When bits & atoms merge: Software is not eating the world, but governing how industries work
Tech companies are not succeeding because they use software to do business in novel ways. They rather create new industry dynamics that incumbents can't compete with because of different world views.
What do companies like Waymo, BYD, Xiaomi, Manna Aero, Amazon Logistic, Somos Internet, Apple, Peloton, Tesla, SpaceX, Anduril or Boom Supersonic have in common? I claim that they all offer physical, real-world products or services (cars, ride services, drones, shipping, internet services, phones, rockets, jets) and at the same time are actually software (or tech) companies at heart. This claim obviously begs the question: What exactly is a tech company?
In a 2019 blog post, tech analyst Ben Thompson answered the question the following way:
The question of whether companies are tech companies, then, depends on how much of their business is governed by software’s unique characteristics.
Note that he says “governed by software’s unique characteristics” - not merely by software per se. So in order to better understand this characterization, we have to better understand what he means when he says “business” and “software’s unique characteristics”.
What exactly is a business?
A business makes products or services that are of value for a customer. So far, so obvious. But “business” is not just a synonym for the word company (as in the organization, the brand). The “business economics” (or business model) describes the mechanism of how a company defines, creates and captures such value. The mechanism consists of four high level sections:
Value design and creation
Definition of the core value a product or service can offer
for whom & how often?
Design & architecture of products or services that can deliver that value
How their operations/production processes work to
Customer acquisition and relationship management
How marketing and sales are being done
How a customer relationship is managed
External value network dynamics
The supplier relations a company taps into and is part of, up and downstream
The larger industry dynamics (a detailed explanation of the idea will follow later)
Cost / revenue structure management
The revenue model & cash flow management
Capital and operational expenditures
Margins
“Business” is the interaction of all of those elements in order to serve a customer. The problem you solve determines the value for your customer. It also determines the nature of your customer relation (because of how often transactions happen - a one-off sale, a subscription?). Your product value (how well you solve the problem, compared to your competitors) determines how much money you can make (do you have a secret sauce, maybe?). It’s the product or service architecture that determines your supplier network and how much of your offering you are making yourself or buying (whether you are more integrated or more modular determines how close your supplier connections are). It also determines your processes (and vice versa: your existing processes determine what kind of product you can actually make). Together they determine your cost structure and whether you can actually make a profit.
Note that a business design is rather technology-agnostic. There are really good businesses (selling coffee: high margin, daily purchase frequency) that are low tech, and really bad businesses that are higher tech (airlines operate very complicated flying machines and make basically no money at the end of the year).
What makes software unique?
Now let’s look at software. Thompson lists (among a few others) three characteristics that make software unique:
Software has zero marginal costs
Software improves over time
Software enables zero transaction costs
I will explain each in a little more depth now and add one further characteristic myself:
Software now can understand messy data
Zero marginal costs
Creating and running one additional copy of a given application costs you almost nothing. Sure, you need more server space, which costs a little bit (mostly electricity). But the bigger the scale of your product or service, the more negligible those costs become, especially compared to the effort to create the software in the first place. At scale, software marginal costs approach zero.
Continuous improvements
Opposed to a physical good, software is never finished. It can easily be changed and updated - especially in cloud-native world, where software is hosted on someone else’s computer, as-a-service, and accessed over the internet. This fact enables a reinforcing feedback loop: software that is used in production generates and collects data, which then can be analyzed in order to change and improve its functionality or usability.
Zero transaction costs
A transaction is the process of transferring (buying, selling, exchanging) a good, service, or resource (tangible: money, goods; intangible: services, information, rights) from one party to another. The transaction typically creates, redistributes, or preserves economic value of some sort. Transactions occur
Internally (design & operations): the exchanges (e.g. of materials or information) that enable the definition and creation of value
Externally (customer & supplier relations): buying from suppliers or selling to or customers
Transaction costs refer to the expenses incurred to make such transfers happen. They go beyond the actual price of the product or service. They include all the additional efforts required to complete a transaction. (Networked) Software can reduce such transaction inefficiencies dramatically:
Search
It’s now easy and cheap to search for products and services online and even purchase them right away. The same is true for information within a business (database entries, documentation).
