AI basics that every merchandise planning team should know

Even if you’re a data scientist, understanding AI can be tough. We’ve pulled together a collection of the jargon, buzzwords, and slang that defines AI discourse, curated for the merchandise planning community.

What are the AI basics that planning teams should know?

As AI becomes more commonplace in fashion, planning teams shouldn’t be caught off guard. Here are some of the common terms you might hear or see.

Accuracy: One of three common ways to check how well an ML model is working. Looks at how well the model finds true positive and true negative answers.

Algorithm: Detailed instructions on how to complete a specific task, written in a way that computers can follow. The simple building blocks from which AI grows.

Artificial intelligence (AI): A way to describe machines that can do something, well, intelligent. It’s also the name of the field that focuses on giving machines this ability.

Backpropagation: An algorithmic approach to training neural networks. Included here mostly so you can see how out there some of the terms can get.

Data: The pieces of information that we gather when we want to answer a question. Structured data is what you would find in a spreadsheet — e.g. historical sales data. Unstructured data could be text from the web, audio snippets from customer interviews, video feeds from security cameras, images scraped from social media, etc.

Data mining: Digging through data to find patterns. An example of a task that AI is good at completing.

Decision tree: A supervised model that uses a series of if/then statements to make predictions. Commonly visualized using flowcharts that look an awful lot like tree roots.

Deep learning: A field of study that focuses on artificial neural networks. May also refer to “what it feels like it takes to understand AI.”

Ensemble method: Combining multiple algorithms to form a stronger model. Many common forecasting models use this approach.

Error analysis: A statistical tool to help answer the question: “How good is my model at predicting new outcomes?”

Generative AI: Rather than simply making predictions about unseen data sets, Generative AI generates new outputs based on those data sets. The hotness at the moment.

Interpretability: A way of describing how easy/hard it is for a human to understand how a model got to an answer. Can help to think of it as a “black box” — lower interpretability = a more opaque box.

Linear model: A simple supervised model that uses statistical methods like linear regression to make predictions.

Machine learning (ML): Models designed to first learn from known data sets, then make predictions about unseen data sets. ML approaches differ from other AI techniques because they don’t need to be programmed — they can learn!

Overfitting: A way to describe when a supervised model is great at predicting training data but not so great at predicting new data.

Parameters: The instructions and rules that make up a model. They get tweaked automatically by the model as it learns. Not to be confused with hyperparameters, which are the human-provided set of instructions and rules that guide the model from a top-down perspective.

Sensitivity: One of three common ways to check how well an ML model is working. Looks at how well the model finds true positive answers.

Specificity: One of three common ways to check how well an ML model is working. Looks at how well the model finds true negative answers.

Supervision: Training can either be supervised or unsupervised. Supervised training uses labeled data with known inputs and outputs. Unsupervised training used known inputs only — useful when we want the model to identify patterns we can’t see.

Time series: A way to describe data that measures one or more variables at multiple poins in time. Sales data from a POS is a good example.

Training: How ML models develop their abilities. Consists of four general steps: getting the data, defining a task, making the model practice a bunch of times, and providing feedback.

Variables: The pieces of data we think might have an impact on whatever outcome we’re taking a look at. For example, if we’re predicting demand for an article of clothing, possible variables include previous sales, time of year, location, weather, and price.

The bright future of AI-supported merchandise planning

AI has captured people’s attention for a good reason: the opportunities feel endless! But as with anything that is hyped, it’s important to stay grounded. Careful thinking about what AI is good for — and what it isn’t — can help merchandise planning teams find the right ways to bring AI into their workflows.

One of the most exciting ways that AI can impact fashion planning teams is through demand forecasting. You can read much more about the biggest impacts AI has on demand forecasting in our companion blog.