From robotics to artificial intelligence, here's how Amazon gets its products to Houstonians in record time. Photo by Natalie Harms/InnovationMap

Last summer, Amazon opened the doors to its North Houston distribution center — one of the company's 50 centers worldwide that uses automation and robotics to fulfill online orders.

The Pinto Business Park facility has millions of products in inventory across four floors. Products that are 25 pounds or less (nothing heavier is stocked at this location) pass through 20 miles of conveyor belts, 1,500 employees, and hundreds of robots.

The center also has daily tours open to the public. We recently visited to see for ourselves the process a product goes through at this Houston plant. From stowing to shipping, here's how packages go from your shopping cart to your front porch.

Starting with stowing 

Natalie Harms/InnovationMap

A product's first step in an Amazon facility is stowing. There's no categorization of the products — it's not like there's one floor for one type of item or anything.

"It's completely randomly stowed," says Donna Beadle, PR specialist for Amazon. "She could be stowing cat food on this floor, and so could somebody on floor two."

An Amazon employee would scan an item and stow it into an empty bin of her choosing — sort of. To prevent confusion, a light projected indicates bins that are off limits to stow the item. The light identifies bins that have similar products. Keeping similar products apart helps prevents mistakes for the employee who later pulls those items once its ordered.

The system also sees where the employee is putting each item, rather than having to scan each item and the bin as well. This is a newer feature — the facility originally opened with hand-held scanners.

"Our next generation workstation is that they don't have to hold that scanner — they have hands free," says Brenda Alford, regional communications manager at Amazon.

Robots on the move

Once the bins are fully stocked, the robot — which is the orange device on the bottom of the yellow bins — moves about the facility by scanning QR codes on the floor.

Should a product fall out, an employee wearing a special vest can enter to retrieve it. That vest will send off a signal to the robots, which will then decrease their speeds and come to a stop when the employee comes close.

"It's an extra measure of safety so that people can interact with the robots and feel safe," says Beadle.

Picking before packing

Natalie Harms/InnovationMap

Once an item is ordered, the bin with that item appears in the pick process at the center. The system tells the Amazon employee which item to grab and which bin to put it in. The bins will have products for multiple different orders — another employee later will separate it out later.

"Often we describe it as a symphony — our technology and our associates working together," Alford says, noting that sometimes the company might receive criticism about using robots over humans. "We can't do this without these humans.

Amazon employees receive their benefits from day one on the job, Beadle says, and they work four, 10-hour days a week.

"We feel like that way they have more time with their families — they get three days off versus two days off. And that gives them time to heal and rest up," she says.

Bin to bin and back again

Natalie Harms/InnovationMap

Once full, the Amazon associate will push the bin onto a series of conveyor belts. The whole facility has 20 miles of conveyor belts — much of which happens overhead.

The bins then zigzag toward the pack process, which is separated to different stations. There are single-product stations and multiple package stations. The system determines where the bin should go, and some stations pack products that are determined to need packing materials, while others do not.

Single-product packaging

Natalie Harms/InnovationMap

At the packing process, the Amazon employee is told which size box to assemble — he or she can grab a bigger box, but they can't select a smaller one. The tape dispenser doles out the correct size of tape for that box automatically.

Once packaged up, a sticker with a barcode is placed on the box. This code will later be used to print the label for shipping. At this point in the process, no personal information has been revealed to anyone. In fact, most packages leave the facility without any personal information being viewed by employees.

In an effort to reduce packing materials, some products are shipped in the container they came in. In that instance, the packer would just place the barcode sticker on the package before sending it on the conveyor belt.

"If we don't need another box for that product, we don't use one," Beadle says. "We work with companies to make that happen, so we don't have to use more boxes if we don't have to."

SLAM 


While the robotics aren't slamming labels on packages, the SLAM process (short for scan, label, apply and manifest) is the first step in the process that includes a customer's personal information. During this process, the barcode is scanned, the package is weighed, and the label is printed and affixed to the package using a puff of air.

A package might be automatically pulled from the line if something seems to be off in the package's weight.

"Say you bought toothpaste, and it says that toothpaste weighs 20 pounds, we know something's wrong," Beadle says. "Like maybe that it was a pack that didn't get separated."

If the package is kicked off, an Amazon associate, called a problem solver, will assess the situation and make it right before returning it to the conveyor belt.

Kicked into gear

Once labeled, all the packages are sent on their final conveyor belt ride. Using a scanning process, the packages are kicked by an automated foot that sends them into a line to be loaded into an Amazon truck.

If a package misses its chute the first time around, it makes the loop again. The system can tell if a package is caught in the loop for whatever reason, and a problem solver might be called to assess the situation.

Down the slide

Natalie Harms/InnovationMap

After being kicked off the belt, the package then slides down a spiral chute that, despite looking like a playground slide, is off limits to any humans wanting to keep their job.

"People ask if you can go down the slide, and we always say that on your last day of work," Beadle jokes.

