Exclusive: This New AI Model Translates Data Correlations Into Demand Forecasting Insights
Syrup Tech wants to help fashion brands and retailers add something sweet to their bottom lines, not just to their pancakes.
The New York-based technology company announced Tuesday it has launched a new model, slated to help companies with demand forecasting and product allocation. It’s powered by neural networks, a form of artificial intelligence that can ingest data, process it and find correlations between significant points.
More from Sourcing Journal
Neural networks are modeled after the human brain, and they use a variety of data points to come to conclusions. James Theuerkauf, the company’s CEO and co-founder, said that the company’s latest innovation, which it calls a large demand sensing model (LDSM), serves a variety of purposes, from generating cold-start forecasts, to identifying complementarity between items and more.
But perhaps its greatest strengths are in its ability to connect the complex, he said—neural networks can help companies process a slew of data inputs, from color and size of existing inventory, to social media trends influencing business from the outside.
“Typical models, they’re good at predicting your [staple items]—things that are fairly stable,” Theuerkauf explained. “This [LDSM] is really good at responding quickly to highly variable trends, both on demand as well as on supply.”
That kind of prowess is meant to help solve some of retail’s trickiest problems. Theuerkauf said oftentimes, customers pair products together—but when retailers have no way of deducing how or why a customer’s purchase of one item might influence another, they may be losing out on sales.
“One thing that makes retail really hard is that products obviously don’t just work in isolation,” he said.
With Syrup’s LDSM, companies will be able to better understand how products work in tandem.
For instance, if customers consistently purchase a blue blouse and white jeans together, the neural nets can help retailers understand the impact of an out of stock on one of those two items.
It can also recommend how to allocate items from other locations to best serve the needs of customers. With that same example, if one New York City retail store is experiencing high purchase levels on that combination of items, while another store in the same city doesn’t seem to see the same pattern, the neural nets might recommend moving some stock of the blue blouse from one store into the other.
Alternatively, if both stores are nearing out of stocks on the blue blouse, the model might recommend upping stock on an item similar to the original garment to help boost sales of the complementary item. So, in this case, if the company lacks stock of the blue blouse, the nets might recommend moving stock of a similar green blouse into the locations running low on a key item.
The model uses some automated data, like sales volume and social media trends, to help forecast where companies should go next, but it also takes into account how the employees using it have reacted to its predictions and suggestions.
Theuerkauf said that when a person overrides the model’s recommendations, the system uses that information to better the model down the line.
“The edits that you make, all of those are effectively stored in our database [and] can be used as an input. So, in a world where fashion always will be art and science, this art component is really important and can be used to train the engine,” he told Sourcing Journal.
And, unlike a human brain, the AI model doesn’t react negatively to criticism, Theuerkauf went on to say.
“The [tech’s] brain has no ego. You can tell it however many times you want that it’s wrong, and it won’t take that personally. In fact, it’s your decision at the end of the day,” he noted.
While much of the use case for the LDSM seems to be around brick-and-mortar locations, Theuerkauf said early piloting of the solution shows that it will also have an impact on e-commerce, particularly when retailers consider how much to purchase and whether to make re-buys throughout a season.
Though they have not yet adopted the LDSM, customers like Faherty Brands, Hoka and Reformation rely on Syrup’s technology for assistance with buying, planning and allocation. The company already had other proprietary models in place, which did not leverage neural networks technology, but did make use of predictive analytics and other subcategories of AI.
Theuerkauf said now that the LDSM is ready to be used by existing and onboarding customers, it will fit into Syrup’s mix as another tool in its kit; he does not expect the neural network to cannibalize the company’s existing technology—instead, he believes the systems will become composable based on what each individual client needs. That means that clients can use different pieces of Syrup’s technology based on the use case they’re working to meet internally.
As Theuerkauf and his team think about what might be next, the CEO has honed in on making the technology broadly available, with low barriers to entry. While many emerging technology companies ask clients to move away from their existing providers, Syrup wants to use what brands and retailers already have in place, then bolster it with the company’s new technologies.
“A really big [priority] for us is understanding how we can be as flexible and as data agnostic [as possible], so that any merchandising planning organization can make use of this intelligence. And we often get the question from brands that we’re in conversations with, of, ‘Oh, are you going to play nice with [our existing software provider]?’” he said. “From our perspective, the answer is, unequivocally, yes. We don’t care what underlying stack you have; we just want to augment and accelerate that with our intelligence.”
Solve the daily Crossword
