Measuring growth when ROAS drops: MER, CAC, LTV and how to read them

Since January 12, 2026, Meta’s Ads Insights API no longer returns data for two attribution windows: 7-day view and 28-day view. Dashboards that pull numbers through the API show conversions falling, while the order count in the warehouse hasn’t moved. Many marketing teams read that drop as the ads getting worse, and they cut budget. Before cutting, you need a set of numbers that doesn’t depend on how each platform counts for itself, and the marketing efficiency ratio is the first of them.

When in-platform ROAS falls, check three metrics before you touch the budget: MER (total revenue divided by total marketing spend), new-customer CAC (total marketing spend divided by the number of new customers) and contribution margin after marketplace fees and marketing costs (CM3). ROAS tells you how many orders the platform managed to record. MER, CAC and CM3 tell you how much the business actually brought in. LTV, calculated on gross profit and by customer cohort, tells you how much CAC you can still afford.

Why in-platform ROAS is less reliable than it used to be

ROAS on Meta, Google or TikTok is the revenue that the platform itself credits to its own ads, using its own window and its own counting rules. Between 2021 and 2026, all three of those changed:

• In April 2021, Apple switched on App Tracking Transparency in iOS 14.5. Meta changed its default attribution window to 7-day click and 1-day view, and Ads Manager dropped the 28-day windows. For iOS users who opt out of tracking, Meta receives only one event, the highest-priority one, per visit.
• In January 2026, what was left of the 7-day and 28-day view windows was removed from the API as well, which carried over into third-party reporting tools.
• Google Ads uses conversion modeling for users who decline cookies, based on the behavior of users who consent, as long as the site has Consent Mode set up. Modeled conversions sit in the same Conversions column, so part of that number is an estimate rather than counted orders.
• Each platform credits orders according to its own window. A shopper who watches a TikTok video, clicks a Meta ad and then searches the brand on Google before buying gets claimed by all three. Add up the revenue in the three reports and you usually get more than the revenue you actually made.

Undercounting and overcounting are only half the problem. A 2019 study by Gordon, Zettelmeyer and colleagues in the journal Marketing Science compared 15 randomized experiments on Facebook, covering about 1.6 billion impressions, with measurement methods based on observational data. The observational methods often failed to recover the true lift the ads created, and in some cases they were off by several times. Reported ROAS measures how closely ads and orders are associated. It cannot tell you how many extra orders the ads produced.

Marketing efficiency ratio, CAC and LTV: definitions and formulas

Shopify advises keeping one fixed definition of revenue for MER across periods (before or after returns, with or without VAT). Otherwise, comparing this month with last month stops meaning anything. Be explicit about the denominator too: some tools count only ad spend, while Adtek counts all marketing spend, including KOC (key opinion consumer) fees, agency fees and gifts.

The often-quoted 3:1 LTV:CAC ratio comes from David Skok, who wrote it for SaaS businesses with subscription revenue, with the added condition that CAC pays back in under 12 months. Online retail brands don’t get paid by the same customer every month, so LTV should be calculated on contribution margin over a fixed window. Twelve months is enough to make an annual budget decision.

Where marketplace fees eat into ROAS

A ROAS of 5 on Shopee Ads does not mean you are making money. Since May 2026, the transaction processing fee on both Shopee and TikTok Shop has been around 6%, plus a fixed fee by product category and a per-order infrastructure fee. Our TikTok Shop vs Shopee comparison puts the total cost of listing and selling on the marketplaces at about 23 to 24% of revenue, before advertising. That is why Adtek reads CM2 and CM3 next to MER and never reads MER on its own.

Here is a hypothetical example: a skincare brand selling on its website, Shopee and TikTok Shop, with one month of figures:

MER for the month is 2,000 / 500 = 4.0. Now add up the revenue the four ad platforms report for themselves: Meta 900, Google 400, TikTok 240, marketplace ads 450, for a total of VND 1,990 million, almost all of the month’s revenue. Taken at face value, the platform reports say almost no orders came from repeat customers, organic search or referrals. That figure is almost certainly double counted.

The month had 2,500 first-time buyers out of 5,000 orders. New-customer CAC is VND 500 million / 2,500 = VND 200,000. Each order brings in about VND 212,000 of CM2, so a new customer pays back their CAC on the first order. If 12 months of data from earlier cohorts shows a new customer places 1.8 orders on average, LTV on a contribution margin basis is 1.8 × 212,000 = VND 381,600, and the LTV:CAC ratio is about 1.9. By the 3:1 rule that looks “poor”, yet the brand makes money from the very first order. This is why SaaS rules shouldn’t be applied to retail.

