Say a kid in your neighborhood, Sam, starts a lemonade stand. Sam not only wants to share their love of lemons and sugar with the world, they want to leverage the latest in technology to make it a wild success. Having studied a variety of AI techniques in a very technologically advanced elementary school, Sam is ready to put them into practice.
As the lemonade stand gets going, Sam might use time-series analysis to understand, based on how many orders they got in the last week, how many orders to expect the next week. Or perhaps Sam wants to forecast how likely the lemonade stand is to be successful next summer – they could use some time-series data around costs and weather forecasting to make a future time prediction. These sorts of analyses are all very time-specific.
Conversely, Sam may turn to A/B testing to compare the impact of different factors. If the lemonade is presented in a paper cup versus a plastic cup, can they charge more for it? Or if two versions of the lemonade are the same but one has 10% more sugar, which one do customers like better? Two similar scenarios with slight differences are examined in A/B testing to understand which produces more of the desired outcome.
Now, let’s say Sam’s lemonade stand really starts to take off, and Grandma invests $1,000 to make it even more successful. Optimization would help Sam to understand how to make the most of that money. Do they spend it on better quality ingredients? If so, should they increase the quality of the lemons, or the sugar? Do they get higher-quality cups? Optimization looks at how to allocate scarce resources to reach an improved outcome, which requires being very clear about what the goal is. Does Sam want more customers? More consistent customers? Larger sales per customer? And so on.
A recommendation system could help Sam suggest the best-fit lemonade to each customer, or perhaps the next internal action for a person on the lemonade-stand team. If Sam offers a hundred varieties of lemonade, and they’ve learned that kids like one kind and those looking to lose weight like another kind, then it certainly would be helpful to point folks who match those profiles to the lemonade they’re likely to prefer. But also internally, if Sam has hired their cousin to take over for a day, a recommendation system can suggest the next action to take to keep the lemonade stand running smoothly. Recommendation systems help with identifying the most effective choice when there are a lot of potential options. This requires knowing what those options are and being able to measure the impact of each in different contexts.
Computer vision is all about making computers visually understand things. Perhaps this lemonade stand exists in some dystopian world in which Sam has a camera capturing customers’ facial expressions to detect whether they like the lemonade or not. Or Sam may use computer vision on the back end – a system to spot the juiciest lemons at a store, allowing Sam to spend less money to get a higher-quality product. Or more creatively, maybe Sam mounts a camera in the neighborhood and happens to see that runners often go by the park on the corner, indicating that moving the lemonade stand there may be a good idea. In that case, Sam isn’t using computer vision to drive day-to-day operations – it’s more about understanding the landscape broadly to best position their operations.
Natural Language Processing (NLP) is similar to computer vision, but is focused on understanding text. Once Sam’s lemonade stand has a bunch of reviews, NLP can be used to understand if they’re overall positive ones, or detect a change in tone over time that indicates declining satisfaction. If there’s suddenly a spate of reviews, NLP can identify which are likely from bots versus real people. Or Sam could go look at the menus for 20 other lemonade stands in the city and use NLP to try to understand if they’re all offering similar drinks, even if they use slightly different words, and use that information in approaching future menus and drink offerings.
Lastly, because Sam is up-to-date with the latest in AI, they might leverage generative AI, which actually intersects with a lot of the aforementioned techniques. Generative AI is not just about understanding the world, but also contributes back, creating new content. Sam could generate new mouthwatering lemonade images to lure in customers (GenAI meets computer vision), or they could have a large language model (LLM) write up a new menu (GenAI meets NLP). Possibilities for new content using LLMs and other GenAI models are wide open.
Your business probably has more moving parts than a neighborhood lemonade stand. But just like success in business often comes down to fundamentals, these AI building blocks are the core ingredients powering most modern business applications – from simple automation to complex decision-making systems. The key is knowing which recipe will help your business thrive.