Communication
Messaging apps or wikis like Whatsapp, Slack, or Confluence make it very cheap to exchange information asynchronously with anyone, anywhere on earth with an internet connection (both within and between businesses). That’s true for machines as well: Standardized languages (e.g., JSON, XML), APIs and protocols let them "talk" without custom hardware links.
Negotiation
Hiring a taxi involves calling a dispatcher and negotiating a pickup time. Software can transform this by automating and digitizing the process, reducing the cost to nearly nothing.
Decision-making
Software automates decision-making based on defined rules and data as input and automatically produces an output. Sensors for example can stream data to a central system that adjusts machinery instantly based on algorithms.
Coordination
Uber’s app matches riders and drivers instantly using real-time data on location, driver availability, and pricing.
Enforcement
A DevOps team at a tech firm uses CI/CD pipelines (e.g., Jenkins) to deploy code multiple times a day. Transactions (code reviews, testing) are automated and instant, accelerating releases.
Handling unstructured data
Enter AI, or software 2.0. While software 1.0 until now took structured data (think lists, tables) as input and calculated the output of an operation based on a deterministically defined rule (i.e. algorithm), AI works differently. It takes in a lot of unstructured data (think images, video, audio, text etc.), looks at them (or at a mathematical encoding of them) and then approximates the rules that are hidden in that data (if there are any), probabilistically. The fundamental shift is moving from "what is the output?" to "what is the distribution of possible outputs and how certain are we?”. Predictions will sometimes be wrong, because it trades mathematical exactness for empirical effectiveness. But in principle software and machines now can “understand” (or rather convincingly guess) what they see, read or hear.
Implication #1: The nature of software changes how businesses create value
So putting the two sections together, when is a company a tech company? Tech companies reshape the way business is being done in a given industry by leveraging software’s inherent upsides (traded off with its inherent constraints) to do it. They are not merely companies that do business the way it was done in pre-software & pre-internet days and now are using software as a tool to make it slightly more efficient. It’s not about “the business people” defining requirements (more or less) tailored to a pre-existing model and then technology being built to meet them. Instead they consider value design and creation, customer acquisition and relationship management, cost / revenue structures and how the external value network dynamics influence a business holistically. They then (re-)design their overall business approach accordingly. That’s a big difference!
Facebook (now Meta Platforms) is probably one of the first archetypical big tech companies, as it checks all the boxes above. It monetizes its user's attention via advertising - the 21st century kind of advertising. In the analog past, ads in newspapers or magazines were simply printed next to and between editorial content pieces. This more or less matched the readers’ demographics profiles, because the newspapers or magazines roughly knew which people in which age groups and income brackets would buy their products. So they could offer those slots to advertising customers. But it was a one-size-fits-all approach, which was as good as it would get in a physical world.
Facebook translated the ad logic into a tech-first approach, just like a tech company would do.
Continuous improvement and AI
In 2005 (after some trial and error) it invented the news feed. That is a dynamic, algorithmically driven stream of content that is ranked by relevance, for each individual user (based on the user’s social graph: friends, interests, fan or business pages liked, demographics and so on). One image, posting or video follows after the next when scrolling down the feed. Since ads are seamlessly woven into the user experience next to organic feed content, it’s a never ending engagement stream that gets meticulously analyzed and better on a daily basis due to a real time feedback loop (showing the right ads involves a lot of software 2.0). This gives advertisers immediate feedback about whether users click on ads and lets them adjust campaigns on the fly based on those performance metrics.
Zero transaction costs
All of this is mostly automated, from the way businesses purchase ads (a reverse auction system) to the way Facebook places them next to relevant user content.
Zero marginal cost
Behind the curtain, Facebook built a world-class distributed infrastructure, enabling billions of users to scroll seamlessly - and because of Meta's high infrastructure utilization at almost zero marginal cost. Free global access allowed the company to scale rapidly and profitably, with gross margins exceeding 80%.
