
Every delivery capacity decision gets judged on the wrong invoice. Reserve too many rider-hours and the waste is itemized: idle time, billed every month, easy to circle in a review. Reserve too few and the cost never appears as a line item. It surfaces weeks later as customers who quietly order less after the evening their 60-minute order took two hours.
That imbalance pushes many D2C brands towards the same mistake. They size capacity to an average day, then break the promise in the hours that matter most.
The short answer: reserve delivery capacity up to the level of demand at which one more unit of unused capacity costs as much as the expected loss from one more missed delivery promise. Work it out separately for each city and hour band, keep planned utilization below the point where your on-time rate starts to fall, and decide in advance what happens to orders above the line.
What follows is a practical way to decide how much delivery capacity to reserve for quick commerce fulfilment on your own website, built on established operations research and illustrative numbers rather than universal benchmarks.
Planning around the daily average fails because delivery capacity is consumed inside the promise window, not across the day. A brand averaging 400 orders a day can still run short at 8 pm if a third of those orders land between 7 and 9, and spare riders at 11 am cannot deliver a 60-minute order placed that evening.
Averages also hide variability. Researchers who modelled last-mile fleet sizing as a cost trade-off found that, in their stylized example, sizing the fleet to expected demand raised total cost by 8.2% compared with sizing to the full demand distribution. In a second example, separate fleet sizes for weekdays and weekends cut cost by 41.4% against one fleet size for the whole week. Both results come from simulated next-day delivery in Chile, so treat them as directional. The lesson still travels: the shape of demand, not its average, sets the right amount of capacity.
The same research shows why there is no universal capacity number. It models delivery time as a function of the area served, the number and density of drops, the time spent at each doorstep and the distance from the depot. Your requirement therefore depends on how spread out your customers are, order density around each dark store, rider productivity, delivery distances, the promise you make and how volatile your demand is. Two brands with identical daily volumes can need very different reservations.
Brands should treat capacity as a trade-off between two costs and reserve up to the point where those costs balance. This is the logic of the newsvendor model, a long-established operations research tool for deciding how much stock to hold before demand is known, which researchers have recently applied to last-mile fleet sizing. Adapted for delivery capacity planning for D2C brands, it becomes a simple reservation ratio:
Reserve enough capacity to cover the demand percentile equal to Cu ÷ (Cu + Co)
| If a missed promise costs | Reserve for this percentile of demand in the hour band |
| The same as an idle unit (1:1) | 50th |
| 2 times an idle unit | 67th |
| 4 times an idle unit | 80th |
| 9 times an idle unit | 90th |
| 19 times an idle unit | 95th |
These percentiles are the arithmetic of the ratio, not operating benchmarks. Your answer depends entirely on your own costs.
Co is easy to find because it sits on the invoice. Cu is where most brands go wrong, because the lost margin on the missed order is only part of it. A study of an online retailer found that after a fulfilment delay, previously loyal customers ordered less often and in smaller amounts. A 2025 study of an omnichannel grocer found that a failed fulfilment delayed the customer's next order by 7.22% on average. A missed promise taxes future orders, which is why a kept promise often matters more than a faster one. That tax never appears on a delivery invoice, so brands that price only the visible cost under-reserve consistently.
The figures below are invented for illustration only. Suppose an unused unit of reserved capacity costs ₹40. A missed 60-minute promise costs ₹160: ₹100 of contribution margin on the order plus ₹60 as a probability-weighted share of the future orders that customer no longer places. The ratio is 160 ÷ (160 + 40) = 0.8, so you would reserve enough to cover the 80th-percentile evening hour.
Now add a fallback. If a third-party option can still deliver an overflow order on time 90% of the time for ₹50 extra, the expected cost of under-reserving falls to (0.9 × ₹50) + (0.1 × ₹160) = ₹61. The ratio drops to 61 ÷ 101, about 0.6, and the right reservation falls to roughly the 60th percentile. The reliability of your fallback matters as much as the price of your reservation.
The ratio is a reliable anchor when these conditions broadly hold:
Where these assumptions break, use the ratio as a starting point and apply judgement on top.
