Menu

Resources

How does Nextail use Artificial Intelligence and Machine Learning to help retailers?

Luis Pérez Vázquez, Nextail Data Scientist, co-authored this frequently asked question.

It’s not magic, it’s data science!

We’re often asked if and how Nextail applies AI, so there are likely even more people who wonder the same thing but who feel a little shy about asking.

That’s likely because even though it’s pretty common to hear about concepts like “artificial intelligence” and “machine learning”, there may be some lingering doubt as to what they mean and how they are helping retailers.

While Nextail is definitely sophisticated technology, isn’t a blackbox solution and it doesn’t require magic. It’s data science! So, here’s a quick overview of how these concepts are defined, followed by how Nextail uses them within the platform of inventory management solutions.

Artificial Intelligence and Machine Learning in a nutshell

Artificial intelligence, or AI, is a blanket term for algorithms (calculations and other problem-solving operations) that make it possible for certain tech to “think” like a human. Specifically, AI instructs technologies and machines to function or react like a human would when presented with certain circumstances or information. Hence the term “intelligence”.

Machine learning, or ML, on the other hand, is what helps the tech learn to become intelligent in order to do things like classify and group information or make predictions. Not only that, but machine learning algorithms also to continue learning on their own and get smarter over time. Again, much like a human would.

For example, a regular algorithm might be able to recognize a human face among many different images, based on a predefined set of criteria (e.g. round shape, existence of areas resembling eyes, nose, mouth, etc.).

But an artificial intelligence “machine” made up of ML algorithms can be “trained” through thousands and thousands of images (data) to draw its own conclusions about what a human face looks like. Eventually, this “intelligent” technology is able to tell us not only that we are looking at a face, but also who it belongs to, a.k.a. facial recognition. This is much closer to how a human would process this information, getting smarter and smarter on its own.

So, you can think of AI as a student, books as the data, and ML as the brain getting more intelligent over time.

Nextail solutions use AI & ML to help retailer digitally transform decision making for better results

In retail, Nextail solutions use AI and ML to forecast hyper-local demand and automate very complex decisions, providing a layer of intelligence behind them that, we as humans, just aren’t capable of processing quickly, efficiently, or even at all.

The more time that goes by, and the more data that Nextail algorithms can “train” with, the smarter they get and the better they are at doing things like predicting demand, deciding where each and every product should be located across a retail network at any given time for a higher likelihood of sale, or identifying similar products. This is because AI and ML allow Nextail solutions to take a bottom-up approach to decision making.

In the past, retailers have relied on traditional top-down approaches for making merchandising decisions (i.e. WSSI, open-to-buy, etc.). But this is getting much more complicated now with the proliferation of online channels and constantly changing customer demands.

These old approaches, which required retailers to act on decisions they’d made up to a year in advance, just aren’t flexible, efficient, or dynamic enough to help retailers understand what will happen in the future and what to do when things change. AI and ML approaches are.

Predicting sales probability for each and every item in every location

Nextail inventory management solutions use AI to make predictions about the sales probability of any number of products for sale in different locations at different times. For example, they can determine how much more likely a red shirt in size S will sell in Store A versus Store B or Store C in the near future.

Sure, you could try to do that on Excel, but it could take hours or even days to get to the level of granularity that AI allows for. And you’d still likely wind up with human errors and overgeneralized predictions. No one could blame you. It’s a gargantuan task, and you could most definitely find more value added tasks to fill your time.

Since the computing capacity of artificial intelligence is exponentially greater when it is cloud-based (stored, managed, and processed on a network of servers on the internet) it can bind otherwise disconnected pools of data to work with many more data sources, such as a retailer’s own data for thousands of products and stores, Nextail-generated data (e.g. product attributes, size curves, etc.), and even external data (e.g. seasonality patterns, aggregate data from other retailers, etc.).

This means that Nextail is able to deliver SKU-POS level demand forecasts in the short-term. What’s more, solutions apply optimization algorithms to maximize the probability of sales of each unit. In other words, of all of the possible decisions that can be made based on the forecast, the algorithms will decide which are the very best ones based on specific business considerations.

Forecasts that get even stronger over time

As explained above, ML is used to train AI algorithms to get smarter and arguably better over time. Since Nextail applies ML, the longer it is able to collect and use (“train” with) a retailer’s data, the better it gets at understanding specific and subtle patterns about product and point-of-sale performance.

