Engineering Last Mile Delivery

What happens when you take an engineer trained on jet engines, run him through Tesla’s Model S and Model X production hell, let him watch startups from the investor side, and then drop him into the last mile? You get Adam Bryant, CEO of Jitsu, a technology-driven carrier rethinking how e-commerce packages move from warehouse to doorstep. Bryant did not start in parcel. He designed engines at GE Aviation, worked at Danaher and Proterra, joined Tesla in 2012 the week the first Model S rolled off the line, and later became a partner at Eclipse Ventures, where he invested in Jitsu and sat on its board. He joined the company full time as an operating partner, and two weeks later the pandemic hit. The rest, as he puts it, is history.

From investor to operator

Bryant’s interest in Jitsu started with the founder, Daniel Sokolovsky. Sokolovsky grew up in a courier business in Los Angeles, worked with paper-based routing and manual processes, studied applied math at Berkeley, and started the company right out of school. When Bryant met him, Jitsu was doing low double-digit millions in annualized revenue, bootstrapped, having raised far less capital than a typical venture-backed startup. Bryant calls that an anomaly in Eclipse’s pattern of full-stack, capital-intensive investments, and that made it interesting. What hooked him personally was the Uber model, dynamic matching of supply and demand in an asset-light structure, applied not to point-to-point food delivery but to batched, aggregated package delivery. FedEx and UPS had done batching for decades, but nobody had combined it with gig-economy supply that could flex up and down. And since he had never worked in parcel, Bryant says, he came in without legacy handcuffs and asked the stupid questions freely.

The problem Jitsu is solving

The national carriers, in Bryant’s words, are expensive and rigid. Shippers have a fundamental need to cut costs because shipping is a major line item, to drive quality and performance that build customer lifetime value, and to find flexible, responsive partners that grow with them. Jitsu did not start by attacking all of e-commerce. It started with perishables: next-day delivery, cold chain constraints that made traditional hub-and-spoke unusable, and shippers willing to inject freight directly at the final node. That leveled the playing field because the company did not need an extensive network on day one. Only around 2021, after growing through the pandemic, did it expand into broader e-commerce like apparel and beauty, which brought new problems: smaller parcels, less predictable manifests, and brutal peak-season spikes. That evolution forced the company to add coverage, connections, and utilization tricks so clients could fill trucks together.

Tesla lessons, applied to parcels

Bryant draws direct parallels between taking on the auto incumbents and taking on UPS and FedEx. Tesla could not beat GM, Ford, and Toyota head-on, so it started small: the Roadster, a niche sports car at $100,000 to $200,000, then the Model S, then the Model X, and only on its fourth car the mass-market Model 3. Starting with mass market first would have bled the company to death. Perishables were Jitsu’s Roadster, a smaller, demanding market with a specific profile where the company built its systems before expanding. The operating lessons he carried over are specific. Move with urgency: Tesla ran on something closer to a 50-50 rule than 80-20, valuing speed and fast learning over caution. Reason from first principles: when someone said “Volkswagen did it this way,” the answer was “I don’t care what Volkswagen did,” and you design from the atom level with today’s tools. Set stretch goals: the annual build plans looked impossible, teams often missed them, but got far closer than they would have with comfortable targets. And put relentless pressure on bottlenecks: during Model X launch, when Falcon door quality stalled the line, Bryant was not allowed to cut the build plan, so material kept flowing until about a thousand tractor trailers sat in the parking lot, forcing the door problem to get solved. Vertical integration was the Tesla answer to suppliers who could not keep up. Jitsu’s version is owning the end-to-end chain: its own WMS, its own routing, its own driver app, plus line haul. Bryant compares a delivery network to dominoes: change the sort process and it hits the driver. Owning the whole system means the company understands those linkages and moves fast instead of waiting on a third party.

Reliability, speed, and unit economics

Bryant treats all three as non-negotiable, just improving at different speeds. On unit economics, he says the biggest lever is utilization and aggregation: search for it everywhere. The company deliberately skipped the 45-minute delivery fad because instant delivery destroys density; your neighbor does not want a package in the same hour you do. Fixed costs are kept to a minimum, the gig model flexes up and down, and expansions must prove demand first rather than layer on capacity ahead of it. Speed comes from throughput and cycle time discipline, making sure sorts clear for dispatch, plus prioritization, since clients have different SLAs and not everything must go out today. Reliability comes from measurement: everything is measured, root-caused, and stacked on an 80-20 Pareto, with redundancy built in on driver supply and sort waves.

The 99 percent on-time claim

Jitsu advertises 99 percent on-time delivery, and Bryant admits the team sometimes cringes at the number because so many providers embellish it. But he says the perishables era built it into the DNA: miss a food delivery and the product perishes, which means expensive claims. That was, as he puts it, a very hard punishment. Sustaining it rests on three things. First, measure it daily and share it with clients; you cannot improve what you do not measure. Second, the asset-light gig model creates latent supply in the wings, and demand data lets the company crystallize much of the next day’s plan before the day starts, with reaction time to pull levers if supply looks short. Third, proprietary tools handle the surprises: unmanifested parcels that just show up, late trucks, bad proof-of-delivery photos, fat-fingered addresses. Redundancy through second sorts and cascading driver waves covers the rest.

Where AI actually helps

On artificial intelligence, Bryant splits the work into three buckets. Today: AI analyzing driver proof-of-delivery photos to catch a wrong-doorstep photo in the moment and nudge the driver before the delivery fails, and customer support automation for the simple questions like where a package is. The one he is most excited about: demand prediction, taking all the noise (promotions, historical data, late trucks, seasonal spikes, orders trickling in all day) and extracting the signal to predict what will show up three hours from now. Autonomous final-mile delivery, he thinks, is a ways out. A 53-foot truck on a highway is a simpler problem than the last mile, which he calls edge case city: barking dogs, bad access codes, cracked sidewalks, people. His CTO’s framing for AI work: make every human minute count, automating the administrative and repetitive so humans do what they do best. That also feeds Jitsu’s internal “mom test”: the driver app should be so intuitive that you could hand a route to your mother and she would succeed on the first delivery.

The regional experience at scale

Looking a decade ahead, Bryant’s hope is that Jitsu gives shippers “a regional experience at scale,” the ultimate easy button. The old easy button is handing your customer experience to a rigid duopoly with a Henry Ford menu: any color you want as long as it is white. His North Star is the carrier as an extension of the shipper’s team, using modern tools like smartphones and real-time data to prevent a claim from happening in the first place rather than arguing about who pays for it afterward.

---

This post is based on the Logistics Remix episode “Engineering Last Mile Delivery” with Adam Bryant, CEO of Jitsu. The views are the guest’s and the host’s own.