Which costs belong to one order

You met the method in pm-27: the average 30-minute "Now" order has an $18 basket and, after goods, rider, packaging, payment fees and a $1.50 discount, loses $0.40. Without the discount, a repeat order earns $1.07. The finance question is which costs belong in that line.

Every cost falls into one of three groups for a given decision:

• Variable costs come with each order and disappear without it.
• Fixed costs stay the same this month whatever the volume: engineers' salaries, rent, software licences.
• Allocated costs are slices of shared costs handed out by a rule, such as share of revenue or orders. The total is real money; the slice stays when an order goes.

Here is how that goes wrong at Mangrove. In fin-01 you saw technology opex of $24M in FY2025. A slide in last year's budget pack divided it by about 21.6 million app orders a year and called the result "technology cost per order": about $1.11. Next to the $1.07 repeat order, every app order seems to lose 4 cents, and the slide concluded that app growth destroys value.

Three things are wrong with it. First, most of the $24M is staff, licences and network, which stay the same if Mangrove takes 1,000 fewer orders tomorrow. Second, the same technology runs 320 stores and the Partner app. Divide the $24M by every order and store checkout (about 146 million a year) instead and the "cost per order" falls to about $0.14. When the allocation key decides the answer, the answer is not about the order. Third, the technology cost that really grows with orders is small: the cloud services behind the customer app cost about $210,000 a month for about 1.8 million orders, about $0.12 an order.

That $0.12 is the marginal cost of technology for an app order, or a little less, since part of that cloud bill is a base load that runs at 3 am too. It belongs in contribution, and pm-27's order line left it out. With it, the repeat order earns about $1.07 − $0.12 = $0.95, the number we use below.

The rule follows. Decisions about volume, price and channels use contribution, built from variable costs only. Fixed costs are compared in total with total contribution, at the level where someone could cut them: a team, a store, a product line.

A partner order, line by line

About 4,000 partner orders arrive a week from corner shops, about 62% of them in the Partner app and about 1,500 by phone to the sales reps. The average order is about $250, and Mangrove's wholesale margin is about 7%, so each order leaves $17.50 of gross margin. Delivery from the distribution centre (DC) costs about $11 per drop. Then come the costs that depend on the channel, which pm-27 measured as cost to serve: $0.20 for an app order and $1.75 for a phone order ($1.00 of rep time plus $0.75 for the 1 phone order in 12 corrected at about $9).

One more cost hides here. Shops have 14 days to pay, and on average they pay after 23. Until then Mangrove holds a receivable instead of cash, money that cannot pay a supplier or reduce a loan. Mangrove's finance team charges money tied up at 12% a year, its rate for projects. So the cost of credit on one order is $250 × 12% × 23 ÷ 365, about $1.89.

One $250 partner order               App       Phone
Wholesale margin (7%)              +17.50     +17.50
Delivery drop from the DC          -11.00     -11.00
Cost to serve: handling             -0.20      -1.00   (phone: 10 min of rep time)
Cost to serve: corrections           0.00      -0.75   (1 in 12 x $9)
Contribution before credit          +6.30      +4.75
Cost of credit (23 days at 12%)     -1.89      -1.89
Contribution after credit           +4.41      +2.86

Two things stand out. The drop is the biggest cost, so route density moves partner economics more than anything in the app. And the cost of credit is larger than the whole phone-versus-app difference. It is not an operating cost in the P&L, since interest sits below operating profit, but 4,000 orders a week at $1.89 is about $390,000 a year, roughly the $3M of partner receivables on the balance sheet at 12%.

Now total it by channel for a normal week:

That is about $800,000 a year after credit, which has to cover the reps' salaries, the Lagoon Digital contract (about $390,000 a year) and the Partner app team. Moving 500 phone orders a week to the app adds about $40,000 a year, as pm-27 found. Getting shops to pay on their 14-day terms would save 9 days of credit, about $0.74 an order or roughly $150,000 a year, so paying in the app on delivery is a product feature with a finance payoff.

