You can’t have a conversation about business or technology without the term Artificial Intelligence creeping into the dialog. I work with large financial institutions and they all have teams working to figure out how AI can be used as a differentiator. These initiatives tend to fall into two broad categories: big data infrastructure and machine learning research.
There is the potential for these initiatives to miss the forest for the trees. AI (the data science powered kind vs. the Ray Kurtzweil singularity kind) is not new. It has been used for decades to make data informed predictions in finance, insurance, manufacturing and retail. The teams that created these successful early applications did not start by thinking about data, or even about predictive analytic technologies. Instead, they started with a pressing business problem that needed to be solved. AI initiatives that start with clarity about a prioritized business need will deliver value faster than generalized AI initiatives.
AI teams must learn this agile principal: if a customer doesn’t need it, then don’t build it.
When a team starts with clarity about their business goal a few positive things happen. The most important is that business stakeholders (or ‘customers’ in agile terms) are involved in defining the AI objectives from the beginning. In the software development world, we’ve learned that focusing on customer needs first is the most effective way to create solutions. If a customer doesn’t want a feature, don’t build it. AI projects need to learn this lesson as well. When AI teams stay focused on solving the customer’s problems, they can discover all sorts of things they don’t need to do. The result is they often choose to leverage data and analytic techniques that solve the problem the simplest way possible. This reduces unnecessary investment, avoids rework and results in an easier to maintain solution.
What’s powerful about a business solution focused AI project is that the team considers from the beginning how operational actions will be taken based on analytic insights. By thinking of your project’s primary goal as delivering a functioning business application, teams tend to solve the challenges around collecting data and deploying the resulting analytics early in the project. This can dramatically reduce the cost and delays associated with putting AI into production and can make it easier to collect the new data and insights that are needed to improve solution accuracy over time.
Incorporating human intelligence feedback loops into AI applications will optimize learning.
Finally, a solution approach to AI allows teams to build tools for human decisions and feedback from the start. For the foreseeable future, human intelligence will continue to outperform artificial intelligence when presented with new patterns of information. Many applications will continue to require human input for handling outlier situations, and it is important that AI applications be designed to consider human input as it learns. By thinking of your AI project as an end-to-end application that includes human interfaces from the beginning, it is easier to deliver a system designed to learn over time.
A shift in focus from a data-first to a solution-first approach to developing AI applications will have a big impact on the investments organizations are making in skills and technologies. As an industry, we have experienced a surge in investment in data science skills on the business side, and data management expertise in IT. As low-code AI application tools mature there will be a shift back towards prioritizing business expertise over these technical skills. When that happens, the value delivered by AI will finally begin to accelerate.
Stop focusing on the data and technology and start focusing on the problem to be solved. That’s Agile AI.
The AI industry is beginning to mature and will soon demand a return on investment. Teams responsible for implementing their organization’s analytics strategy have an opportunity to learn from lessons our industry has learned the hard way. We now know, for example, that building a huge data warehouse in anticipation of delivering some unidentified future value is a bad idea. Now teams collect data in low effort lakes and extract what they need when they have a real business requirement. Software teams have learned to build minimum viable products (MVPs) based on the most important business needs instead of running death-march software projects. AI teams need to learn from these examples. Stop focusing on the data and technology and start focusing on the problem to be solved. That’s Agile AI.