On to the shipping process

Natalie Harms/InnovationMap

The packages leave the facility in Amazon trucks and head to one more pit stop before making it to the customer.

"They don't go directly to your house after this process," Beadle says. "They go to a sortation center."

This could mean a USPS or UPS stop, but it depends on where the customer lives.

Ad Placement 300x100
Ad Placement 300x600

CultureMap Emails are Awesome

11 Houston researchers named to Rice innovation cohort

top of class

The Liu Idea Lab for Innovation and Entrepreneurship (Lilie) has named 11 students and researchers with breakthrough ideas to its 2026 Rice Innovation Fellows cohort.

The program, first launched in 2022, aims to support Rice Ph.D. students and postdocs in turning their research into real-world ventures. Participants receive $10,000 in translational research funding, co-working space and personalized mentorship.

The eleven 2026 Innovation Fellows are:

Ehsan Aalaei, Bioengineering, Ph.D. 2027

Professor Michael King Laboratory

Aalaei is developing new therapies to prevent the spread of cancer.

Matt Lee, Bioengineering, Ph.D. 2027

Professor Caleb Bashor Laboratory

Lee’s work uses AI to design the genetic instructions for more effective therapies.

Thomas Howlett, Bioengineering, Postdoctoral 2028

Professor Kelsey Swingle Laboratory

Howlett is developing a self-administered, nonhormonal treatment for heavy menstrual bleeding.

Jonathan Montes, Bioengineering, Ph.D. 2025

Professor Jessica Butts Laboratory

Montes and his team are developing a fast-acting, long-lasting nasal spray to relieve chronic and acute anxiety.

Siliang Li, BioSciences, Postdoctoral 2025

Professor Caroline Ajo-Franklin Laboratory

Li is developing noninvasive devices that can quickly monitor gut health signals.

Gina Pizzo, Statistics, Lecturer

Pizzo’s research uses data modeling to forecast crop performance and soil health.

Alex Sadamune, Bioengineering, Ph.D. 2027

Professor Chong Xie Laboratory

Sadamune is working to scale the production of high-precision neural implants.

Jaeho Shin, Chemistry, Postdoctoral 2027

Professor James M. Tour Laboratory

Shin is developing next-generation semiconductor and memory technologies to advance computing and AI.

Will Schmid, Electrical and Computer Engineering, Postdoctoral 2025

Professor Alessandro Alabastri Laboratory

Schmid is developing scalable technologies to recover critical minerals from high-salinity resources.

Khadija Zanna, Electrical and Computer Engineering, Ph.D. 2026

Professor Akane Sano Laboratory

Zanna is building machine learning tools to help companies deploy advanced AI in compliance with complex global regulations.

Ava Zoba, Materials Science and Nano Engineering, Ph.D. 2029

Professor Christina Tringides Laboratory

Zoba is designing implantable devices to improve the monitoring of brain function following tumor-removal surgery.

According to Rice, its Innovation Fellows have gone on to raise over $30 million and join top programs, including The Activate Fellowship, Chain Reaction Innovations Fellowship, the Texas Medical Center’s Cancer Therapeutics Accelerator and the Rice Biotech Launch Pad. Past participants include ventures like Helix Earth Technologies and HEXASpec.

“These fellows aren’t just advancing science — they’re building the future of industry here at Rice,” Kyle Judah, Lilie’s executive director, said in a news release. “Alongside their faculty members, they’re stepping into the uncertainty of turning research into real-world solutions. That commitment is rare, and it’s exactly why Lilie and Rice are proud to stand shoulder-to-shoulder with them and nurture their ambition to take on civilization-scale problems that truly matter.”

Houston startup debuts new drone for first responders

taking flight

Houston-based Paladin Drones has debuted Knighthawk 2.0, its new autonomous, first-responder drone.

The drone aims to strengthen emergency response and protect first responders, the company said in a news release.

“We’re excited to launch Knighthawk 2.0 to help build safer cities and give any city across the world less than a 70-second response time for any emergency,” said Divyaditya Shrivastava, CEO of Paladin.

The Knighthawk 2.0 is built on Paladin’s Drone as a First Responder (DFR) technology. It is equipped with an advanced thermal camera with long-range 5G/LTE connectivity that provides first responders with live, critical aerial awareness before crews reach the ground. The new drone is National Defense Authorization Act-compliant and integrates with Paladin's existing products, Watchtower and Paladin EXT.

Knighthawk 2.0 can log more than 40 minutes of flight time and is faster than its previous model, reaching a reported cruising speed of more than 70 kilometers per hour. It also features more advanced sensors, precision GPS and obstacle avoidance technology, which allows it to operate in a variety of terrains and emergency conditions.

Paladin also announced a partnership with Portuguese drone manufacturer Beyond Vision to integrate its Drone as a First Responder (DFR) technology with Beyond Vision’s NATO-compliant, fully autonomous unmanned aerial systems. Paladin has begun to deploy the Knighthawk 2.0 internationally, including in India and Portugal.