Software powers every part of Facebook’s business, from value design and creation to user engagement. Tight product and process integration as a platform to scalability and cost structures. Facebook didn’t just digitize old ad workflows — it reimagined advertising for a zero marginal cost world, creating a flywheel:
Implication #2: The advent of tech companies changes industry dynamics
The rise of such Big Tech giants had broader implications for the entire advertising and publishing industry though. Meta, along with Google and Amazon, basically generates almost all of the global digital advertising revenue. Digital advertising itself took almost half of the advertising share. Online advertising went from a niche channel to taking almost 50% of the ad share within a decade after the feed was invented.
Many of the old big publishing giants that attracted the most ad spend have dwarfed or went out of business entirely. It’s not even that their product got worse. It’s just that their advertising clients went away, which made the economics of their products stop working. It’s critical to understand why.
Tech companies do not just participate in the industry game more effectively. They change the nature of the game. The “game” in this case is best described with the concept of a “value network”. A value network is the interconnected system of companies, suppliers, and partners in an industry that collectively create value to meet customer needs. It is within this context where a firm identifies and responds to customer needs, solves problems, procures input, reacts to competitors, and strives for profit. This context drives the network’s competitive dynamics and incentives:
Customer needs shape the types of products a firm must deliver, which in turn determines the processes required to build and distribute those products. These processes then dictate the cost structure — including overheads, margins, required volumes and product release cycles — that the firm must operate within.
Within this dynamic, firms take specific roles: more modular suppliers provide components (e.g., chips), while integrators orchestrate full solutions (e.g., phones). Depending on their position, firms have more or less economic power and can realize higher or lower margins. But they don’t necessarily choose their position freely. Over time, the value network becomes self-reinforcing: only innovations that fit the existing cost structure and processes are seen as viable, while disruptive innovations that require a different structure are rejected.
So firms need to think about the capabilities they do or do not need to build up (like being an integrator) and decisions they or do not need to make based on that (about products/services architectures, make-or-buy, processes and supplier ties). If they don’t, their profitability or survival might be at stake because their cost structure might not fit their pricing power and margins. It’s also not a one-off decision, since the network is dynamic, evolving with industry maturity, market size, shifting customer preferences, and technological innovations.
The conclusion: Software is not merely eating the world - it's its new governing force
In 2011 Marc Andreessen said that software is eating the world. He was kind of right. But it turns out that the world is still there. So software is not just something that completely eats up legacy business models. It changes how entire industries make sense of the world, due to its inherent characteristics:
Its malleable, dynamic nature makes it fast-paced, with rapid product release cycles. Tech firms can update software daily, even minute-by-minute.
This requires operational processes with high automation to manage frequent changes. Firms must also carefully decide what to make in-house versus outsource, as pace and adaptability are key.
Software is intangible but tightly tied to the computer hardware and networking value network. In tech, both software and hardware must be considered together.
Last but not least it incentivizes global scalability strategies due to its immense economic leverage potential. This demands high upfront capital investments and hence a different approach to risk assessment of markets and opportunities.
Those traits are now creeping into basically every traditional industry value network. They are the new governing force of the economy - and they often clash with what they meet there irreconcilably. So in a sense, software is much more cultivating the world than eating it.
Purely information-based industries like music, movies or advertising were just the beginning. More and more real-world product and services industries are getting transformed into tech-driven industries because software-first people took over. They have leveraged the nature of software, AI and the mobile and cloud infrastructure the tech industry has put in place over the last two decades. Automotive is a striking example.
Automotive
Tesla or BYD electric vehicles are computers on wheels (or software-defined cars) that get sold directly to the customer online, with just a credit card. It’s a total business model disruption.
They are tech companies that not only have re-thought how a car should look like and function in the 21st century
Most legacy auto companies until this day haven’t shipped software-defined cars (so they are missing out on all of the value that such products can generate - such as over-the-air updates of the entertainment suite or the brake performance). However, they have built sales software tools for their existing dealership partners. By doing so they have extrapolated their non-tech legacy business approach of an indirect dealership sales model. This made them better than their own old selves, while the new EV competitors are not better - they are different.
