When Thomas Melia, an engineer for the City of Salinas Fire Department in California, was dispatched earlier this year on a medical call to an assisted-living facility, his routing app told him where to park the truck. But when he pulled into the sprawling multi-unit complex, the ambulance service that had already arrived was parked near the front door, far from the apartment in need.
“It’s just a super long walk,” Melia said. His team had time to reposition the ambulance near the back door. “We looked and knew this unit is going to be right by the elevator by the back door.”
In the past, in order to make sure they parked in a spot where they could easily access a particular apartment and nearby standpipes, hydrants and sprinklers, the firefighters would have to flip through a “target hazard” binder full of printouts while frantically adjusting masks and toolbelts en route to an emergency.
Now, their fireproof mobile devices tell them not only the best route to take there, but the best spot to park the truck. That means Melia’s captain doesn’t have to flip through a binder as often. “It increases safety by having more eyes on the road,” said Melia. “You can start to come up with a plan before you even get there.”
They’re not using Google Maps, though.
Instead, the firefighters rely on routing technology from Beans.ai, one of many companies building unique data sets and AI-based software that make the final steps to a destination speedier and more efficient. After the pandemic fueled use of quick-commerce services and heightened delivery expectations in general, the routing data and applications devised by companies including Beans.ai, Here and even pizza purveyor Domino’s aren’t just about the last mile. They’re about the last few feet.
FOIA requests and building photogs
Founded over three years ago, Beans.ai was not originally intended strictly as a service for emergency responders. However, it turned out fire departments and government agencies were useful places to look when foraging for data that would help food and package delivery workers find the closest parking spot to a destination.
Not only did the company temporarily borrow the Salinas Fire Department’s binders of maps to input data into its system, but Beans.ai even used freedom-of-information requests — it petitioned 180 municipal agencies such as building inspection and occupational safety departments in all 50 states and 118 cities — in the hopes of retrieving information such as physical maps that show building layouts. In the end, only 31 of those agencies had something to share that the company could use to feed what it calls “ground-ops” data, said Nitin Gupta, its co-founder and CEO.
In fact, most of its data does not come from old-school analog sources. Instead, more than 90% of it has been supplied by people the company pays to submit photos of maps posted inside large apartment complexes, mobile home parks and assisted-living facilities through its 100ft Surveyor app and driver-routing app. To date, Beans said it has mapped about 70% of U.S. apartments that have 40 or more units.
Binders used by the Salinas Fire Department.Photo: Salinas Fire Department
Today, the company uses that information to refine machine-learning models for features in its apps that optimize delivery dispatch, routing and other logistics-related activities. For instance, its technology is integrated with FedEx Ground, allowing delivery contractors that pay $25 per month per driver to upload daily delivery locations in order to generate routing information for their drivers. But Beans.ai also aims to sell to government and public safety agencies through integrations with software used by emergency responders, including Tablet Command and BCS Marvlis.
Stairway to 15-minute deliveries
Hunting for last-foot data could pay off as the rise of quick-commerce delivery services from companies such as Buyk, Gopuff and Gorillas spur competition and development in the delivery-routing tech arena. More-established companies including DoorDash have been gathering delivery logistics data to train models used in their delivery fulfillment apps for years.
Last November, Coresight Research estimated that retail sales in the overall quick-commerce market would total $20 billion to $25 billion in the U.S. in 2021, around 10% of the research company’s share of estimated U.S. online consumer packaged goods sales for the year. New players like Fridge No More, JOKR and 1520 that promise 15-minute deliveries “have intensified instant needs in terms of speed promises,” according to Coresight.
Companies in the quick-commerce sector will need accurate data and tools to get items to people at a rapid speed or they’ll risk losing out on customers who can easily switch to another service, said Christoph Herzig, head of Fleet Applications for Here Technologies, which sells its routing application for delivery drivers to businesses.
“The value of location technology becomes even more important because it’s all about early user conversion,” Herzig said.
But other sorts of companies also are interested in data showing building minutiae. Herzig said Here is working with a package-delivery carrier to gather building data including details about elevators, staircases and mailboxes.
The company uses a combination of automation, machine learning and human analysis to test and refine delivery routing maps created using a variety of data sets such as information about road surfaces and speed limits: Data that is especially relevant to heavy commercial vehicle drivers who can only travel on certain roadways. Here also sells its proprietary data, and lets customers sell their own data sets or build applications using data available in its exchange.
Keeping tabs with workers’ personal devices
Pizza giant Domino’s has also attempted to mine data to help streamline deliveries. The company’s more than 10,000 delivery drivers use a delivery app that its manager of Data Science Zack Fragoso told Protocol is “a key piece” of its last-mile routing efforts.
The app tracks driver locations while they are en route in order to alert customers when they are nearby. But he said that when the app was first introduced, it needed better data to improve problems such as false notifications telling customers an order was delivered when it was not. In the hopes of achieving higher accuracy, the company added features to help gather more specific location information during deliveries, Fragoso said.
Over the last couple years, Domino’s delivery drivers have given the Domino's Delivery Experience app — which many drivers say their managers demand they install and use on their personal phones — mixed reviews on Reddit. Some have said they found it useful, while others complained that they didn’t like having to enable tracking functions on their phones for work, and worried about battery life and phone data usage.
“The only reason it's there is to make sure we're not off fuckin’ around during deliveries,” wrote a Reddit poster last year.
Domino’s delivery drivers have given the Domino's Delivery Experience app mixed reviews.Photo: Domino's
While helping to make emergency calls and deliveries more efficient, emerging technologies used for delivery routing and fleet management also create ethical questions related to worker privacy and civil rights.
Apps used by workers that merge personal and professional device use are “undermining workers’ basic human right to disconnect” especially “when workers are required to use personal devices that deliver data to employers, which can be used against them,” wrote Wilneida Negrón in a 2021 report for Coworker.org. The report pointed to threats to worker rights that can be exacerbated by data collection and algorithmic technologies such as increased worker monitoring, wage theft and labor-organizing surveillance.
When FedEx drivers or other delivery drivers use Beans.ai apps to assist in routing, they can turn off location tracking while still accessing static routing information, said Gupta, who added that Beans.ai does not sell its data as a separate product.
Still, as Beans.ai pushes for ways to turn its data technologies into valuable services for business customers in the delivery game, the company has added features and capabilities that some drivers might find invasive. For example, the company uses phone accelerator data to determine whether someone is driving, walking or idle, as well as camera footage data from a dashcam provider to help determine whether deliveries happened, if drivers parked where they say they did or if they were driving while using their phones.
That data gets fed into the Beans.ai system to improve its machine-learning models for routing, but is also used for driver safety reports provided to customers to keep track of driver activity. “With the camera footage we’re getting, we’re able to constantly reevaluate the data,” Gupta said, noting that the Beans.ai app only tracks drivers while they’re working. “When the driver clocks out, it automatically clocks out of any type of tracking.”
The additional driver-tracking capabilities were integrated to ensure the company’s services stay relevant to potential customers, such as delivery contractors that are required to monitor driver incidents or behavior, he said. “Our integration with our partners helps streamline the user experience,” Gupta said. “Every other data company we looked at in the space does not have a strong feedback loop on their data.”
