have been building one factory per year. Many of them have factories that have been established for years – Ford, GM, and even Toyota – and they keep doing the same thing. I'm pretty sure none of those factories, before Tesla, even thought about the volume efficiency of the space. - Former Manufacturing Manager, Tesla

The two interviews published this week are interesting to read on TSLA in parallel.

ServiceNOW Stickiness

ServiceNow has been one of the most successful Enterprise SaaS companies of the last decade. When asked about the risks the company faces for future growth, a ServiceNow consultant at DXC Technologies referred to people rather than competition as the largest risk:

I would say the number one risk is people. There simply aren't enough trained, certified, and skilled individuals to do the things ServiceNow is doing. They're investing in programs like ScaleUp and RiseUp, but they'll need many more people over the next five years. They currently have around 750,000 trained individuals, but being trained and being a pro are different. If that number needs to increase to 1.5 or two million people, that's a lot of people to bring into the fold.- Current ServiceNow Ecosystem Leader at DXC Technology

A healthy ecosystem of trained professionals plays a large role in realizing the value of an enterprise SaaS platform. Partners and certified professionals invest in the platform and ultimately increase the number of workflows adopted by customers. This drives retention and ACV for NOW. The executive explains this mechanism further:

the dollar value of a deal with two workflows is three times that of a deal with one workflow. If you add a third workflow, the dollar value jumps to almost nine times. This is because taking advantage of the capabilities within each workflow provides the biggest bang for the buck. - Current ServiceNow Ecosystem Leader at DXC Technology

Bergman & Beving

Bergman and Beving is a Swedish serial acquirer of tools and construction products across Europe. The company is trading at ~15x NTM FCF and has a relatively new CEO who used to run M&A at Lagercrantz.

We’ve recently covered B&B from multiple angles:

1. Interview with the Current CEO

2. IP Analysis: Bergman & Beving, Lagercrantz, Nordic B2B Distribution & M&A

3. Interview with a Former B&B Board Member

In this audio-only analysis, we explore why we’ve spent time on B&B, how it compares to other acquirers, why we think it’s cheaper than competitors, and its opportunity and challenges ahead.