How to create defensible competitive advantage in organisations driven by artificial intelligence

With the buzz that AI is generating today across most organisations and business sectors, it’s important to understand what we’re really talking about.

Generally, there are two types of organisations in this field:

1. Tech organisations that are furthering the field of AI and creating tools that others can use
2. All other organisations that are applying AI to an industry sector to create a competitive advantage, either through better products and services, customer targeting, or operational excellence.

The second group makes up the majority of organisations engaging with AI, and range from large organisations to start-ups such as in FinTech. This is the group I’ll be focussing on here.

Creating artificially intelligent programmes and applications is expensive. Prohibitively expensive in some cases and certainly generally expensive for large parts of the economy.

The cost is driven principally by three things:

• AI experts that can create and write the programmes are very hard to find and can command high salaries. 
• The volume of data required to take advantage of state-of-the-art AI algorithms is extremely large, gathering these data sets is not easy.
• The computational power required to train and in certain cases run AI programmes can be significant, although most of these organisations deploy AI on the edge once algorithms are trained. 

So how can these businesses make commercial sense out of developing AI? And how can they create valuable and defensible competitive advantage (and in some cases intellectual property) through the use of AI?

For the most part, the answer lies in the data being programmed, rather than the AI programming itself as the AI tools themselves are often made freely available.

The AI developer community is increasingly working on open source principles. This is great for the development of the technology, bad for companies that are trying to develop competitive advantage solely based on algorithms.

If a company has proprietary data on top of which it builds an AI programme to interpret the data, then the company’s AI-enabled proposition has defensible value.

The AI algorithm trained on this data in this scenario cannot be grabbed for free on open source platforms or with third-party tool providers and replicated in other businesses, because other businesses don’t have access to the same data.

Sometimes the data doesn’t have to be proprietary, it can just be really, really hard to gather. Gathering significant volumes of data can erect a barrier to entry. However, this requires significant investment.

If, for example, you wanted to write an AI programme to automatically identify the building materials of a given structure solely from photographic images, you’d first need to take photos of hundreds of thousands of structures and map those photos to the relevant building materials. The vast majority of the workload in building this AI programme would go into gathering and organising the data (assuming this data doesn’t already exist elsewhere). It’s not proprietary data, but the scale of effort required to get the data adds value to your AI proposition.

Where the data is proprietary, any AI programme you develop will be even more defensible and therefore much more valuable.

Naturally, the AI programme itself needs to be interpreting the data in such a way that its outputs also have intrinsic business value. But the long-term value of that proposition from an investor perspective is tied up with the nature of the data it is analysing.

Investors and venture capitalists are still getting to grips with how to value AI driven businesses. However, most have begun to recognise the importance of defensible competitive advantage and IP; and this is most commonly created by the existence or gathering of proprietary data.

Of course, if you are a tech firm such as Google and have the budgets there are other ways to build defensible IP in artificial intelligence, just look at the work Google DeepMind is doing, which for some applications uses readily available data (albeit lots of it), and through clever use of research and novel model architectures, has been able to create something truly spectacular.

The challenge is that this kind of development requires deep pockets and patience.

For most, the real value in AI can only be derived through proprietary data. So don’t go building AI algorithms or buying AI black boxes off the shelf; think first about your existing data or how you will collect proprietary data.