The company raised a $5.2 million seed round in 2024 and another round for an undisclosed amount earlier this year. In 2019, Houston’s Memorial Villages Police Department piloted Paladin’s technology.

According to the company, Paladin wants autonomous drones responding to every 911 call in the U.S. by 2027.

Rice research explores how shopping data could reshape credit scores

houston voices

More than a billion people worldwide can’t access credit cards or loans because they lack a traditional credit score. Without a formal borrowing history, banks often view them as unreliable and risky. To reach these borrowers, lenders have begun experimenting with alternative signals of financial reliability, such as consistent utility or mobile phone payments.

New research from Rice Business builds on that approach. Previous work by assistant professor of marketing Jung Youn Lee showed that everyday data like grocery store receipts can help expand access to credit and support upward mobility. Her latest study extends this insight, using broader consumer spending patterns to explore how alternative credit scores could be created for people with no credit history.

Forthcoming in the Journal of Marketing Research, the study finds that when lenders use data from daily purchases — at grocery, pharmacy, and home improvement stores — credit card approval rates rise. The findings give lenders a powerful new tool to connect the unbanked to credit, laying the foundation for long-term financial security and stronger local economies.

Turning Shopping Habits into Credit Data

To test the impact of retail transaction data on credit card approval rates, the researchers partnered with a Peruvian company that owns both retail businesses and a credit card issuer. In Peru, only 22% of people report borrowing money from a formal financial institution or using a mobile money account.

The team combined three sets of data: credit card applications from the company, loyalty card transactions, and individuals’ credit histories from Peru’s financial regulatory authority. The company’s point-of-sale data included the types of items purchased, how customers paid, and whether they bought sale items.

“The key takeaway is that we can create a new kind of credit score for people who lack traditional credit histories, using their retail shopping behavior to expand access to credit,” Lee says.

The final sample included 46,039 credit card applicants who had received a single credit decision, had no delinquent loans, and made at least one purchase between January 2021 and May 2022. Of these, 62% had a credit history and 38% did not.

Using this data, the researchers built an algorithm that generated credit scores based on retail purchases and predicted repayment behavior in the six months following the application. They then simulated credit card approval decisions.

Retail Scores Boost Approvals, Reduce Defaults

The researchers found that using retail purchase data to build credit scores for people without traditional credit histories significantly increased their chances of approval. Certain shopping behaviors — such as seeking out sale items — were linked to greater reliability as borrowers.

For lenders using a fixed credit score threshold, approval rates rose from 15.5% to 47.8%. Lenders basing decisions on a target loan default rate also saw approvals rise, from 15.6% to 31.3%.

“The key takeaway is that we can create a new kind of credit score for people who lack traditional credit histories, using their retail shopping behavior to expand access to credit,” Lee says. “This approach benefits unbanked applicants regardless of a lender’s specific goals — though the size of the benefit may vary.”

Applicants without credit histories who were approved using the retail-based credit score were also more likely to repay their loans, indicating genuine creditworthiness. Among first-time borrowers, the default rate dropped from 4.74% to 3.31% when lenders incorporated retail data into their decisions and kept approval rates constant.

For applicants with existing credit histories, the opposite was true: approval rates fell slightly, from 87.5% to 84.5%, as the new model more effectively screened out high-risk applicants.

Expanding Access, Managing Risk

The study offers clear takeaways for banks and credit card companies. Lenders who want to approve more applications without taking on too much risk can use parts of the researchers’ model to design their own credit scoring tools based on customers’ shopping habits.

Still, Lee says, the process must be transparent. Consumers should know how their spending data might be used and decide for themselves whether the potential benefits outweigh privacy concerns. That means lenders must clearly communicate how data is collected, stored, and protected—and ensure customers can opt in with informed consent.

Banks should also keep a close eye on first-time borrowers to make sure they’re using credit responsibly. “Proactive customer management is crucial,” Lee says. That might mean starting people off with lower credit limits and raising them gradually as they demonstrate good repayment behavior.

This approach can also discourage people from trying to “game the system” by changing their spending patterns temporarily to boost their retail-based credit score. Lenders can design their models to detect that kind of behavior, too.

The Future of Credit

One risk of using retail data is that lenders might unintentionally reject applicants who would have qualified under traditional criteria — say, because of one unusual purchase. Lee says banks can fine-tune their models to minimize those errors.

She also notes that the same approach could eventually be used for other types of loans, such as mortgages or auto loans. Combined with her earlier research showing that grocery purchase data can predict defaults, the findings strengthen the case that shopping behavior can reliably signal creditworthiness.

“If you tend to buy sale items, you’re more likely to be a good borrower. Or if you often buy healthy food, you’re probably more creditworthy,” Lee explains. “This idea can be applied broadly, but models should still be customized for different situations.”

---

This article originally appeared on Rice Business Wisdom. Written by Deborah Lynn Blumberg

Anderson, Lee, and Yang (2025). “Who Benefits from Alternative Data for Credit Scoring? Evidence from Peru,” Journal of Marketing Research.