NILG.AI

There’s a common pattern in artificial intelligence (AI) where large corporations build massive infrastructures to support their AI use cases. The goal is to make quick predictions and constantly update with new data to scale up your infrastructure. However, this approach often overlooks the trade-off between infrastructure cost and the size of the opportunity that the use case is solving.

In this article, we’ll discuss a different strategy called “Old but Gold” predictions. This approach can help you maintain a proper balance between predictive accuracy and infrastructure cost, thus minimizing expenses.

What are “Old but Gold” Predictions?

The “Old but Gold” pattern involves calculating your predictions beforehand and saving them in a table. For example, if you know that you’ll need to make predictions about your customer base tomorrow, you can figure them out ahead of time and store them. This way, when someone asks for a prediction, you already have it prepared and waiting in your table.

Even if these predictions are slightly outdated, they can still be relevant. The pattern then involves incrementally updating these predictions at a pace that is most likely lower than the speed of the predictions you have to generate. This approach allows you to keep a smaller infrastructure, deal with older predictions, and decrease the time it takes to get these predictions.

How Can You Update These Predictions?

There are different strategies for updating these predictions. One approach is to update the oldest prediction first. Another is to prioritize new customers or cases that have not yet received a prediction. A third approach is to update the predictions for the most active cases, i.e., those with a high prediction frequency.

This last approach is particularly useful as it allows you to have relevant predictions for the 20% of your customer base that brings the 80% of your value while also keeping the rest updated in case they need new predictions.

When Can You Use This Strategy?

This strategy is particularly useful in two scenarios:

1. Stable Domains: You can use this strategy on slow-changing domains such as real estate or insurance prediction. In these domains, clients don’t change their minds constantly. So, for example, if you predict the probability of insuring a client next month, it’s probably going to be the same month.
2. Non-critical tasks: You can use this approach in situations where an outdated prediction seems more relevant than no prediction. For example, if you’re sending a recommendation in a newsletter, you might be okay with a prediction generated two or three days ago. But on your website, where you want to account for the session data, you might want real-time predictions.

The Trade-off Between Infrastructure Cost and Predictive Power

As you increase the size of your infrastructure, you reduce the lag of your predictions, making them more up-to-date and relevant. In other words, you increase your predictive power. However, this also increases the cost of your infrastructure. So, you need to balance where your predictions remain valuable enough to make more money than the infrastructure costs.

This strategy allows you to start with a very small infrastructure with delayed predictions and then increase its size as you earn revenue. It’s a smarter move than investing a lot of money on huge distributed systems for a use case that may not even recover the investment in its infrastructure.

Conclusion

The “Old but Gold” predictions strategy is a great way to build AI use cases with minimal risk. Instead of building huge, expensive systems, this method focuses on making predictions in advance and updating them as needed.  It allows you to start small, scale gradually, and make the most of your resources.

If you’re interested in learning more about computing the size of an AI opportunity, book a meeting with us at NILG.AI. For additional information, be sure to check out our Data Ignite Course or download our ebook for a more business-centric approach to AI.