Food is life!
And technology should make life better

Inside the robot-powered grocery store of the future

Fulfil Solutions has rethought every aspect of the grocery fulfillment center to eliminate the pain points of online market shopping.

‍

By Clint Rainey

July 18, 2023

‍

In e-commerce, little is more personal than grocery shopping. A single mistake can torpedo a market order’s whole purpose. Say you want to make a pasta dinner but your order arrives with tomato sauce, rather than the canned tomato paste your recipe calls for. Or the noodles are not gluten-free, as requested. Suddenly the convenience of shopping online has curdled into annoyance. 

Of course, that may be only the beginning of the irritation. After all, you paid a $9.95 delivery fee for your ruined dinner. The exasperation ripples through every node in the grocery commerce system. The shopper who’s trying to make this gig work rifled through 30,000 supermarket SKUs in a scavenger hunt rivaling Guy’s Grocery Games to tackle your shopping list. The store itself isn’t exactly winning either: It may pocket a grand total of $2 or $3, after delivery costs eat into grocery’s already slim profit margins.

“In no other e-commerce experience do you order something and get something else,” says Mir Aamir, president and CEO of Fulfil Solutions, a robotics startup that seeks to fix a growing, but fundamentally broken, segment of the $6.3 trillion e-commerce business. 

‍

Grocery’s online moment—and its problems

‍

Groceries are in the throes of their biggest industry shakeup since conveyors took over checkout counters. COVID-19 caused customers to pivot en masse toward not just “safer” Amazon-style online orders but also to immediate orders that mirrored their own shopping behavior. Supermarkets did their hasty metamorphosis into dark stores, and a niche became a trend, one that has outlived the pandemic. Online orders now exceed 10% of U.S. grocery sales.

‍

And yet, Aamir says of grocery delivery, “Here’s a $100 billion business that for the most part is unprofitable.” Why? Because the industry has applied a fail-safe automation strategy, called goods-to-person, that helped e-commerce pioneers like Amazon reshape the modern warehouse. Think of the Segway-like bots whooshing items to stationary workers, and that’s the model that’s been replicated across the world of fulfillment centers serving digital shoppers. 

Aamir argues that this model is ill-suited for groceries: “The same technology that was used to automate digital cameras and shoes cannot be applied to cilantro and Chobani yogurt,” he says.

There are a surfeit of companies working to solve grocery’s original hangups: idiosyncratic orders that must be accurate; consumers resisting markup on “basic goods”; and the contradiction that lists may comprise 60 items, causing bottlenecks, but it’s to pickers’ advantage to work at breakneck speed. Albertsons partnered with a goods-to-person company called Takeoff Technologies. FreshDirect has worked with a competitor named Fabric. Walmart recently acquired a player named Alert Innovation. And H-E-B and Amazon have both tapped Norway-based AutoStore.

But it’s Aamir, a longtime Safeway executive, who has taken a methodical approach to rethinking the emerging sector within both e-commerce and grocery, developing a system and strategy that he believes can improve the customer experience and make the business worthwhile for retailers.

‍

A new way to robot

Aamir argues that Fulfil’s edge is an affordable, accurate, person-less reconceptualization of the grocery fulfillment center that he dubs “robots-to-goods.” Fulfil’s model begins with a high-tech facility equipped with towers of trays in a 3D grid, surrounded by a maze of chutes, roller conveyors, and paths marked on the floor. Upon command, an army of Fulfil’s proprietary ShopBots dispatch to nodes via those paths where they beckon customers’ items.

‍

ShopBots resemble Xerox photocopiers if their lids were removed and you hollowed out the machines, lined them with totes, and added wheels. They understand the precise size and weight of every packaged item in the 3D grid (loose items get a close estimate), and Aamir says they’ve harnessed years worth of work with algorithms and machine learning to optimize bagging. Sensors and heat maps ensure that a watermelon or Tide bottle never crushes a loaf of bread. To test the bots’ adjustable, cushioned bins, Fulfil picked random spots to drop raw eggs and see if they’d break, but none did.

Notably, ShopBots are the only automated grocery pickers capable of working in freezers. Logistics get tricky in cold environments: Circuit boards become sluggish if frozen, and they can freeze quickly. Aamir says that the simple end around was capping the elapsed time that ShopBots spend in its frozen towers. They’re limited to a minute or two.

