Back when most companies were still trying to get a handle on web-based commerce, smartphones caused significant disruption. Businesses were motivated to meet their customers where they were, on their phones and tablets. Companies that simply made their existing websites mobile compatible struggled to realize mobile’s true potential. Firms that went back to the drawing board and redesigned the customer experience with a native mobile app were able to compete effectively and win. Others had to follow suit. Those that didn’t, suffered massive losses in customer engagement as well as revenues.


We’re in a similar place with AI. We have a choice: Treat AI as a feature by looking at the way we do business now and simply adding AI to parts of it, or treat AI as architecture by realizing its true potential and reimagining methods and decisions. We believe in seeing AI as architecture that can be leveraged to reengineer how business is done across the enterprise. (This feature vs. architecture question originally surfaced in a conversation between Marc Andreessen and Kevin Kelly on startups and the future of technology.)


Here’s an example to illustrate: A marketer wishes to anticipate a customer need, in order to continually suggest offerings that fulfill this need.

A feature approach (illustrated in the figure above) might involve leveraging AI to enhance segmentation and develop dynamic and perhaps micro segments. One can then manually match actions to these new segments or in more sophisticated environments leverage a recommendation engine. In many cases, these two efforts (creating the segments and matching the actions) are largely disconnected. At times, legacy delivery mechanisms such as CRM present challenges in what can be delivered to customers. Ensuring that we get feedback from the customer on whether they liked the offer or if it prompted them to act are at times an afterthought and not planned for.


Keep in mind that this approach of increasing sophistication on individual elements of an existing process has worked for many organizations for many years. Advances in technology, data and algorithms show us that there is a better way.

If we were to take an AI-as-architecture approach to solving this marketing problem, the extent to which AI could deliver value might be surprising. As the figure above illustrates, such a solution could include:

  • Automated data pipelines, managed by AI, that continuously label, cleanse, search and integrate new data at a customer level on a real-time basis;
  • No limitation on the number of segments you can create;
  • An always-on system that generates customer-level insights and recommendations;
  • Real-time micro experiments that drive new experiences for and feedback from customers;
  • Message and offer pipelines delivered to customers through their preferred channels;
  • Real-time feedback loops from customer interactions as they happen;

All of these and likely more could be running continuously, every minute, all day, every day and would provide marketers with a clear sense of the opportunities they could pursue at any given time.


We strongly believe that the cost of an AI-as-architecture approach will be lower than a feature approach. Many processes and manual interventions will be fully automated or eliminated with an architecture approach, which will dramatically reduce costs. But it requires that we fundamentally rethink how such business problems are being solved across the enterprise and redesign existing approaches. As more of this work happens, it will generate assets such as automatic data cleaning and labeling that can be leveraged across the organization.


An AI-as-a-feature approach is a viable place to start. But an organization must pivot quickly to an architecture approach. Those that fail to do this will not realize AI’s full benefits and will suffer the kind of competitive disadvantages that plagued firms that were late to mobile. It’s time to reframe your approach to AI innovation or risk being left behind.