When a user taps order now on a food delivery app, a optimization engine decides which restaurant prepares the meal, which courier receives the offer, what route they drive, and how much they earn—often within milliseconds. The consumer sees estimated arrival time; the platform sees a constraint satisfaction problem involving supply, demand, geography, and labor cost minimization under service-level promises.
This article explains food delivery logistics analytically: dispatch graph construction, batching economics, surge pricing triggers, courier incentive programs, and failure modes (ghost orders, cold food, tip baiting perceptions). Examples generalize patterns across major US/EU platforms; exact parameters proprietary and city-specific.
Couriers experience the system as a sequence of thirty-second decisions; operations teams experience it as weekly EBITDA and customer retention curves. Both views are true. Closing that perception gap is prerequisite to fair debate about minimum pay standards, algorithm transparency, and whether flexibility rhetoric compensates for income variance workers cannot budget around.
Platform economies treat couriers as independent contractors while controlling task assignment tightly—a hybrid labor model with legal and ethical debates beyond this piece’s scope. Focus here is operational mechanics ordinary users and gig workers rarely see documented clearly.
Order Ingestion and Restaurant Side Timing
Order flow begins at restaurant tablet or POS integration receiving ticket with prep time estimate—manual or ML predicted from historical kitchen throughput. Platform adjusts quoted customer ETA dynamically as prep variance propagates. Late restaurant confirmation triggers reassignment or customer notification cascade.
Menu availability sync prevents ordering out-of-stock items; desync causes cancellation downstream hurting courier utilization. Inventory APIs imperfect; human restaurant staff overrides remain friction point.
Commission rates restaurants pay platforms often eighteen to thirty percent of order subtotal depending market negotiation—restaurant economics pressured, leading to inflated menu prices on apps vs in-store. Delivery fee customer-visible portion does not fully map to courier pay—platform retains spread.
Kitchen capacity acts hidden throttle—dispatch can flood couriers while grill line backs up twenty minutes, producing angry customers and idle drivers simultaneously in parking lots streaming dashcam boredom to TikTok. Optimizing only courier graph without prep truth yields bad ETAs no surge multiplier fixes.
Geographic zoning splits cities into cells; demand heatmaps color cells red during lunch peaks. Staffing incentives target red cells with surge pay to attract courier supply before ETAs degrade and customers churn.
Couriers see the offer—not the logic behind it.
Dispatch Algorithms and Courier Offers
Dispatch builds candidate assignments pairing available couriers near restaurant with pending orders. Objective function minimizes total customer wait while constraining courier idle time and avoiding excessive detours. Inputs include GPS location, vehicle type (car vs bike vs walker), historical acceptance probability, current batch load, traffic API estimates.
Couriers see offer card: guaranteed base pay, estimated tip, distance, merchant name sometimes blind until accept depending platform policy. Accept window short—often thirty to forty-five seconds—before offer routes to next courier. Decline or ignore affects internal acceptance rate metric influencing future offer quality opaque to worker.
Upfront tip visibility experiments changed courier behavior measurably—when tips hidden until after delivery, some markets saw higher acceptance on low-base orders; when shown upfront, cherry-picking increased leaving food idle on restaurant warming shelves. Platforms tune visibility like dial seeking equilibrium customers couriers restaurants rarely simultaneously happy.
Batching assigns multiple orders one trip if paths compatible—raises efficiency for platform, complexity for courier juggling hot bags and rating risk if second delivery delays first. Batch premium pay inconsistent; couriers debate fairness on forums with anecdotal evidence.
Ghost offer perception—offer disappears before accept—sometimes explained by restaurant cancel or algorithm reoptimization; sometimes app latency bug. Trust erosion quick when pay transparency lacking.
Effective hourly wildly variable; net after vehicle costs lower still.
Pay Structure, Surge, and Tips
Typical courier pay components: base delivery fee per order, distance supplement, time supplement in some markets, tip from customer, surge multiplier during demand spikes, challenge bonuses (complete twelve deliveries lunch window for extra forty dollars). Effective hourly wildly variable—fifteen to thirty-five dollars gross common US urban anecdotes; net after vehicle costs lower.