But we have to give the legacy car companies the benefit of the doubt. In the past, if you were in the car industry, you were not in the software and computation industry, from a value network perspective. Both industries followed separate approaches of doing, thinking about or prioritizing things. They had their own respective customer needs, value creation logic, product architectures, product release cycles, company processes optimized for those cycles, cost structures and so on.
Sure: Cars have been using software since the 1980s. But since CPU speed was still so slow and expensive back then, it was just a minor side note. Vehicles were not software-defined, so they could not be part of or form an ecosystem. Offerings could not improve over time, so the revenue logic was more transaction-based. Since customer relationships were mostly analog, customers had to go to a dealership in order to buy a car. Before the cloud as a business concept, cars had to be driven back to dealerships in order to get firmware updates (to the degree they were possible in the first place). Zero transaction cost processes were just not something that was part of the automotive industry’s core logic. Software as a true product and operations differentiator only became possible step by step since 2000. That’s when the two value networks of software and cars started to overlap and clash more and more, until this trend sped up dramatically around 2010, when EVs really became a thing.
BYD and Tesla had a head start here, because they started from scratch with the cultural awareness that they have to blend the two value networks into one (with great initial difficulties themselves).
Software is eating the mobility industry, and that is even before we throw in autonomous driving. Now legacy OEMs around the world are trying to catch up. Some are succeeding, some are struggling.
Logistics
Consider logistics. Manna Aero from Dublin offers last mile delivery logistics services via autonomous, battery-powered drones. They currently can deliver up to 3,5 kg worth of food, drinks, groceries or even electronics device orders over distances of 3-5 km within a few minutes (e.g. 3 minutes for a hot coffee). They already serve up to a million people in the suburbs of Dublin, Ireland, Helsinki, Finland and Austin, USA. But it’s much more of a tech business than a traditional logistics business. It even blends three value network domains: logistics operations, software and aerospace.
Manna had to go through an aircraft and airline regulations process in the EU and US in order to get their flight permissions. This is obviously something traditional logistics companies would never touch. It’s simply an approach that’s impossible from within their traditional value network logic.
Last-mile delivery is a high-throughput business - the more deliveries you can get done with one delivery unit (drone, car, bus, bike), the better. Manna’s drones are specifically designed to facilitate fast loading operations on the ground (e.g. swappable batteries that charge separately, managed by a battery management software system for maximum battery life. This minimizes drone downtimes and maximizes deliveries per hour). So the technology facilitates the business economics, not vice versa.
Order dispatching, flight routing, avionics, environment scanning at the point of delivery and so on is done via software and sensors like lidar. Integrating sensor data in real time and deciding what to do autonomously is obviously a case where software 2.0 can shine. Since each drone collects tons of data per flight, the service improves over time, in every aspect.
This enables a marginal cost structure per flight that is already much cheaper and better than a road-based delivery (due to the autonomous nature of the drone no other costs than a few cents electricity per flight, plus pro rata costs for the ground operator wage. The loading and battery exchange is currently still done by humans, but will get automated at some point in the future as well).
Manna enables a wider (food) delivery ecosystem. It exposes its drone availability and scheduling data via APIs to aggregator partners like UberEats or Doordash. They then can easily make use of Manna’s drones to serve their customers, at scale, and with cheaper, faster and better service. Food orders can simply be placed via (its own or) partner apps. It's a business relation with basically zero transaction cost thanks to software. Everyone wins. Since the system is basically completely software-defined (the flight planning, the drone navigation, the flight scheduling, the supply-demand matching etc.) it can more cheaply and easily scale to other cities across Europe and the US.
It will be hard for traditional logistics players to compete with this asymmetric logic.
The 21st century playbook: Transform into a tech company or get commoditized by one
Software and AI already today are the unique business differentiator for all kinds of real world services and products. What all these companies have in common is that they have reshaped their business models with a software-first approach to every aspect of their holistic business design.
