CAC by channel and cost per retained customer
Meta wins customers for less, but Google Search costs less per customer who stays 90 days.
The company and data are fictional.
Loading the interactive dashboard. The figures below don't depend on it.The interactive dashboard didn't load. The figures below don't depend on it.
An interactive Vista dashboard. In the line chart, the red line is Meta and the gray line is Google Search. Lower is cheaper. Try 30, 60 and 90 days: at 90 the cheapest channel to keep changes.
Which channel gets the next pound of media budget?
Sam Okoye runs growth at Quillboard, which sells project software to agencies and consultancies. He has next quarter's media budget to place.
He starts with CAC, customer acquisition cost: what you spend to win one customer. He splits it by channel, where each customer came from.
First-quarter spend and customers won.
| Channel | Spend | Customers won | CAC |
|---|---|---|---|
| Meta | £4,779 | 25 | £191 |
| Google Search | £9,034 | 30 | £301 |
On this table the budget looks settled. Give it to Meta.
Why the lowest CAC isn't the whole answer
CAC counts customers the day they start paying. It doesn't ask who is still there later. So Sam divides the same spend by the customers he kept.
Cost per customer kept, for customers who started in the first quarter.
| Kept after | Meta kept | Meta cost | Google kept | Google cost | Cheapest |
|---|---|---|---|---|---|
| 30 days | 20 of 25 | £239 | 26 of 30 | £347 | Meta |
| 60 days | 15 of 25 | £319 | 26 of 30 | £347 | Meta |
| 90 days | 9 of 25 | £531 | 24 of 30 | £376 | Google Search |
6 of Meta's customers leave in their third month, and that changes the order.
The sample is small: 25 customers from Meta and 30 from Google Search. LinkedIn and YouTube won too few to rank with confidence.
How to calculate CAC by channel
Sam builds a flow in Strata, Oplema's browser app for data prep, from his ad spend and billing exports.
Use spend from the same months as the customers.
Spend keeps arriving after the quarter ends. Mix it in and CAC nearly doubles.
Filter, Summarize
All channels, first-quarter customers.
Spend used Spend CAC January to March £23,158 £260 All spend so far £45,270 £509 Mark who stayed.
A Formula checks each customer against the days set in the dashboard.
Filter, Formula
First-quarter customers, kept after 90 days.
Channel Won Kept Meta 25 9 Google Search 30 24 Keep every channel on both sides.
A plain match would drop spend with no customers, and customers with no spend. A Union puts them back.
Summarize, Join, Union
The channels a one-sided list would lose.
Channel Spend Won TikTok £412 0 Organic - 12 Referral - 6 Cost each outcome.
Spend over customers won, then over customers kept. This feeds the dashboard above.
Formula
Meta, kept after 90 days.
To win To keep £191 £531
Start with a small test
Sam doesn't move the whole budget. He moves a small share from Meta to Google Search in some regions or weeks, and keeps the rest as a comparison. After 90 days he checks the next cohort, the customers who started in that period.
Each £1,000 moved keeps about three-quarters of a customer more. If a kept customer is worth £800 to you, that's worth about £620.
Each customer is credited to the last channel they came through, and only ad spend counts. The test will tell him more than this table can.
Sam found which channel's customers stay. Lena asks what leavers have in common, on a separate sample from the same company.
How we built it
Media spend
25 monthly spend rows from your ad platforms.
Required columns: channelmonthspend
Customers
172 customers from billing, with the channel from your CRM or analytics.
Required columns: customer_idchannelpaid_startcancelled_at
See the steps run
Select Join and open Output Right: that's TikTok, with spend and no customers. Output Left holds Organic and Referral.
Loading the interactive flow. The steps above don't depend on it.The interactive flow didn't load. The steps above don't depend on it.
Left to right: the two exports, the spend and cohort filters, totals by channel, the Join and Union, then the costs. One days setting feeds the kept check.
The formulas in this flow
| Column | Expression |
|---|---|
| Kept | CASE WHEN cancelled_at IS NULL OR CAST(cancelled_at AS DATE) >= CAST(paid_start AS DATE) + to_days(CAST(params.days AS INTEGER)) THEN 1 ELSE 0 END |
| Per customer won | round(nullif(spend, 0) / nullif(customers, 0), 2) |
| Per customer kept | round(nullif(spend, 0) / nullif(retained, 0), 2) |
What the data can't show
A customer who leaves on day 89 counts as lost at 90 days, even after months of paying. Keeping someone for 90 days isn't lifetime value either.
Customers from April to June are in the export but not the cohort, because some haven't had 90 days yet.
The data can't say why anyone left. We wrote these patterns into the sample, so test the method on your own data.
Definitions
- Blended CAC
- All spend divided by all customers, including organic and referral.
- Paid CAC
- All spend divided by customers from paid channels.
- Per customer kept
- Channel spend divided by its customers still there after the days you set.
- Cohort
- Customers who started in the same period.
- How it ties out
- 71 paid customers, 12 organic and 6 referral make 89 customers. TikTok spent £412.35 with no customers yet.
FAQ
- How do you calculate CAC by channel?
- Divide each channel's spend by the customers it won, using the same months on both sides. State whether you include salaries, creative and agency fees.
- What is the difference between blended and paid CAC?
- Blended divides all spend by all customers, including organic and referral: £260 here. Paid divides by customers from paid channels only: £326 here. Paid is the one to use for deciding where the next pound goes.
- How is this different from LTV to CAC?
- LTV to CAC compares lifetime value with cost to win. This uses a simpler observed outcome, customers kept after a set number of days, which needs fewer assumptions and says less about the long run.
- What won't it do?
- It doesn't prove a channel caused a customer, measure brand effects or tell you the cost of the next customer rather than the average one.
Build yours from exports
Rerun it each quarter. Local runs are free.