Fulfil also catalogs products individually. The industry standard is to lump all large, organic Hass avocados under one code; all small, conventionally grown McIntosh apples under another; and so forth. “Every tomato has an identity in our system,” Aamir notes. 

One cool way in which this translates into its service is that Fulfil also logs physical scans of each item. By pairing those images with the items’ specific codes, it enables online shoppers to choose the actual piece of produce they want. They can sort tomatoes by size, bananas by ripeness—and, in theory, a whole pallet’s worth, not just what’s visible in the display. This feature hasn’t yet been implemented, but it’s only a matter of time. “Technically it’s possible tomorrow,” Aamir says. “The technology is there.”

To let consumers choose specific products, Fulfil must update inventory by the nanosecond, and that innovation also has a side benefit: No more back-and-forth texts with shoppers after an order is underway, and no more well-meaning, split-second but often soul-crushing substitutions. What online grocery shopper hasn’t received gardenia-scented votive candles for the “4” and “0” birthday candles in their cart? Items go on hold, but orders are pulled minutes before pickup. Pickup occurs on the customer’s terms—in a few minutes, or later when convenient, as opposed to three prescheduled slots, all in the middle of tomorrow’s workday.

‍

Live in Mountain View

Fulfil’s first fulfillment center opened in November in Mountain View, California, in partnership with the grocery chain Save Mart. It’s available to shoppers via the site Luckynow by Fulfil. The startup’s setup is 6,500 square feet—call it the footprint of two to three suburban homes, but with inventory stacked to the building’s ceiling. According to Aamir, it’s a dark store whose sales volume is capable of reaching $15 million a year, rivaling a grocery store three to four times its size. 

‍

Fulfil’s modular concept means facilities can take any form: nanosized in a compact urban grocery store or scaled up to a sprawling rural effort. “This is something that touches every household,” Aamir says. “It’s a massive market where every bit of improvement is helpful to everyone.”

‍

Fast Company is a registered trademark of Mansueto Ventures LLC.

Fulfil Provides A Glimpse Into A Profitable Online Grocery Fulfillment Future

When it comes to online grocery fulfillment, the industry stands at a crossroads.

‍

When it comes to online grocery fulfillment, the industry stands at a crossroads. As online sales declined 7.6% YoY in March to $8bn, we are far a far cry from peak pandemic predictions of 20% penetration by 2025. As of last month, e-grocery pickup and delivery accounted for ~11% of total grocery sales according to Brick Meets Click’s monthly report as well as US Census MARTS data. This growth has coincidentally led to a pullback in automation, with companies like Fabric and Takeoff struggling to find retail partners as larger companies like Walmart and Kroger acquire or partner with Alert Innovation and Ocado, respectively. Meanwhile, recently unveiled stealth startups like Israel-based Super Duper and, more recently, Silicon Valley-based Fulfil have claimed that they have cracked the code to positive contribution margins, in some cases generating as high as 10% operating margins.

‍

Last week, HNGRY took a tour of Fulfil’s first Mountain View, CA robotic dark storeoperating a same-day grocery service in partnership with Save Mart’s Lucky bannercalled LuckyNow. The four-year-old startup, which has raised nearly $127mm at a~$267mm valuation according to PitchBook, claims it has surpassed over 50% of theoperating and capital expenditures of a manual-pick dark store on a blended basis. Onthe customer side, the value proposition is to get a weekly basket of groceries at instore prices with 100% accuracy in under an hour for a nominal delivery fee.

‍

‍

“Our robots store items more densely so we need a smaller building. Robots do most of the work so we require a much smaller staff than most grocery stores,” states LuckyNow’s FAQ. “For every item, our system tracks the freshness, expiration, supplier, and volume. Our algorithms let us drastically eliminate food waste. Dark stores just cost less than consumer retail stores. All of these factors allow us to pass the savings on to you.”

