AI and Helmer's 7 Powers
Artificial Intelligence has made its way past the hype and into a lot of products. I've been working on AI in healthcare since 2016 as a founder of Maxwell Plus. I've seen the good, the bad and the ugly but still believe that AI will speed up human progress for the better. About a year ago, I started writing about AI in the context of the 7 Powers framework. I changed writing platforms and never finished the series. This post is my condensed and completed summary of how AI plays into the 7 Powers framework. I am writing this assuming you're familiar with 7 Powers. If you aren't check out this post from NFX.
Before getting into each of the powers I am working with some assumptions here. These shouldn't be hard to align with but rule out some of the low hanging rebuttals to these arguments.
- The AI I am talking about here is narrow AI. AGI is a topic for another day.
- The team designing and implementing the AI is aware of potential bias and regulates updates to the model.
- The business model used to package this AI captures both the raw data and the final gold standard label, allowing for continuous learning.
- User's of the described AI can opt-in or out of use. Most of these work without this being true but it paints a darker picture.
Power 1 - Scale Economies
Once AI is trained and a pipeline for continuous learning exists, a moat starts to grow. The more users (and data) an AI company gains, the larger that moat grows. Beyond this, having large pools of data can enable the expansion of capabilities. For example, moving from diagnostic AI in healthcare to adding prognostic capabilities. For this to work assumptions two and three above are critical.
Power 2 - Network Economies
With a business model that allows closed-loop learning AI will improve over time. Some applications act on cohorts as well as individuals. In healthcare, for example, the more data the AI has about your relatives, the more it can deliver for you.
This power remains valid even if raw data is portable. Most AI created secondary data through its use. This data and the weights of the model are not released even if the user and their data move to another provider.
Power 3 - Counter Positioning
Many applications of AI are a replacement for work done by humans. This is particularly true in healthcare where AI replaces clinical decision-making. We're not at a stage of total replacement but, AI applications could cannibalise the existing business models.
Beyond the initial disruption, the move from human cognitive labour to AI means that the barrier to future switching is lower. An AI company can spin up new services with a lower concern for cannibalisation.
Power 4 - Switching Costs
Many high-value AI systems operate with a local context. By that, I mean that they have some fine-tuning to the individual. Your video recommendations on youtube are based on your viewing history. AI companies may allow you to export raw data and take it to another service but the second layer data is likely kept. As a result, new systems, AI-driven or otherwise need time to learn your nuance. For the end-user, this may be enough to keep users on the platform. The more AI can individualise, the more it increases switching costs.
Power 5 - Branding
AI is not a significant player here. There is likely some ability to stand out as the company using AI but that is a short-lived branding boost. Long term, AI may have the ability to deliver higher customer satisfaction and a larger sense of quality. Google's search recommendations have the perception of being the most accurate but AI alone didn't build the brand.
Power 6 - Cornered Resource
Most AI companies will lock up either some data or some labels if AI is a core part of their business. In some cases, raw data can still be released. To give an example here medical scans can be released if the user leaves but, the annotations on those scans may remain internal. This gives the original company some cornered resource.
Power 7 - Process Power
The learning loop of AI means that any gains from a better process can be more rapidly realised. Large organisations dependent on human input can take a colossal effort to change. AI-based companies can loop faster.
Any company running AI this way will still need a process to update and control changes but these cycles happen quickly. A well-organised company using AI can develop systems to train, report on and update existing models in the company. In some cases, this may need additional regulatory oversight but this too can be rapid.
AI isn't magic. It can give serious competitive advantages over non-AI solutions. AI will likely continue to improve in the future. Today's bottlenecks of data and clean labels are going to be different to those in the future. In a lot of ways, you could rewrite this post, replacing the terms AI with "more staff" and see many similar gains. The difference of course is that AI can be scaled at a significantly lower cost than a human.