All these data points mean that artificial intelligence algorithms, thanks to ML, are going to be much better at making specific predictions than you or I would be able to do. And why not let them carry out the hard labor instead of merchandising teams?

A seasonality model that considers how internal and external factors impact demand patterns

Nextail demand forecasts combine not only past performance data, but also account how different external factors impact demand. That way, retailers can always ensure the best availability across all channels without having to increase coverage levels.

This is where data science comes in. The Nextail seasonality model uses machine learning to identify and consider three different elements: trends, events, and the effect of recurring patterns over a period of time.

  • Trends – General upward or downward changes in shopping tendencies which aren’t related to any type of pattern. For example, think of well-known trends such as bell-bottom vs. skinny jeans, or the surge in athleisure wear during the COVID-19 pandemic.
  • Events – One-off and non-recurring events, usually over a short or defined period of time that affect demand. Some examples of this are the demand spikes around one-off events like weather changes that affect foot traffic in physical stores or even the pick-up in demand for football jerseys on game-days in stores located near stadiums.
  • Finally, the Nextail model captures recurrent, periodic changes over years, seasons, months, and weeks/weekends, etc. (applying the “Fourier Transform” mathematical tool). Also referred to as “seasonality”, examples of these external impacts are the annual back-to-school uplift in kidswear toward the end of the summer, spikes in demand during the winter holiday season.

Nextail uses a machine learning model called “Ridge Regression” to assign weights to these trends, events, and seasonality, identifying the best and least complex combination of the three elements. This results in the final demand forecast that is both robust – meaning it’s built based on sufficient data points – and explainable.

Not enough data? AI helps identify comparable items and works with “dynamic fallbacks”

Generally speaking, Nextail will calculate size curves at a given point of sale with the use of historical data for product performance at each point of sale.

But what if you don’t have much data on a certain product? This might be the case if we have a new product that we haven’t sold before, or a rare product that hasn’t generated much sales data yet.

On the one hand, Nextail applies computer vision, a type of AI, as well as data from master files to extract product attributes (e.g. family, price range, product description, material, color, print, etc.) for identifying comparable products and store performance.

That way, if you’re about to introduce a product which doesn’t have any stock or sales history to draw from in a particular point of sale, Nextail will build a preliminary demand forecast for different sizes based on similar products sold in the past. You can read more about that here.

And now what if no comparable products exist within a given point of sale? Data science can solve for that too.

The Nextail seasonality model works with products and stores grouped with business intelligence data from retailers in hierarchically ordered categories (e.g. region, family, etc.). This means that even if data is scarce, Nextail algorithms can expand the analysis, through “dynamic fallbacks”, to look for comparable products from this wider pool of products or group of products at the regional or country level.

What’s next: Deep Learning to further decrease buy quantities and lost sales

As with any other science, data science is always in a constant state of change and evolution, and the Nextail Data Science team is always investigating how to push models even further and take forecasts to the next level. In this case, we’re getting more profound – with Deep Learning (DL).

In a nutshell, DL is a subset of machine learning based in artificial neural networks. These algorithms are VERY data hungry, meaning that they need tons and tons of data to generate acceptable results.

While these models are “needy” in that sense, their strength lies in the fact that many different types of data sources can be incorporated into them. And once they are trained, not only are they able to capture the most non-linear volatile patterns in data with amazing accuracy, they do so very quickly.

For retailers, this translates into an even clearer picture of demand, allowing them to buy even more intelligently – reducing quantities and avoiding lost sales, ultimately increasing the ROI of their largest investment – their inventory.


In the “Ask Nextail” series, we’ve compiled the top most frequently asked questions that we receive. Team members from across Nextail help to answer these questions, bringing expertise from their work helping retailers understand, implement, and be successful managing the radical change that comes with digital transformation in retail.

Contact us for more information on how Nextail and artificial intelligence can help you automate and streamline your inventory management!

Nextail FAQ series demand forecasts

What data does Nextail use for forecasting demand?

Giona Capurso, Product Manager, provided insights to answer this frequently asked question. In the face of uncertaint...

Nextail FAQ series distorted demand

How can Nextail prevent unusual demand patterns from dist...

Alba Vacas, Nextail Change Manager, provided insights to answer this Nextail FAQ on preventing data and forecast dist...

Nextail FAQ series identical products

How does Nextail manage multiple SKUs for identical produ...

David Vicente, Nextail Integration Project Manager, provided insights to answer this retail merchandising question. M...