Why did Blackberry's market share go to zero?
They went from perfect product-market fit to zero market share in less than a decade. What can we learn from this as AI transforms technology now?
Blackberry was a product of its time, specifically the economic constraints of the infrastructure, technology, and consumer demand. These constraints created Blackberry's success — and later doomed it.
Blackberry became a success because they found a way to circumvent the limited bandwidth of mobile networks. Inventing a server-client architecture instead of a single channel per user disrupted the economic model with unlimited messaging vs. pay-per-message texts. This empowered and freed business users to work from anywhere.
In many ways, the story of SaaS is similar. SaaS became successful by unlocking the value of the server-client model over the cloud of applications instead of expensive on-premise installations. SaaS also changed the economic model by introducing recurring subscriptions vs. upfront licenses, empowering a new generation of businesses to leverage software more easily.
The downfall of Blackberry was, in many ways, out of their control; their organization and business model were a product of the wider constraints at the time. It echoes Conway's law, which states,
“ Organizations which design systems… produce designs which are copies of the communication structures of these organizations."
So when Steve Jobs introduced the iPhone, they could only understand it through the lens of their product; they saw a Blackberry without a keyboard, not something completely new. They missed that Steve Jobs had hardballed the mobile networks to get massive concessions to high bandwidth data at lower costs, had direct-to-consumer distribution, and, most importantly, was not selling a “phone” with email but a computer in the pocket.
Apple did this because their organization approached a new market as a bottom-up design exercise to deliver the best possible user experience, not just circumvent current limitations or slightly improve the previous economic model. Conway's law, in full effect, allowed them to “think differently”.
We are at a similar moment today in software. SaaS applications are mostly an evolution of what was before, disrupting by optimizing within the constraints of the architecture of existing enterprise software, business, and organizational models. This also means that as AI becomes a reality, they see it through the lens of their products today. They are prisoners of Conway's law.
The answer so far has been copilots and AI chatbots as an addition to existing applications. This is like Blackberry's thinking that to compete with Apple, all they had to do was remove the keyboard. By the time Blackberry came out with a phone with no keyboard Apple had put a full computer in its customers' pockets.
Just like Blackberry we might look back at legacy SaaS in 10 years and consider it a transitional technology.
Like the iPhone, the current AI revolution is a result of breakthroughs in distribution, economic models, and chips. Had we not had a huge surplus in GPUs from the crypto bubble, it would never have been economically viable to build the current generation of foundation models. Just like if we had not had the huge surplus of fiber and internet backbone from the dotcom bubble, we would not have had the iPhone a decade later.
With SaaS, a huge leap was the multi-tenancy of data models allowing many customers to share the same database (just like Blackberry and their server-client model). With AI, the new technology stack involves vector databases, RAG, MLops, model routers, and much more, which, put together, are completely new paradigms for how to structure software.
Further, with less focus on the classic SQL databases, single-tenancy might also be the better model as it makes it much easier to manage data privacy and proprietary learning loops for customers. These differences are also at the heart of the problems combining legacy SaaS applications with new AI-first architectures.
Copilots are a hybrid between two incompatible models, the AI-first interface and the legacy SaaS application architecture. Copilots are not good enough to replace the user interface; neither do they easily work with data across applications nor allow the user to change workflows or how work is done. The reason is that all the data, front-end architecture and security models of existing SaaS applications represent a huge sunk cost and is structured in a way that is hard for the new AI architecture to interface with.
The legacy SaaS application architecture is in the way of unlocking AI's value for most users.
This is why I think that copilots are ultimately an evolutionary dead-end (and lagging user retention seems to confirm this already), and we need a new paradigm. While this new paradigm is emerging I think it is hard to say what it ultimately will look like, but I do think a few things are clear, it's going to be almost impossible for legacy SaaS players to invent it, given Conway's law.
It will take far more than technology— you will have to design your organization, distribution model and user-experience from first principles to be true AI-first.
At Beyond Work we have been trying to live these lessons by shaping our organizational principles from the beginning to support a different way of building and delivering software. By changing the distribution and economic incentive models, and most importantly by learning as fast as possible from real enterprise customers what their experiences and expectations are as they get their hands on something completely different from the previous generation of software and how you solve work.
So far the feedback has been amazing, and it's been truly energizing to see how different the user expectations are for the AI-first enterprise software, just like with the iPhone most users get it intuitively and want more.
First published on LinkedIn, 18 February 2024.