Sprinkles and cherries

There are two ways to embellish a cake: scatter a handful of sprinkles or one or two cherries. Similarly, when integrating AI into products, you can apply AI throughout the product or focus them on one or two key subcomponents.

For large, complex systems like Facebook Ads, Airbnb, or YouTube, which are already finely tuned, there are seldom quick, easy wins. One needs to enhance many underlying components with advanced algorithms to improve effectiveness and efficiency. This is akin to adding sprinkles: each small improvement contributes to a greater whole. Fifty 0.2% improvements adds up, and if the baseline is already counted in millions (of users to dollars), the large investment pays off.

For smaller or less optimized products, it’s better to narrow down to one or two subcomponents that lag behind – candidates for a transformative “cherry”' addition. This can be moving from heuristic to data-driven approaches, upgrading from basic to advanced ML models (e.g. LLMs), or by combining strategies, e.g. leveraging both LLMs and specific datasets, leading to substantial, non-incremental leaps in performance.

The first critical step is to figure out whether your product needs “sprinkles” or a “cherry”. Then, identify precisely where, and finally, decide on the potential solutions. So what about your cake (=product)?