As a delivery operation approaches full load, delays rise sharply rather than gradually. Queueing theory illustrates this through Kingman's approximation, which estimates average waiting time in a single-server queue as the product of a variability term, the service time and a utilization term, ρ ÷ (1 − ρ). In that model, moving utilization from 80% to 90% more than doubles the utilization term (from 4 to 9), and moving from 90% to 95% roughly doubles it again (to 19).
Those figures cannot be lifted onto a delivery fleet. Kingman's approximation describes one server in a steady state; a fleet is many riders working in parallel, with travel, batching and changing routes, and a pooled fleet can usually run at higher utilization before waits build. So 80%, 90% and 95% are not rider utilization targets. What transfers is the shape of the curve: near full load, delays escalate non-linearly, and a 30-minute or 60-minute promise has little room to absorb them.
The practical step is to find your own ceiling. Plot hourly on-time delivery rate, one of the fulfilment KPIs every D2C ops lead should track, against hourly utilization for the last few months, city by city. Where the on-time rate starts to fall away is your ceiling. Reserved capacity then becomes the demand level from the reservation ratio, divided by that ceiling.
Brands should reserve a higher share of expected peak demand and book it earlier, because overflow capacity is scarcest exactly when it is needed. The usual advice to flex for peaks assumes extra riders are available at a modest premium. In India's festive season, that assumption is at its weakest.
Adecco India expects a 10 to 15% demand-supply gap in frontline hiring during the 2026 festive peak, most pronounced in major metros, alongside 30 to 35% hiring growth in logistics and last-mile delivery. In October 2025, Mint reported that companies employing gig workers had raised incentives by 15 to 20%. When fallback capacity becomes scarcer and costlier, Cu rises, and the reservation ratio pushes the right reservation up.
Return to the illustrative example. If festive fallback reliability drops to 60% and its premium rises to ₹80, the expected cost of under-reserving becomes (0.6 × ₹80) + (0.4 × ₹160) = ₹112. The ratio rises to 112 ÷ 152, about 0.74, and the right reservation moves from roughly the 60th to the 74th percentile. Same brand, same margins, different answer, because the fallback changed.
Two practical implications follow:
Orders above your reservation get whatever you decided in advance, or whatever improvisation produces on the night. Every reservation deliberately leaves some demand above the line, so those orders need a planned answer. That is the job of the three-layer capacity model, the central practical takeaway from this framework.
| Layer | What it covers | How to size it | When to commit | What it costs |
| 1. Baseline reservation | Normal weekly demand, by city and hour band | Reservation ratio for each hour band, divided by your utilization ceiling | Monthly review, resized at least quarterly | Some idle capacity in quiet weeks, by design |
| 2. Event layer | Known peaks: sale days, festive weeks, launches | Hourly curve of the last comparable event, adjusted for growth and promotion depth | Weeks ahead, linked to the marketing calendar | Paid for whether or not the event performs |
| 3. Overflow and promise control | Demand above layers 1 and 2 | A priced fallback where it is reliable; otherwise an adjusted promise at checkout | Live, hour by hour | A fallback premium, or some lost conversion on a slower promise |
Layer 3 is where most brands improvise, and accepting a 30-minute order you cannot fulfil is the worst outcome available: it turns a sale into the missed promise that costs future orders. Showing the next deliverable window at checkout, such as a 2-hour or same-day option, trades a little conversion for a promise you can keep. That window must be real, the same discipline behind deciding whether to show a delivery date on your product page.
The model also gives you a review discipline:
Zippee provides Quick Commerce-as-a-Service infrastructure to 100+ consumer brands through a network of 140+ dark stores across 21+ cities, powering 60-minute, 2-hour and same-day delivery on brands' own websites. Its delivery partners are full-time employees rather than gig workers, which makes delivery capacity something brands can plan and audit rather than hope for on a busy evening. We have examined the trade-offs of that choice in employed riders versus gig riders. Because orders are placed on the brand's own channel, the delivery promise, and how it is kept, stays the brand's decision.
The right amount of delivery capacity is not a number you can borrow from another brand. It is the output of inputs you own: the cost of idle capacity, an honest cost of a missed promise, the reliability of your fallback and the shape of your demand by city and hour. Get those right, keep utilisation below the point where your on-time rate breaks, and decide what happens above the line before the evening you need it.
If you're ready to turn your fulfilment into a competitive advantage, join our waitlist.