Pick the unit, then sort every cost

The unit is the thing you sell or serve again and again. For a delivery app it is an order; for ride-hailing, a trip; for a subscription, a subscriber-month; for wholesale, a delivery to a shop. Choose the unit where decisions are made. If you set prices and fees per order, analyse per order.

Then sort every cost into one of three groups:

• Variable costs come with each unit: the goods in the basket, the rider's pay for that delivery, the bag, the payment fee, any discount.
• Fixed costs stay the same for the month whatever the volume: rent, the store manager's salary, software licences.
• Step costs are fixed until you cross a threshold, then jump: a second picker on the evening shift, another rider shift on a rainy night, a bigger delivery van.

Contribution margin is revenue per unit minus the variable costs per unit. It is what each unit contributes towards the fixed costs. Once the fixed costs are covered, contribution becomes profit. It differs from gross margin, which only takes away the cost of the goods. A grocery basket can have a healthy gross margin and still lose money once you pay for the ride to the customer's door.

One rule follows directly. If the contribution margin is negative, more volume makes the loss bigger. Spreading fixed costs over more orders helps only when each order earns something to spread.

One "Now" order, line by line

Mangrove's busiest city has a dark store: a small warehouse, closed to shoppers, that only picks 30-minute "Now" orders for the app. It handles about 600 orders a day. Tomás, the data analyst, pulled last month's average order:

Basket value                         $18.00
Goods margin      18.00 x 22%       + $3.96
Delivery fee charged                + $1.50
Rider cost                          - $3.60
Packaging                           - $0.40
Payment fees      2% x (18.00 + 1.50 - 1.50)  - $0.36
Discount (promo codes, first orders) - $1.50
Contribution margin per order        - $0.40

Now add the fixed costs. The dark store's rent, power, pickers and supervisor cost about $18,000 a month. Break-even volume is fixed costs ÷ 30 days ÷ contribution margin per order. With a margin of −$0.40 there is no break-even: at 600 orders a day the store loses about $25,200 a month, and at 1,200 it would lose more. If the discount were removed, the margin would be +$1.07 and break-even would be about 561 orders a day. At 600 orders the store would make about $1,260 a month. Positive, and very thin.

Averages hide the story, so split the margin by segment. Tomás grouped the month's orders by basket size:

The rider and the bag cost the same whatever is in the bag, so a $7 basket can never pay for its own trip with a $1.50 fee. Small baskets are where the losses are. Other useful splits are time of day (rain and dinner peaks need more riders), distance, new against repeat customers, and payment type.

Real quick-commerce companies have lived this. Blinkit, the quick-commerce arm of India's Zomato, reported a contribution margin of −7.3% of gross order value in the July to September quarter of 2022. A year later it reported +1.3%, with average order value up to about 607 rupees, and it had first turned contribution-positive in June 2023. Getir went the other way: in April 2024 it said it would leave the Netherlands, Germany, the UK and the US, markets that were about 7% of its sales and heavily loss-making, and focus on its home market, Turkey.

What moves each line

Each line has its own levers, its own owner and its own side effects. Write them down before anyone argues about which to pull:

Two of these hide a mechanism worth knowing. Rider cost per order is rider cost per hour ÷ orders per rider-hour. In the dark store's area, riders are paid about $7.20 for each hour on shift and complete about 2 orders an hour, so each order costs $3.60. If picking is slow, riders stand at the counter waiting, and fewer orders fit in each hour. That is why the Store app's picking screens are a unit-economics lever: get riders 2.25 orders an hour and the cost falls to $3.20, which on its own brings the order to break-even.

Payment fees often have a fixed part. Stripe's standard price for US domestic cards is 2.9% + 30¢ per payment (checked October 2026). On a $50 basket that is about 3.5%; on a $7 basket it is about 7.2%. Fixed per-payment fees, like riders, punish small baskets most.

Customers: CAC, LTV and payback

Orders come from customers, and winning a customer costs money. Three numbers describe the customer layer:

• CAC (customer acquisition cost) = acquisition spend ÷ new customers it brought. Count paid channels against the customers they brought; customers who arrived on their own don't belong in the divisor.
• LTV (lifetime value) = contribution margin per order × orders per month × expected lifetime in months. With steady monthly retention r, the expected lifetime is about 1 ÷ (1 − r) months.
• CAC payback = the months of contribution it takes to earn back the CAC.

Mangrove pays about $12 to win a customer. Customers order about 3 times a month, and 80% of them are still ordering the next month, so the expected lifetime is 1 ÷ 0.2 = 5 months. With today's −$0.40 margin, LTV is −$0.40 × 3 × 5 = −$6: every customer won loses $18 in total, counting the $12 CAC. With the discount removed, LTV is $1.07 × 3 × 5 = $16.05, and LTV:CAC is about 1.3.

The curve flattens because retention compounds: after 6 months only about a quarter of the customers are left. That shape explains a common surprise. Paying more to acquire customers looks safe while you watch the early months, but if CAC rises above about $16, the curve never gets there.

Three LTV mistakes come up again and again, and Bill Gurley's 2012 essay on "the dangerous seduction" of the LTV formula lists them:

1. LTV on revenue. $18 × 3 × 5 = $270 looks wonderful and means nothing: Mangrove keeps a few cents of each dollar. Always use contribution margin.
2. Counting free customers in CAC. Dividing paid spend by all new customers, including those who came through word of mouth, makes paid channels look cheap.
3. Assuming everything improves together. Raising fees tends to lower retention, and spending more on ads tends to raise CAC. Gurley's point is that these variables pull on each other, so a forecast where they all get better at once is unlikely to hold.

Rules of thumb come from software companies: David Skok's widely used guide says the best SaaS businesses have LTV more than 3 times CAC, often recover CAC in 5 to 7 months, and look weak if payback takes more than 12. Treat these as a starting bar. Grocery delivery has thinner margins and more frequent orders, so finance teams set their own bar. Mangrove's finance team asks for 3 : 1 before scaling paid marketing. Retention curves that flatten, which you met in pm-23, are a better input than a single retention rate once you have cohort data.

Unit economics for internal tools: cost to serve

Internal products have no revenue, but every unit of work they handle has a cost. Cost to serve is that cost per unit: per partner order, per goods received note (GRN), per wholesale delivery, per support ticket. A good internal tool lowers it, and the product's value is roughly the cost per unit saved × volume, plus the errors it prevents.

Here is the Partner app's version. About 4,000 partner orders arrive a week, and about 1,500 of them still come by phone to Mangrove's 40 sales reps. Mangrove's numbers for each channel:

Phone order   rep time 10 min x $6/hour (fully loaded)             $1.00
              about 1 in 12 corrected after delivery x $9 per fix  $0.75
              cost to serve                                       $1.75
App order     support, servers and payment handling               $0.20
Difference per order moved to the app                             $1.55

Target: app share from 62% to 75% = about 500 fewer phone orders a week
500 x $1.55 x 52 weeks = about $40,000 a year

Two cautions keep this honest. First, the $1.00 of rep time becomes real money only if the reps' hours go to something that pays, such as signing new shops, or if hiring slows. Say which, and who decided. Second, the corrections are a real cost today: credit notes, a second delivery, and an annoyed shop owner. Count those even when nobody's salary changes. The business case lesson, pm-28, turns numbers like these into a decision a managing director can sign.