Reading the metrics together: a diagnostic table

Any single metric can mislead you. Put them side by side and the cause usually shows up:

The first row is the case teams most often get wrong. The example at the top of this article, Meta removing view-through windows from the API, lands exactly on that row. Cutting budget there is a reaction to a change in counting, not to a change in business results.

Which metrics to check daily, and which quarterly

This is the measurement cadence Adtek uses with e-commerce and D2C clients. Each layer serves a different kind of decision:

Gartner reports that marketing budgets in 2025 stood at 7.7% of company revenue, and 59% of CMOs say their budget is not enough. When the budget isn’t growing, decisions about splitting money across channels have to rest on the weekly and monthly layers, not on each platform’s ROAS.

Checking the numbers with incrementality tests and MMM

MER tells you whether the business as a whole is profitable. It does not tell you which channel produced that profit. There are two ways to answer that question without trusting the platforms’ own reports.

The first is an incrementality test: split users or regions into a group that sees ads and a group that doesn’t, then compare revenue. Meta has Conversion Lift, and Google has a geography-based Conversion Lift in beta. A mid-sized business can run a simple version itself: switch off one channel in a group of comparable provinces for 4 weeks and compare revenue with the remaining group, after removing seasonal trends. The test measures how much revenue you would lose by turning the channel off, which ROAS cannot measure.

The second is Marketing Mix Modeling, a regression model built on weekly spend and revenue data. Google opened Meridian to all businesses in January 2025, and Meta has Robyn; both are open source. MMM needs at least 1 to 2 years of clean weekly data, so for many Vietnamese brands the first step is to standardize data along data-driven marketing lines and bring customer data into one place, for example with a CDP. BCG recommends combining MMM, multi-touch attribution and experiments instead of relying on a single method.

The data for both approaches is bound by user consent. Pixels, Conversions API and customer lists uploaded to ad platforms can only be used for people who have agreed, under the rules covered in our article on Vietnam’s personal data protection law for marketing. Signal will keep shrinking, so MER and testing will become even more important than ROAS.

LTV depends on repeat buyers

The claim that “a 5% increase in retention lifts profit by 25 to 95%” comes from Frederick Reichheld’s research (Bain) on service businesses in the 1990s, and Harvard Business Review cited it again in 2014. It shows the direction, not a benchmark for an online retail brand in 2026. A more practical measure is the 90-day and 12-month repeat purchase rate for each cohort, grouped by the month of first purchase.

In the example above, if the average number of orders per new customer over 12 months rises from 1.8 to 2.2, LTV goes from VND 381,600 to VND 466,400. The brand can then pay about 20% more in CAC and keep the same profit margin, which lets it outbid competitors who only look at ROAS in ad auctions.

The market is growing fast. According to Metric, Vietnam’s four main marketplaces, Shopee, TikTok Shop, Lazada and Tiki, reached about VND 429.7 trillion in 2025, up 34.75%, with TikTok Shop holding roughly 41%. As marketplace and advertising costs rise together, the brands that keep growing are the ones that know what they pay for a new customer and what that customer brings in over a year.

Adtek builds MER, CAC, CM3 and cohort LTV dashboards for brands selling on their website, Shopee and TikTok Shop, combining ad, marketplace and CRM data. If you want to review how you measure today before locking next quarter’s budget, contact the Adtek team.

References

• Meta for Developers (October 2025). Ads Insights API metric availability updates.
• Meta for Developers (January 2021). Actions for partners and mobile and web advertisers.
• Google Ads Help (accessed October 2026). About consent mode conversion modeling.
• Gordon, Zettelmeyer, Bhargava, Chapsky (2019). A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook. Marketing Science.
• Shopify (accessed October 2026). Marketing Efficiency Ratio.
• David Skok, For Entrepreneurs (accessed October 2026). SaaS Metrics 2.0.
• Gartner (May 2025). 2025 CMO Spend Survey.
• Google (January 2025). Meridian marketing mix model is now open to everyone.
• Meta for Developers (December 2024). Announcing the Python version of Project Robyn.
• BCG (2019). Marketing Measurement Done Right.
• Harvard Business Review (October 2014). The Value of Keeping the Right Customers.
• Thanh Niên, citing Metric data (February 2026). Four e-commerce marketplaces took in nearly VND 430,000 billion in 2025 (in Vietnamese).