Pathr.ai

The insights help retailers improve sales and operations, reinvent customer experiences, prevent theft, manage the flow of people, and answer questions like these:

• Who’s standing in line where—and for how long?
• Does merchandise placement impact sales?
• Where are salespeople most needed?
• What action do customers take after interacting with a salesperson?
• Where are people spending the most time?
  
  

“Our tagline is, ‘You can learn a lot from a dot,’” says Pathr.ai’s Product Marketing Manager Nicole O’Keefe. “To simplify, we turn the people on camera feeds into dots. From the behavior and movement of those dots, you gain insights that connect to business decisions and solve operational problems.” As one small example of learning a lot from a dot, Pathr.ai is integrating real-time notifications to help employees quickly get to waiting customers so stores can capture sales rather than risk those customers walking away. “We’re inventing a whole new category in machine learning,” she says.

Head of Data Science Jason Sadowski, Ph.D., agrees. “We’re bringing the kinds of insights, analysis, and experimentation people are doing in online environments into the physical world,” he says. “As an example, take those long, snaking lines at the register. Where you put different products along that queue has an impact on the sales of those products. Our customers use the data we provide to benchmark and experiment in their spaces at scale across multiple locations and regions—something they can do without installing more equipment or constantly repositioning cameras.”

Their Challenges

At its inception, Pathr.ai aspired to stand out in the computer vision, spatial intelligence, and retail AI fields by being flexible and delivering insights from existing cameras to customers in a matter of weeks while offering cost-effective pricing. But as the business grew, it became hard for the small team to manage annotation while also scaling and delivering high customer ROI.

“For one customer, for example, we have to build a dataset across, say, five cameras, making sure we have the correct annotations,” says Sadowski. “That time investment across many cameras and many locations and customers meant we wouldn’t have been able to spend time scaling, or developing and improving the product.”

At the top of its list of priorities, the Pathr.ai team was after three improvements:

1. Set up and install quickly so customers could launch new initiatives faster and reduce time to market.
2. Quickly calibrate cameras based on what each camera sees—important because Pathr.ai relies on a customer’s existing infrastructure; customers don’t have to add new cameras or reposition cameras.
3. Give customers the most accurate analytics possible.

Our relationship began in June 2020, with CloudFactory performing entrance counts and simple annotations for entrance analytics. As Pathr.ai continued to grow, the services CloudFactory provides expanded accordingly.

“When we started, we used CloudFactory’s data annotation to retrain our computer vision algorithms,” says Cayetano. “We then moved on to camera calibration, allowing us to expand our partnership together." Camera calibration, which involves creating one-to-one mappings between points on store video feeds to points on floorplans, turns videos into a format the Pathr.ai platform can use. Today, as Pathr.ai onboards new customers, CloudFactory performs custom camera calibrations to pinpoint camera locations on a floor plan and identify what each camera sees, information Pathr.ai then uses to project what the camera sees to a mapping of moving dots on a floor plan.

CloudFactory also provides data annotation for Pathr.ai’s analytics, which are based on dots moving on floor plans, such as customers entering the store, heading to specialty departments, or browsing through the jewelry cases. In this process, CloudFactory annotates videos for Pathr.ai’s validation of analytics.

And because Pathr.ai designs custom analytics to help customers with the decisions and optimizations each is after, CloudFactory also labels images and video footage to show traffic patterns, fitting-room entrances, check-out queues, customer interactions, and people tracking across multiple cameras throughout a space. Pathr.ai then uses this information to train its in-house models and build up better person detectors.

“We appreciate CloudFactory’s flexibility, whether it’s expanding services beyond annotation or downsizing or upscaling in a matter of days,” says Cayetano. “It’s been a game-changer. If we sign a deal with an enterprise customer and need to deploy into hundreds, if not thousands, of locations, CloudFactory scales up right away based on the number of hours we need.”

Sadowski says he, too, appreciates the ability to shift to new use cases and annotation types, as well as the ability to shift tools. “We started off doing camera calibration with one in-house tool, and then we shifted over to another,” he says. “CloudFactory moved seamlessly across those two tools, which was very helpful to us.”

The Results