Surge algorithm raises pay when order-to-courier ratio exceeds threshold in zone—elasticity experiment tuning how much surge restores supply without overpaying. Surge displayed to customers as busy pricing increases order fees funding partial courier uplift—not always one-to-one pass-through.
Tips post-delivery on some platforms; pre-delivery tip prompts on others influence accept probability if couriers see tip amount before accept—policy changes periodically amid fairness debates. Tip baiting—high tip shown then reduced—reported behavior erodes morale; platforms implement tip integrity policies with uneven enforcement.
Prop twenty-two style ballot measures and local minimum pay standards (NYC delivery worker minimum per hour active time) reshape net pay floors regionally—legal layer atop algorithmic layer workers must track separately.
Multi-apping is common—workers cherry-pick the best offer across platforms.
Utilization, Ratings, and Deactivation Risk
Platform metrics track courier acceptance rate, completion rate, customer rating average, on-time percentage, fraud flags (GPS spoofing, never delivered claims). Fall below thresholds triggers warnings or deactivation—effectively termination without employment benefits. Metric weighting opaque; appeals process variable.
Utilization target maximizes deliveries per courier hour to spread fixed onboarding costs. Idle couriers cost nothing directly but may switch apps; multi-apping common—workers run DoorDash, Uber Eats, Instacart simultaneously cherry-picking best offer.
Insurance and liability: accidents while delivering create coverage gaps between personal auto policy and platform contingent insurance layers. Workers often discover limits after incident—financial risk externalized.
Weather spikes demand and accident risk simultaneously; surge may not fully compensate added time and hazard—couriers describe storm shifts as high gross, negative net emotionally.
Customer rating systems originally designed to police courier behavior now double as restaurant quality proxy and address accuracy score—courier punished for restaurant packing soup without tape because stars attach to driver profile visible next offer. Reputation portability across apps limited; deactivation on one platform does not always block others, but local courier communities share warning lists informally.
System Failures and Optimization Side Effects
Failure modes include restaurant never started order, courier assigned before food ready causing idle parking lot wait unpaid, wrong address apartment complexes, customer unavailable leading to waste disposal protocols. Each scenario has playbook; playbooks imperfect—support chat bots frustrate urgent field problems.
Algorithmic optimization reduces average ETA year over year industry-wide yet individual experience variance huge—tail latency matters psychologically more than mean. One forty-five minute late pizza erases memory of ten on-time orders.
Environmental externalities—packaging waste, vehicle miles—accumulate at city scale; platform sustainability pledges partially address with optional utensil opt-out and bike courier subsidies in dense cores.
Understanding logistics clarifies courier grievances not as isolated complaints but predictable interaction with dispatch objectives misaligned worker stability—piecework flexibility vs income predictability tradeoff structural not accidental.
City geometry shapes earnings more than individual hustle. Dense downtown grids with short restaurant-to-apartment distances produce more deliveries per hour; sprawling suburbs with highway strips pay similar per order but consume windshield time unpaid. Couriers learn micro-geography—where restaurant prep is honest, which apartment complexes lack door codes, which corporate campuses security delays kill ratings. That knowledge is skilled local labor platforms treat as interchangeable until a worker leaves and ETAs degrade.
Food delivery platforms run real-time logistics engines treating human couriers as stochastic mobile nodes in a graph optimization problem. Surge, batching, and metrics translate business goals into individual shift experiences invisible on customer tracker map.
Seeing mechanics helps couriers strategize multi-app usage and zone timing—and helps customers understand why tips and honest addresses materially affect another person’s hourly reality.
Restaurant prep delay is the silent variable couriers cannot control yet absorb in ratings when food arrives cold. Some platforms begin paying wait time after threshold minutes; thresholds differ by city and change quarterly in ways support articles rarely announce clearly. Reading courier forums is occupational literacy official help centers omit.