Real-world application of CAC, LTV & ROAS

In the first quarter of 2017, Blue Apron served roughly 1,036,000 customers. It never served more. The company had spent about $144 million on marketing in 2016, would spend around $155 million in 2017, and closed that year with something like 750,000 customers. Eighteen months of the biggest marketing budget in the meal kit category bought a smaller customer base than the one it started with.

This lesson stays inside that single business: a subscription model whose acquisition costacquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition → passed the value of what it was acquiring, with the evidence sitting in cohort data that was public before the shares were. Take the three metrics and the payback window as the foundations lesson sets them out. What matters here is which one moved first, how long it stayed visible before anyone acted, and what the delay cost.

The arithmetic in the S-1

Divide the disclosed marketing spend by new customers and Blue Apron's acquisition cost sat near $94 across 2016 and around $147 in the first quarter of 2017. On the other side of the ledger, the filing showed an average order value of roughly $57, about four orders per customer per quarter, and average revenue per customer near $236 a quarter. Gross marginGross marginGross margin is the share of revenue left after subtracting the direct cost of producing goods or services, expressed as a percentage of revenue.View full definition → ran around a third.

So an active customer threw off roughly $75 of gross profit per quarter. At $147 to acquire, the customer needed to keep ordering for two full quarters before the company saw a cent, and that is before product, technology and general overhead, which Blue Apron carried as separate lines. Two quarters of ordering is not a demanding assumption in software. In a weekly-delivery food business where the customer has to cook the thing, it is a heroic one.

Retention was the variable, and the average hid it

Third-party card panel work in 2016 and 2017 put the share of Blue Apron customers still ordering six months after signup at roughly a third. The median customer, in other words, never reached the point where acquisition cost was recovered.

Here is the reporting trap, and it is the part most teams miss. Average revenue per customer is calculated on customers who are still there. As the weakest half of a cohort leaves, the average per remaining customer holds steady or ticks up, because the survivors are the enthusiasts. The line on the slide looks stable while cohort value is falling underneath it. A flat ARPC alongside a flat or shrinking customer count is not a plateau, it is churn being masked by selection. Blue Apron's per-customer revenue was never the alarming number. The alarming number was how few customers the metric was being averaged over.

The discount that never showed up in CAC

Meal kit acquisition ran on first-box offers: $30 or more off the first delivery, later stacked across the first two or three. That has an accounting consequence with teeth. Promotional discounts are usually booked as a deduction from revenue, not as marketing expense. The discount therefore lands in gross margin, not in the acquisition cost line.

The practical effect is that reported CAC understates the real cost of a customer, and gross margin absorbs the difference, so both numbers look better than the cohort deserves. On Blue Apron's structure, a $40 first-box incentive against roughly $75 of quarterly gross profit means the first delivery contributes close to nothing. The cohort only crosses zero somewhere around the third or fourth box, and most of it is gone before then. If your promo budget is contra-revenue, add it back before you compare acquisition cost to anything.

What reactivation did to the customer line

Blue Apron reported total customers, a figure that mixes fresh cohorts with win-backs. Reactivated customers are cheap to reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → and respond to discounts, which makes the headline count easy to defend for a quarter or two. They also tend to churn faster than a first-time cohort, because they already quit once for a reason the product has not fixed. A customer line propped up by reactivation looks like demand and behaves like a refund. Separate the two in your own reporting, or your best-performing "channel" will turn out to be your own lapsed list.

The decisions, in order

On 16 June 2017, Amazon announced its purchase of Whole Foods. Blue Apron priced its IPO thirteen days later at $10 a share, cut from an indicated range of $15 to $17, valuing the company near $1.9 billion. The second quarter of 2017 carried the heaviest marketing spend in its history, most of it committed before the offering.

The correction came after. Spend was cut hard in the second half of 2017, while the Linden, New Jersey fulfilment centre was still working through operational problems. Cutting acquisition without having fixed retention removes the inflow and leaves the leak. Customers fell to about 750,000 by the end of 2017, roughly 646,000 by the end of 2018, and around 350,000 by the end of 2019. Revenue went $795 million (2016), $881 million (2017), $668 million (2018), $455 million (2019). Net loss in 2017 alone was about $210 million. HelloFresh claimed the leading US position in 2018. A 1-for-15 reverse split in June 2019 kept the NYSE listing. Wonder Group bought the company in November 2023 for roughly $103 million.

The cohort evidence did not arrive late. The response did.

The second-order squeeze

Physical fulfilment makes the cut-your-way-out plan fight itself. Blue Apron had built capacity, staffing and a supply chain sized for a million customers. Run that footprint for half as many boxes and fixed cost per box rises, which pushes on the exact gross margin that funds acquisition in the first place. So the cure for a broken ratio (spend less, keep the good cohorts) degrades the denominator of the value side while it works.

Anyone running a subscription with real fixed cost should model the shrink case explicitly: at what customer count does per-unit fulfilment cost eat the margin you were protecting? A pure software business can shed acquisition spend and watch margin hold. A business with freezers cannot.

The counterfactual worth costing

Suppose the same read had been taken in mid-2016, when acquisition cost was near $94 and the six-month retention curve was already known internally. Cutting spend by a third then and putting the money into the ordering experience buys eighteen months at a lower burn, with a customer base still above a million and a private valuation intact. The moves Blue Apron did make in that direction, on-demand ordering without a subscription, retail distribution, the WW partnership in 2018, all came after the customer base had halved, when there was no cash and no patience left to fund them properly.

The cost of being late is not the difference between a good decision and a bad one. It is the difference between making the decision with a million customers and making it with 646,000.

CMO action items

• Add promotional discount back into acquisition cost this quarter, whatever the accounting treatment says, and recompute your payback on the loaded number. In a heavy first-order-offer model, expect the honest figure to be 20 to 40 percent worse.
• Report first-time and reactivated customers as two separate lines, with their own retention curves. If the blended count is the only one on the board deck, you cannot tell growth from churn recycling.
• Model the shrink case: what your unit cost and gross margin do at 60 percent and 40 percent of current volume, given the fixed capacity you have already committed to.

Common mistakes that kill results

• Reading average revenue per active customer as a health signal. It rises as the weak half of a cohort leaves. Pair it with cohort-level revenue against original cohort size, or you will misread selection as improvement.
• Cutting acquisition spend before fixing the retention problem that caused the cut. Blue Apron did both in the wrong order and lost 300,000 customers in eighteen months.
• Treating a category shock as the cause. Amazon buying Whole Foods did not create Blue Apron's cohort curves; it removed the time available to fix them. Structural weakness sets how much an external event costs you.
• Assuming the acquisition cost you can see is the marginal one. The last tranche of a $155 million budget buys the least willing buyers, and the average number politely hides them.