When Standalone AI Apps Meet Existing Systems of Record
More of the AI-powered applications I'm seeing have a new challenge - working with core systems of record that weren't designed for AI applications and services
In January 2023, I wrote a post about my questions and concerns about investing in AI-powered applications. While we have invested in AI-powered applications and will continue to do so, I remain cautious about how much of the economic opportunity is available to early-stage startups. The big, incumbent tech companies have been early, active participants in this ecosystem. It feels that much of the surplus value created by AI-powered software will flow to the software buyers, not the firms producing that software. If you’re a software buyer, this is good news; the software you use will get better and more powerful for a modest potential price increase.
Many of the AI applications I’ve seen in the past year were largely standalone; they were built to perform AI-enabled actions, such as assessing candidates for an open job role, responding to customer service issues and requests, or creating content for future marketing and communications campaigns. These systems were not asked to interface with other parts of the company’s workflow, as the first-order question was to see whether these applications could produce useful, actionable output. The question of integration would come later.
In my conversations with founders who have built and sold successful AI-powered applications, the question of integration has come to the forefront. These standalone applications that excelled on their own now have to connect to existing workflows and systems of record that weren’t designed to work with AI-powered apps and services. This is particularly true of AI-powered products that need to plug into workflows that touch core systems of record, such as:
HR solutions that now need to interface with human resource information systems (HRIS) and applicant tracking systems (ATS)
Finance solutions that need to communicate with inventory management, enterprise resource planning platforms, or core accounting solutions
Customer service and sales products that need to read and write data to and from customer relationship management (CRM) and sales automation systems.
There is a world where enterprises could decide to redesign their workflows and systems around these AI-powered products and services. I have not seen much evidence that this is happening at scale yet; for the most part, these new products will have to play nicely with the existing infrastructure, even if that infrastructure wasn’t designed to work with these new products.
In past software cycles, the ability to connect with core systems and interface with the other tools and tech an organization used was one way software vendors competed for customers and differentiated their offerings. In some cases, those integrations were table stakes for deployment, and you couldn’t get in the door without them. I am curious to see how this plays out for this new wave of apps and services, and this need to integrate and interface with existing infrastructure creates room for more differentiation and defensibility.
Well said…and exactly how we are seeing the use of AI in our work. First, build the system of record, then an AI-first API, test our own models based on pressing client needs, open to other external products that are doing great things/adding value for our clients.
This is one of the last mile problems of AI. It’s gonna be hard