CAC by channel and payback

CAC is a channel's acquisition spend divided by the customers it brought in, and CAC payback is the months of contribution it takes to earn that back. pm-27 used a CAC of about $12 for the app. Tomás split last month's paid acquisition by channel:

Payback uses pm-27's app customer (3 orders a month, 80% still ordering each month) at $0.95 an order. In month 1 a customer earns 3 × $0.95 = $2.85. In month 2 only 80% of the cohort is left, so expected contribution is $2.28, then $1.82, and so on. For a $12 CAC the running total passes zero in month 9, and it can never pass 3 × $0.95 × 5 = $14.25, because the expected lifetime is 1 ÷ (1 − 0.8) = 5 months. Social ads at $19 never pay back. A blended CAC of $12 makes all three look acceptable.

The quick version, "CAC ÷ monthly contribution", gives $12 ÷ $2.85, about 4.2 months, because it assumes nobody leaves. The cohort payback takes more than twice as long, and it is the one finance teams use.

A new partner shop works the same way, with bigger numbers and a ramp. Signing one shop takes a rep about 10 hours, counting visits to shops that say no: about $60 at $6 an hour fully loaded. Add a $30 welcome credit on the first order (counted here, so it stays out of that order's contribution) and CAC is about $90. New shops place orders in the app and start small: about 3 orders in month 1, 5 in month 2, 7 in month 3, then about 9 a month like an average shop (4,000 orders a week across 1,800 shops). About 98% of shops still order each month.

New partner shop, app orders at $4.41 after credit, CAC $90
Month  Orders  Shops left  Contribution  Cumulative
  1      3       100%        $13.23       -$76.77
  2      5        98%        $21.61       -$55.16
  3      7        96%        $29.64       -$25.52
  4      9        94%        $37.35       +$11.83   pays back
 12      9        80%        $31.78      +$284.98

A partner shop pays back far faster than an app customer, because it brings about $40 of contribution a month against $2.85. That argues for signing shops. The limit is that the 1,800 shops Mangrove serves were probably the easiest to sign, so the next shop usually costs more to win and may sit off the usual routes.

Cohort LTV, with and without discounting

LTV is the total contribution you expect from a customer over the time they keep buying. Build it from a cohort: all customers who joined in the same month, followed month by month, as in pm-23. Each month's contribution is orders × contribution per order × the share of the cohort still buying.

A dollar in month 30 is worth less than a dollar today, which could repay debt or fund a project earning 12% a year. Discounting divides each future month by (1 + r) to the power of the month. A 12% yearly rate is about 0.949% a month, since 1.12 to the power of 1/12 is about 1.00949.

                         Undiscounted   Discounted at 12%   Drop
App customer (5 months)     $14.25          $13.60           about 5%
Partner shop (lifetime)     about $1,930    about $1,290     about a third
Partner shop (36 months)    about $970      about $830       about 15%

For the app customer, discounting barely matters, because most of the money arrives in the first months. For a partner shop that stays for years (98% monthly retention means an expected lifetime of about 50 months), it removes about a third. Long lifetimes are also where the forecast is weakest, so many finance teams cap LTV at a horizon such as 36 months. Discounted and capped, a new shop is worth about $830 against a $90 CAC, and that is the number to bring to a budget meeting.

Three mistakes turn LTV into a sales slide:

1. LTV on revenue. $250 × 9 orders a month × 50 months is $112,500 per shop, and means nothing when Mangrove keeps a few dollars an order. Use contribution after credit.
2. Blended CAC. One CAC across channels, or across old cities and new ones, hides the channel or city that never pays back.
3. Averages that hide segments. The average shop pays in 23 days; a phone shop that pays at 45 days keeps about $1.05 an order instead of $2.86. The average drop costs $11; a shop off the usual routes may cost $14. Split by channel, payment days, city and route before you trust an average.