Planning for a new AI initiative? Start with the right AI development culture
So, you have committed to building an AI company, designed your own competitive advantage, and created a good AI and product strategy. Then comes the implementation. Beware: certain facts about AI systems are counterintuitive to conventional engineering wisdom.
In his book, Good Strategy / Bad Strategy, Richard Rumelt, a professor at the UCLA school of business, describes an interview he conducted in 1996 with Jean-Bernard Lévy, the Chairman & CEO of Matra Communications, a French telecommunication hardware company. Those were the golden years of telcos and the advent of the Internet era. The companies invested heavily in fiber optics, and the promise of network bandwidth as the blockbuster product of the digital age was still alive and well. Professor Rumelt and monsieur Lévy discussed the phenomenon of Cisco Systems, a then-recent startup taking the telecommunication hardware industry by storm. While analysts predicted the global telecommunication equipment market to become a battleground between heavyweights AT&T and IBM, Cisco Systems emerged out of the blue and apparently was unstoppable in capturing this new, red-hot niche.
“I am trying to understand the forces that are changing the industry structure, said professor Rumelt. — For instance, look at the amazing success of Cisco Systems. As you said, scale is the critical barrier to becoming a major player in telecommunications equipment. Yet, Cisco Systems has broken right through this scale barrier. It has grabbed the internetworking equipment market from under the nose of giants like IBM, AT&T, Alcatel, NEC, and Siemens. And Matra. How has that happened?”
“We have had Matra engineers working on inter-networking equipment. – replied monsieur Lévy. The basic principles are well understood. Yet we cannot replicate the performance of Cisco’s multi-protocol network routers. The heart of the Cisco router is firmware — software burned into read-only memory or implemented in programmable arrays. Cisco’s product embodies, perhaps, one hundred thousand lines of code that is very skillfully written. It was created by a very small team — maybe two to five people. That chunk of very clever code gives the product its edge.”
Since then, mainstream engineering has changed beyond recognition. Cloud technologies made computing power readily available on demand, and the storage price decreased dramatically. In turn, it lifted the need for nitty engineering design and led to thousands of handy and easy-to-use engineering tools, frameworks, and technologies. Apart from the few niches, today, we expect engineers to implement product features as quickly and cheaply as possible and encourage them to reuse as many existing solutions as they can. We do that directly as staff engineers by laying the system designs and architectures and indirectly, as managers, by measuring and incentivizing the engineering team’s velocity. In many ways, this new engineering culture created the Internet as we know it today. The flip side, though, is that replicating an engineering solution is much easier today than it was in 1996.
Building an AI competitive advantage, though requiring engineering, rests on very different premises.
1. Implementing your AI models with a limited amount of data should be hard, and the initial version quality should be inferior.
The crucial characteristic of competitive advantage is that replicating it should be hard. Ideally, impossible without access to your data. Collecting the data should be hard, too. Ideally, impossible without your or a similar product.
This premise leads to the conclusion that, given the best effort of the engineering team, implementing your AI models with a limited amount of data should be hard, and the initial version quality should be inferior.
If you implement a product feature fast and cheaply, it is a sign of great engineering. If you implement an AI solution with ease and little available data and get a widely appealing product of high quality, it is a red flag. It means a competitor can implement a similarly appealing solution at the same pace and take away a part of your market share.
2. Your AI models should not be easily transferable across domains.
Next, you want your product features to be as widely applicable as possible. If your product team comes up with an easily scalable solution across markets, it is a sign of excellence. On the contrary, if your AI models trained on the retail domain data are easily transferable to financial services or other domains, it is a red flag. If you can build a viable AI solution for the financial services domain without accessing a lot of financial services domain data, a competitor can build their solution on yet another domain and, with the same ease, bring it to the financial services, too, threatening your market share.
It is a good sign if your solution works well for one domain but underperforms for another without injection of the new domain data. Although it means expanding into a new domain will be time-consuming and costly, you will have a defensible position against competitors there.
3. The expected future improvements should primarily come from the increase in data.
You expect your product team to come up with better algorithms that, in turn, make your product faster, cheaper, and easier to use. Apart from fixing bugs, engineering teams work on designing, implementing, and testing new systems and algorithms that improve product quality.
For AI competitive advantage, your leverage should be data, not the algorithm. You may invest in researching and building better machine-learning algorithms for your market niche. Still, if you expect the future value of the product to be driven primarily by the algorithm improvement and not by the increase in the relevant data, it is a red flag. Your algorithms could be replicated by a similarly smart team of AI engineers and researchers unless they heavily rely on the variety and volume of data in your possession.
4. AI business has the fastest, not the first-mover, advantage.
Some industry structures ensure that the first company that brings a product to the market gets a formidable competitive edge. For example, in pharmaceuticals, your small molecule or biologics formula may be protected by a patent that gives you a 20-years monopoly. In this case, it is essential to develop the solution and file the patent first to ensure the competitors will not be able to steal its design and dilute your revenue.
The AI software, though, works differently. To win the market, the company should accumulate a huge amount of the right data — not just any data, but the data your machine learning models can benefit from. Being first in the market without ensuring the proper AI, product, and marketing design does not mean much. It only exposes the market niche to potential competitors and leaves your company vulnerable to the latecomers who get things right.
A shared understanding of these differences by leadership, product and engineering teams is essential for success. Misunderstanding the difference between AI and product development and applying regular product development patterns to AI initiatives would set them up for failure.
Some time ago, I worked with an enterprise software company. Co-founder James (not a real name) had a powerful product vision and an aspiration of turning his startup into an AI company. James did not have any experience with AI in his previous career. Yet, he assumed the chief product officer role for a complex machine learning-based product positioned as the company’s new flagship.
James believed in the first-mover advantage and wanted to bring the product to the market as soon as possible. The development was held in extreme secrecy. The product team was not allowed to share what they were working on, even internally. As you see now, it violates point four: AI products need not be the first on the market to succeed but must be properly designed. The secrecy and absence of cross-functional work undermined the chances of getting the AI and product design right.
Next, James personally put together a very aggressive product development roadmap, assuming the team would deliver a superior quality product in a short amount of time, having little concern about the available data at hand. It obviously violates points one and two: the quality of the product should come from data. And even if there is a way to build an exceptional AI product in just six months, it would have a limited value anyway, as competitors would replicate it with the same ease.
As the release date approached, James launched a powerful marketing campaign promising exceptional product quality and wide applicability across markets. We can see now that it violates points one and two: the initial product version is likely to be inferior even for a single market and unlikely to be transferable across markets at all.
Delivering a high-quality and domain-scalable machine-learning-based solution using a small dataset was a contradiction in terms obvious for any AI engineer or strategist. Although the company did release the product, the initial model quality was way below the marketing promise. As a result, the board replaced James with a professional manager. The company shut the product down and turned to another niche. Today the market is dominated by a competitor who launched a similar product after watching James’ attempt fail.
This blog is dedicated to AI competitive advantage, and we are doing our best to explain how it works and how you can build one for your product or company. You can check our other posts to get an extensive explanation of what the network effect is and how AI enables it, how to build an AI competitive advantage for your company, what culture helps you build the right AI products, what to avoid in your AI strategy and execution, and more.
If you need help in building an AI competitive advantage for your business, look no further. Our team of business experts and AI technology consultants has years of experience in helping technology companies like yours build sustainable competitive advantages through AI technology. From data collection to algorithm development, we can help you stay ahead of the competition and secure your market share for years to come.
Contact us today to learn more about our AI technology consulting offering.
If you want to keep posted on how to build a sustainable competitive advantage with AI technologies, please subscribe to our blog post updates below.