Consumers around the world know Gap Inc. for its timeless clothing that appeals to many generations. After more than 50 years in business, Gap Inc. has also honed its supply-chain systems to process millions of items each year at maximum efficiency. Given its scale, the company actively researches and tests innovative solutions to stay ahead of the technology curve and the competition.

In 2016, Gap Inc. started a far-reaching transition to a new fulfillment network. The Cross-Channel Logistics Optimization project was designed to position it ahead of its competition, as well as bounce back from a destructive fire at its northeastern distribution center that summer.

To meet the goals of the program, Kevin Kuntz, senior vice president of global logistics fulfillment, was looking to pilot innovative technologies to add to the automation solutions the company already used, such as automated storage and retrieval systems. Moreover, like all big retailers, Gap Inc. was continuing to grapple with the challenge of finding enough employees to support its operations during peak sales seasons.

Kuntz became aware of SORT, a picking technology from Kindred that uses artificial intelligence. SORT is a robotic system that employs vision, grasping and manipulation algorithms to pick and sort merchandise. Using AI methodologies to continuously improve the robot’s capabilities, the system evaluates millions of data points to calculate and execute an optimal pick strategy for each task in real time.

“After seeing the working demonstration of SORT, we felt this technology was further along than other solutions,” says Kuntz.

To begin evaluating the technology, Kuntz placed a pilot SORT system at Gap Inc.’s distribution center in Gallatin, Tenn., marking the first time the company had used AI technology in any of its D.C.s.

As with many D.C.s, Gap Inc.’s are built to meet peak demand. The pinnacle of demand arrives during the holiday season, with the highest traffic in stores and online during the week of Black Friday and Cyber Monday. A typical D.C. has manual, labor-intensive sortation stations that process customer orders.

Each D.C. houses a giant oval belt known as a Bombay sorter, which takes thousands of individual units and sends them to individual stations. Chutes attached to the Bombay sorter act as tributaries, funneling merchandise off the conveyor belt and to sorting stations.

At manual sorting stations, merchandise is dropped into a pile, and an employee picks up each piece and scans it with a barcode scanner. A corresponding light on the putwall, a vertical grid of cubbies, lights up, signaling to the employee which cubby corresponds to that order. The routine is repeated again and again as the merchandise is processed. When the last item in an order is scanned, a different color light flashes on that cubby, signaling to the worker that cubby is complete. Then the employee takes the contents of the cubby and passes it off to another station to be packed and shipped.

“The person-powered sort station is very manual,” says Zach Gomez, director of operations for customer success at Kindred, and Kuntz’s main point-person at Kindred.

“The process is monotonous, and as time on the shift goes by, the speed and accuracy of the employee diminish. It’s just human nature.”

Automated sorting processes supported by Kindred SORT can make that employee twice as productive, Gomez claims. “The SORT units are essentially plug-and-play,” he adds.

To start, Gap Inc. reconfigured the chutes from the Bombay sorter to send items directly into the bin of the SORT system. Next, the system’s cameras look at the items in the bin to compare them with tens of thousands of images it already has stored in its memory. Using AI, it matches what it sees to those existing images to recognize the merchandise, and identify where one item starts and the other one stops.

After the cameras record the location of each item, the robotic arm in the middle of the circular system picks up an object using propriety Autograsp technology — a combination of pinching, grasping and suction. As the robotic arm holds the item in its gripper, it quickly swings it around to expose any area where the barcode could be to one of four cameras inside the system. As soon as one of the cameras scans the item, the arm matches that item to the appropriate order, and drops it into the corresponding cubby.

If the robot isn’t able to secure the item, the SORT system notifies a “robot pilot” at Kindred’s headquarters to request help. Remotely, the pilot looks at a live view of the machine and helps guide the robotic arm to make the pick. Sometimes, an item ends up in the SORT bin that wasn’t supposed to be there — something too big to fit in the cubby, for instance. In that case, the robot pilot notifies the operator at the D.C., who can remove the item from the bin.

Using machine learning, SORT learns over time the shapes of items and other details — such as where the barcodes are — to become faster and more efficient at picking and sorting. After months of operating at Gap Inc. D.C.s, the SORT system can pick 98 percent of merchandise autonomously, and rarely requires intervention from the robot pilots.

Despite its machine learning and AI capabilities, SORT still relies on human participation for the last step of the picking process. After the robotic arm finishes adding items to an order, it triggers a light to notify the operator that it’s complete. The operator pulls the completed order out of the cubby and transfers the items to the packing area.

One employee is assigned to at least two SORT systems to oversee operations. A single employee can pack more orders at a robotic SORT station because the robotic arm is doing most of the work. Workers overseeing multiple picking stations can be responsible for two to three times as much throughput.

Employees at the Gap Inc. D.C.s. play a key role in the success of the technology implementation. They’re in charge of managing the SORT equipment, and making sure that all processes are flowing smoothly from point to point.

“All of our employees find it very easy to interact with the SORT stations and feel very safe working in and around the machines,” says Kuntz. “Training is easy, and for the most part, we can spend an hour or less with our employees and have them working at rate. That includes learning all the exception processing. The design and simplicity of the machine make it very easy and quick to train new hires.”

Gap Inc. has subsequently added SORT to its facility in Fresno, CA and plans to install the system in its northern distribution center in Fishkill, NY by the end of 2019.

“Kindred’s willingness to improve all aspects of the system exceeded our expectations,” says Kuntz. “They enhanced the grippers and scanners to improve the accuracy of the machines. We’ve seen impressive speed improvements in the 18 months since we began implementation, and the percentage of autonomous picks has increased.”

Zach Gomez is director of customer success with Kindred.