Should you run your own delivery riders? Do this maths first

"Should we get our own riders?" is usually argued with opinions. It is actually an arithmetic question with a clean answer, and the answer differs by restaurant.

The two cost structures

Marketplace delivery is a variable cost. You pay a percentage per order. Ten orders costs ten units; a thousand orders costs a thousand. It never gets cheaper per order, but it never costs you anything on a slow day either.

Your own fleet is mostly a fixed cost. A rider costs roughly the same whether they do eight deliveries in a shift or twenty-five. That makes it expensive per order at low volume and progressively cheaper as volume rises.

The whole decision is finding where those two lines cross.

Working out your break-even

You need three numbers.

1. Your commission cost per delivery order. Average order value × commission rate. At 1,400 and 25%, that is 350 per order.

2. Your fully loaded cost per rider shift. Not just wages. Include fuel or a fuel allowance, bike maintenance or rental, phone data, and any insurance. Say a shift costs 2,800 all in. TODO: use your own local figures here — these are illustrative.

3. Realistic deliveries per rider shift. This is where people get optimistic. Actual throughput depends heavily on how spread out your delivery area is. A dense urban radius might support 18–22 drops per shift; a spread-out area might be 10–12.

Take 14 as a middling figure.

2,800 ÷ 14 = 200 per delivery.

Against 350 of commission, in-house wins — provided you can actually keep that rider busy for the whole shift.

The trap: idle time

That last condition is where in-house delivery usually goes wrong.

If you have a rider on shift and only nine orders come in, your cost per delivery jumps to 311. At six orders it is 467 — worse than commission, and now you also own the scheduling problem.

The break-even is not about total volume. It is about volume density during the hours a rider is on shift.

A restaurant doing 400 delivery orders a month spread thinly across twelve hours a day is a worse candidate than one doing 250 orders concentrated into two clear peaks.

The middle path most restaurants should take

Very few restaurants should go all-in on either option. The sensible structure is usually:

• Own riders during your peak windows, where density is high and the per-delivery cost is genuinely low
• Marketplace or third-party for the off-peak tail, where you would otherwise be paying a rider to wait
• Own riders for repeat customers, where the relationship and the experience matter most

This keeps your fixed cost matched to your reliable demand and lets the variable cost handle the unpredictable part.

What in-house buys you beyond cost

Three things that do not appear in the arithmetic but matter:

The customer relationship. Your rider, your branding, your service standard at the door. It is one of the few remaining physical touchpoints.

Accurate delivery times. You control dispatch, so your estimates can be honest. Nothing damages repeat ordering like a promised 30 minutes that reliably takes 55.

Data. You learn your real delivery times by area, which lets you set zones and fees that reflect reality rather than hope.

What it costs you beyond money

Also worth being honest about:

You now manage people. Scheduling, absence, disputes, cash handling. That is a real management load on someone who is probably already stretched.

You carry the risk. Accidents, bike breakdowns, a rider quitting on a Friday. Marketplaces absorb that for you.

Slow days hurt. A fixed cost on a quiet Tuesday feels very different from a variable one.

A reasonable way to start

Do not commit to a fleet. Start with one rider during your two busiest windows, and route only your direct orders to them.

Track for a month:

• Deliveries per shift
• Actual cost per delivery
• Delivery time versus your marketplace average
• Whether repeat ordering from those customers changes

You will have a real answer in four weeks, based on your own numbers rather than anybody's general advice — and if it turns out not to work, you have unwound one rider rather than a fleet.