But the impact actually hits deeper: The value network of the software industry assimilates traditional industries and shakes them up from the ground. It changes customer preferences, what is considered a differentiator and what a commodity. It alters where value is created and captured. It shifts a company's position in an industry. And most importantly it changes the risk profile of investments, towards high first copy costs and zero marginal costs. Tech companies tap into the software- and compute-heavy value network as well as in industry specific value networks with ease.
So what is the endgame here? So will every company be a technology company? Probably not. Will there be a technology company in almost every industry? Probably yes. Will this company be the one that reshapes the industry value network according to a new logic? Definitely yes! Will this company be the one that makes the outsized returns in that category? Maybe. Will it be the one that needs by far the most upfront investment? Definitely yes! So will the winner-takes-most characteristics from Silicon Valley slowly but steadily spill over into many other industries? Possibly - with all the implications this brings in terms of jobs, competition etc.
But whatever the future will bring, the key implication is this: Apart from deeply understanding business and business design, companies need to become really good at understanding
Software and its core characteristics
modern software delivery practices
the implications of the cloud
the dynamics of computation and networking
how global competition is shaped between China, the US, Europe
How software governs their industry's dynamics and value network
who has which economic power
how software is changing this dynamic
who becomes an integrator and who becomes commoditized
whether the investment risk appetite might need to adopt
how the industry pace changes due to software
That is less obvious than it sounds. Currently, in digital transformation work, the only focus lies on the software project output, mostly delivered by external suppliers. This lack of deeper value network awareness e.g. led to mistaken decisions that caused to the chip shortage in the auto industry. So the backwardness of many industries already has very expensive consequences. It’s a price that most companies can’t afford to pay for much longer.
Some more case studies
Somos Internet
Let’s move on to an even less obvious example: Somos Internet has rethought how high speed fiber internet can be deployed and an internet service provider can be run with a tech-company mindset. They currently do that in different cities in Colombia and aim to scale their business across many developing countries around the world, where billions of people still don’t have home internet access. But how does this make it a software business? Explaining that requires a short history lesson:
Around the year 2000, when internet infrastructure was laid out in developed countries, it was done by using a complicated physical cable setup of one-directional coaxial copper cables and slow passive optical networks (PON) telecom equipment for transmission. Here are the trade-offs that were made back then: Copper cables transmit data using electrical signals, which degrade over distance due to resistance and interference (e.g., crosstalk or electromagnetic noise). But they were already there, originally installed for TV signals. PON means that future improvements depend completely on hardware upgrades, which are complicated and expensive. But operating the system is very low on maintenance and power costs, which is an upside. This trade-off choice came with a few more system-level downsides though.
Bandwidth is shared among users on the same link (which serves potentially hundreds of end users in a neighbourhood). So the more people “are online”, the worse your connection gets. PON’s point-to-multipoint design (from one internet backbone to many end users) means the layout is optimized for splitting the signal along the way. To increase capacity significantly you might need to rewire parts of the network. Furthermore, the speeds are generally determined by the current standard of the hardware used across the network (e.g. specific data rates like G(iga)PON at 2.488 Gbps downstream). Upgrading to a faster standard requires replacing all of the devices, which makes upgrades a coordinated, network-wide effort that isn’t trivial, as splitters along the way are often buried in the field. And the expensive investment then locks the network into a new static state (until the next upgrade).
The main reason though why such a complicated physical setup was picked back then is that logical capabilities like IP addressing, routing or encryption are software-based: They rely on CPU power to make decisions about where data goes. So the limited computation capacities around the year 2000 forced network engineers to handle the necessary complexity physically (in the form of very complicated and “dumb” cable networks) to ensure acceptable performance. The network could not be smart, so it had to be more physically complicated.
Fast forward 25 years, the approach that Somos Internet takes in Colombia is reversed. The physical network is now simple, relying mostly on cheap (redundant) fiber cables, while the complexity has shifted to smart logical components - active and intelligent routing at endpoints, driven by commodity Ethernet hardware and powerful computing.
