Lean AI

Chapter 12. How to Manage Complexity

There will always be complexity with any AI project because of the codependency on different technologies and coordination required of cross-functional teams necessary to be successful. In Part II, we covered in detail the concept of Customer Acquisition 3.0, which showcases the core dependencies required to get the consistent flow of good, clean first-party customer data to Athena Prime—our “intelligent machine” from Nectar9—which enables the AI to perform to its fullest potential.

Our challenge at IMVU was very similar to most startup growth teams. Imagine the complexity that goes into managing all the different user acquisition experiments and campaigns. First, you need to think through your overall strategy. You need to take into consideration setting up and managing all the different experiments across different platforms and campaigns with the right budgets, bids, goals, audiences, landing pages, and creative assets that need to be monitored and adjusted depending on the results.

There is a practical limit to how many of these experiments humans can run and make sense of. In addition, human-driven experimentation is typically a sequential endeavor—you design your experiments, you run them for a period of days, weeks, or months, you consider the data, make adjustments, then start the process over again. Not only does this lead to longer feedback cycles, but the more experiments you conduct simultaneously leads to exponential complexity ...