‍

The decision to automate depends on retailers’ online demand and their CFOs’ appetites for capex. Solutions like AutoStore tend to run in the range of $5-10mm in upfront capex and are designed to handle at least ~$25mm to $30mm of online orders per facility for an all-in per order cost of ~5-6% excluding delivery costs, roughly half as much as they would typically pay Instacart. According to Fulfil, grocers operating dark stores with manual labor to pick and pack orders could become ripe for automation at ~$5mm in annual sales. The startup makes money by charging retailers for hardware, services and recurring licensing fees and is currently looking to expand beyond its Lucky partnership to a myriad of grocers with varying desired levels of investment. At the low end, Fulfil says its partners could pick a lower-capex model that results in as little as an 18-month payback. For larger grocers with deeper pockets, this investment could be capitalized over a seven year life and result in much higher upside over a longer period.

‍

Whereas typical automated storage and retrieval systems (AS/RS) like Ocado, AutoStore, and Fabric rely on a “goods to person” model that uses humans to pack items from totes picked by robots, Fulfil is betting on a “robot to goods” one. This means that the only labor required is for induction, or in layman’s terms, breaking palettes of ambient, refrigerated, and frozen items to be stowed in automation. For many micro-fulfillment systems, additional labor is required to pick fragile or perishable items like eggs or produce from nearby shelves or inside an attached store. Fulfil boasts three fully-automated temperature zones where its 50 robots can pick everything from frozen pizzas to a carton of eggs. A typical order requires two bots picking twenty items in about six minutes. Just as in-store shoppers and gig workers fulfilling manual orders, Fulfil's robots are distributed in parallel throughout the store for maximum speed, throughput, and availability.

‍

‍

The system is fully modular, able to scale vertically and horizontally given a retailer’s space requirements. Prospective customers are evaluating a handful of different configurations ranging from dedicated warehouses to micro-fulfillment centers attached to existing stores. The facility I toured sits at ~6,500 sqft of automation and 8k sqft overall. According to Fulfil’s website, its system can handle as much as $10mm in GMV from a 5k sqft footprint all the way up to $60mm from a 20k sqft one. SKU counts can go as high as a typical grocery store ~30k.

‍

‍

The core modules of the Fulfil system are proprietary autonomous ground vehicles (AGVs) and dispense stations. These stations are comprised of trays, elevators, and pulleys that marshal products from vertically-stacked trays down to an AGV lined with a shopping bag. The startup uses computer vision to determine the picking order and specific location within each robot’s bag to gently drop each item using a mechanical pulley system. Each bot can adjust the height of the bag to ensure that fragile items can land safely within each shopping bag without falling inches to the ground. Computer vision is also used to track each individual item per SKU to ensure the sale of first-in-first-out inventory that ultimately lowers food waste.

‍

“Sensor fusion and computer vision algorithms control custom actuators and mechanisms to dispense items with precision,” states Fulfil’s website. “Cameras and a database inspect and track every item’s origin, location, and expiration profile for a new level of real time inventory that eliminates missing items and substitutions, and minimizes shrink.”

‍

‍

Fulfil’s LuckyNow service is currently delivering most of its orders in less than 30 minutes using a 10-mile, 25 minute driving radius. This delivery zone appears to be slightly cannibalizing online sales from three existing Lucky supermarkets while expanding fast delivery into new neighborhoods like Palo Alto and Mountain View. The service is currently available on DoorDash as well as its own first-party website built on its white-label ordering platform. DoorDash is currently subsidizing delivery, offering $1.99 delivery fees for orders over $10 as well as a 30% discount up to $50 off for new customers' first two orders.

‍

‍

Based on HNGRY’s analysis, the majority of items on LuckyNow were at parity with instore and Instacart prices with the exception of a handful of produce that was 17%-20% cheaper on LuckyNow’s first-party site and DoorDash listing. While the service is currently only available for delivery, Fulfil says it plans to offer customers the ability to order for curbside pickup in advance or in its parking lot. It has received positive driver feedback thanks to its API integration with DoorDash that requests deliveries as orders are picked to eliminate driver wait times.

‍

‍

While it is clear that Fulfil can work for mid and larger-sized baskets, the question remains to be seen what the line between average order value and speed will need to be to generate positive contribution margins for retailers. For example, it may not make much sense to use Fulfil to deliver a $20-25 top-off order of last-minute essentials in 30 minutes. Such orders would likely have to be batched with larger orders to lower down the variable fulfillment costs.

‍

To automate or not to automate? That… is the question.

‍

Disclosure: I am an investor in a limited partnership which owns a stake in Instacart and lack any information rights to the company