This comes with a few great advantages: Internet speeds are now decoupled from physical infrastructure. Fiber allows for fast transmission. However the actual data transmission rate depends on the speed of laser pulses sent into the cables and decoded at the end. Their speed depends on computation and software capabilities. Another advantage of an active network is signal quality and data integrity over long distances. Devices like switches and routers act as regenerators. They take in incoming packets, decode the signal to its original digital form, clean up any noise, and then retransmit a fresh version of the data. Last but not least, bandwidth is not shared among users, and every end-user has a point-to-point connection to the internet backbone itself.
The result? Several GBit/s connections for 50 $ dollars. And here is the overall kicker: Performance improvements across the entire network can be achieved with simple software updates or by updating computational hardware in a few distributed data centers across the city. Laying new cables or replacing switches buried in the ground is no longer needed. Somos is an almost zero-marginal cost internet provider: Once the CapEx for the (cheap) fiber cables (in the streets, into buildings) and the distributed mini data centers (in the buildings) has been invested, maintenance and future expenses are mostly driven by software or Ethernet hardware plus CPUs. Both are basically a commodity in the age of cloud hyperscalers. This also means that network speeds evolve with Moore's law. They become both faster and cheaper over time. Software has eaten wired internet infrastructure, or as its CEO Forrest Heath phrases it:
You're able to sort of trade off the complexity of the physical world for the complexity of a digital system.
Somos is a real-world tech company because it offers a physical, software-defined service whose core offering gets better over time and has low marginal costs per customer (both enabled through software-defined service architecture and by tapping into the value network of the tech industry, hardware and networking).
SpaceX
I wrote a long piece about SpaceX and its tech company approach here. I will copy a short segment from the article, just to make my point: When it comes to access to space, low cost is the absolute north star metric to focus on in that industry. With the first successful landing of its Falcon 9 rocket in 2016, SpaceX initiated a totally disruptive cost curve trajectory in the launch industry that is now topped off by Starship. The Starship system architecture in fact specifically aims for a system-level, globally (almost) optimal cost structure. Here is how SpaceX is trying to get there, among other things:
A fully software-controlled rocket (with algorithms that get better with every flight and more real-world data. Plus this data enables virtualized development and testing)
Cheap materials (cheap to build means cheap to iterate & test)
Modularity that facilitates economies of scale (e.g. 33 engines for the booster, the booster being made of 33 identical stainless steel rings, (almost) same engines for booster and ship etc → a lot of scalable output)
Designed for manufacturing at scale → Starship factory with a high degree of vertical integration
Fully & rapidly reusable - both stages will land again (which they both already managed to do). Scalable ground operations allow for high utilization rates of the asset (think: the scale of the fuel tanks, fueling operations and logistics, stacking the next ship on top of the booster with the chopsticks etc.)
Cheap & effective fuel (It burns liquid oxygen and methane)
Software-defined R&D infrastructure that allows for fast and cheap development and testing (thanks to F9 & FH) → virtualization, HITL testing, simulation infrastructure etc.
Automated booster catch – no landing legs for the ship, which makes the ship easier to build and cheaper, because overall operations get faster (Tower = “Stage 0”). Both the booster and the tower operate the maneuver fully automated, based on real time position data
Musk started his career in software, and it shows: not only is the entire launch and landing approach software-defined and mostly automated. SpaceX operates with agile software delivery principles (fast product iterations, modularity design principles, relying heavily on virtualized and simulated test integrations to enable automated rapid feedback etc.) Obviously, Starship will not be a zero-marginal cost product, and launches will not be zero-marginal cost services. But thanks to a software-defined product and operations processes launch prices will approach a level that is basically as low as it gets for a space transportation service business: SpaceX aims to lower the costs per kg to low earth orbit by a factor of 100 or more, bringing it down to less than 100$ per kg (see image).
This will totally disrupt the nature of the industry.
Boom Supersonic
Boom designs and manufactures the spiritual successor of the Concorde: A supersonic jet that can take up to 80 passengers from New York to London or San Francisco to Tokyo in under 4 hours, for the price of a current subsonic business class seat. While the Concorde never was a viable business model, but rather a political prestige project, Boom has found a plane design that supports a viable operational business model and vice versa. And software is at the core of both the product itself as well as its design and engineering process. Says Boom CEO Blake Scholl:
So Concorde was designed half a century ago with slide rules and drafting paper and it was built out of aluminum and converted military engines and today we have everything that is in the ‘software eats aerospace’ world for digital design, digital optimization. That leads to significantly better aerodynamics. If you look at Overture, there’s hardly a straight line anywhere on the airplane.
Software helps Boom design the plane, optimize that design and pick the right materials that match the best aerodynamics. In the plane itself all functionality is accessible through software and touchscreens, which eliminates most breakers and buttons. Redundant physical controls are built in only for safety-critical functions. Overture software will be able to be updated over-the-air, so new features and improvements will arrive regularly. Airlines can control how upgrades are rolled out across fleets.
So yes: Boom designs and builds supersonic jets, but they basically build flying computers with wings and engines. They have blended the two value networks of the airlines business and the tech business and have formed something uniquely new out of it.
Amazon Logistics
Let’s consider Amazon. I’m not going to get into too many details, because Amazon is one of the best-known and best-covered companies in the world. I just want to highlight some recent developments in their logistics operations of their fulfillment business. Here is what Amazon CEO Andy Jassy said in a November 2024 earnings call about Amazon Logistics:
We continue to innovate in robotics to speed delivery, lower cost to serve, and further improve safety in our fulfillment network. We recently launched our 12th-generation fulfillment center design with the first building launching in Shreveport, Louisiana. This is the first facility that incorporates our newest robotics inventions that simplify stowing, picking, packing, and shipping processes.
Amazon’s fulfillment in their leading edge logistics centers is now completely defined by software and robotics processing (for picking, sorting and packing). This makes the fulfillment much more frictionless, faster and drives the marginal processing cost for each parcel further towards (almost) zero:
Thus far, this new design reduces fulfillment processing time by up to 25%, increases the number of items we can offer for same-day or next-day delivery and is expected to drive a 25% improvement in our cost to serve during peak, within this next generation facility.
Think about it: A fully automated picking and packing process has very low transaction costs (like searching what to pick, where to pick it etc.) and only electricity as a marginal cost. All the software R&D and CapEx has already been spent. But software (and AI) not only brings down processing costs and increases processing speed. It creates a flywheel: Amazon now sells more and more so-called “Everyday Essentials” products, like shampoo or paper towels. Such items have a rather low margin for Amazon and a higher degree of urgency from a customer perspective. So Everyday Essentials as a meaningful category only works from a business perspective if it can be delivered the same day, or the next day the latest, at a low marginal cost. Amazon now is able to fulfill Every Essentials at a profit and quickly. That has further positive effects on the business. Here is how Jassy framed that:
But we take that as a real positive, seeing the growth in everyday essential categories, which are really predicated on speed. So you have to have fast delivery to be able to sell those products to customers. And when you do, it results in a stickier consumer relationship, higher orders, building larger baskets, which help our ship economics, and repeat orders are stronger.
Customers that want to rely on Amazon for fast toilet paper or shampoo shipments need to be subscribed Amazon Prime members. And since they are already paid members, they tend to purchase other things from Amazon on a regular basis as well - which further brings down customer acquisition costs over time. For customers, Amazon Prime plus same-day delivery drastically lowers the transaction cost for such a purchase as well, since it’s basically just one click in an app away. The additional costs per item processes for Amazon? Almost negligible.
Amazon is a tech company not only because it started as an online shop where people could buy books on a website. It’s a tech company because software enables them to fully automate many of their logistics processes and operations, driving them towards a zero-marginal cost tendency, at scale. It’s a tech business because it has lowered transaction costs for its customers and internally thanks to software, which drives additional revenues. And it taps into a value network that is heavily oriented around software, AI, computation, automation plus logistics elements like warehouses, delivery infrastructure etc. This totally shifted the dynamics of both the retail and logistics industry.