Rebuild Attribution for Subscription Businesses in 90 Days

Rebuild Attribution for Subscription Businesses in 90 Days

Meta Ads Manager still shows every subscription brand the same top-line metric: first-order ROAS. It is the most quoted number in subscription eCommerce, and it is also the most misleading. Across the forty-plus subscription brands I have audited since 2022, founders can quote first-order ROAS to two decimal places but cannot tell me their month-three contribution margin by acquisition channel. That is a category of budget decision being made on the wrong denominator. Every quarter spent funding that mistake compounds into a subscriber base dominated by single-box churners.

The Cohort Math Most Subscription Brands Refuse to Run

Two subscription brands, both at $3M annual revenue, both reporting a first-order Meta ROAS of 2.5. Brand A ships a $45 monthly curation box. Brand B ships replenishment coffee at $32 a month. Their cost per subscriber is identical. Their Meta Ads Manager dashboards are indistinguishable. The ROAS number cannot see what happens after the first shipment lands on the doorstep.

By month six, the cohort Brand A acquired from Meta has lost two-thirds of its subscribers. Brand B's cohort has lost about a third. Run the unit economics against actual monthly recurring revenue and Brand A loses money on every Meta-acquired subscriber while Brand B clears a thirty percent contribution margin. Same first-order ROAS, two wildly different businesses, one budget decision being made on numbers that hide the difference.

The root cause is a churn differential the attribution number never sees. Subscription churn benchmarks from eAccountable put curation-box monthly churn at 10 to 15 percent, replenishment subscriptions at 7 to 10 percent, and access models at 5 to 8 percent. A curation subscriber at 12 percent monthly churn has roughly an 8-month economic life. A replenishment subscriber at 8 percent monthly churn has a 12-month life. An access subscriber at 6 percent monthly churn has a 17-month life. Treating those three customer types as identical because they share a first-order ROAS is negligent, not conservative.

Attribution for subscription businesses fails in a way direct-purchase attribution does not. A DTC brand selling a single $60 order only needs to know if that order was profitable. A subscription brand selling the same $60 first order needs to know if the customer will still be a subscriber in month four, because the economics do not clear until then. The lie subscription operators live with is that the platform number is close enough. It is not.

Shopify LTV formula guidance from Saras Analytics pegs a healthy LTV to CAC ratio at 3:1, which requires the LTV side of the equation to be real. LTV equals average revenue per user times gross margin divided by monthly churn. If you are calibrating channel mix against first-order revenue alone, you are using the numerator and ignoring the denominator. The channel that looks cheapest to acquire subscribers is often the channel those subscribers churn out of fastest.

I watch this pattern every month. A Melbourne-based pet-food brand I audited in late 2025 was spending forty percent of paid social budget on a Meta campaign that hit a 2.8 first-order ROAS. Their Recharge dashboard showed the cohort from that campaign had a month-three retention rate of 34 percent. The campaign their finance team was starving, a long-form Google Search placement at 1.9 first-order ROAS, had a month-three retention rate of 71 percent. Month-six contribution margin per subscriber on the Google cohort was almost four times the Meta cohort. The board meeting ran on the first number. The business ran on the second.

The Subscription Revenue Decomposition Model

The Subscription Revenue Decomposition Model replaces first-order ROAS with four buckets of channel-attributed revenue, scored against blended month-three contribution margin. The four buckets are Initial, Committed, Reactivated, and Referred. Every paid and organic channel gets a score in each bucket. The decision input becomes a combined metric, not the front-of-funnel view.

Initial revenue is the first-order value from each channel over a fixed acquisition window. This is what Meta, Google, and Klaviyo already report. The Model does not discard this data. It treats it as one column out of four, not as the answer.

Committed revenue is the month-two and month-three recurring revenue from the same cohort. It is the most-often-ignored column because it sits inside Recharge or Bold Subscriptions, not inside the ad platforms. Committed revenue tells you which acquisition channels produced subscribers who actually wanted what was being sold. Across my own client work I see a consistent pattern: channels with strong Committed revenue are usually the channels that got the sales pitch right at acquisition time, not the channels that drove the cheapest first click.

Reactivated revenue is post-cancel return revenue within 180 days, credited back to the original acquisition channel. A subscriber who cancelled in month two and resubscribed in month five is a different economic outcome from a subscriber who cancelled and stayed gone. The Subscription Revenue Decomposition Model credits reactivation to the original channel because the original channel produced a buyer willing to return.

Referred revenue is the referral-code activity from subscribers in the cohort. Not every channel produces referrers at the same rate. Word-of-mouth-heavy channels like podcast sponsorships and influencer partnerships tend to deliver outsized referred revenue in months three through nine. The Model captures that value explicitly instead of losing it into a generic "referral" bucket that no channel is held accountable for.

The output is a four-column matrix per channel, totalled against the blended month-three contribution margin benchmark. I have deployed The Subscription Revenue Decomposition Model across a dozen subscription brands between $1M and $15M in the past eighteen months. The pattern that shows up in every engagement is the same: channels that look strongest on Initial revenue look weakest on Committed revenue, and channels that look weakest on Initial revenue are usually where Committed and Referred revenue quietly stack up. Budget reallocation after the first pass is typically 20 to 35 percent of paid spend. The Model sits alongside The Contribution Margin Architecture and The Cohort Economics Protocol from the unit economics series, both of which assume you already know which channel produced which cohort. The Decomposition Model is the attribution layer that makes those two frameworks executable for subscription brands.

Phase 1: Instrumenting Cohort-Level Attribution (Days 1 to 30)

The first thirty days exist to prove that your cohort-level data is already present and can be joined into the attribution view. Most $1M to $10M subscription brands already have everything they need inside Recharge, Bold Subscriptions, Shopify, and Klaviyo. The work is wiring, not building.

Day 1 to 7: export three datasets into a single spreadsheet or warehouse. First, Recharge subscriber list with subscriber_id, signup_date, product_type, and status over the last 180 days. Second, Shopify orders with customer_id, order_date, and UTM source where captured. Third, Klaviyo attributed revenue per source over the same 180-day window. Join the three tables on customer_id. Recharge subscription metrics guide lists the canonical subscriber-lifecycle fields and is the right starting point if you are unsure which tables your Recharge instance exposes. Churn calculation methodology from the same source walks through the churn-window choice, which matters here because the Decomposition Model is sensitive to how you define the two-month committed boundary.

Day 8 to 14: bucket the joined dataset into the four revenue columns. For each subscriber, write out Initial (first order value), Committed (order value in months two and three), Reactivated (any order value after a cancel-and-return within 180 days), and Referred (any order value from a subscriber whose referral code generated an additional subscriber). Use the acquisition channel on the original customer record as the row dimension. For a 3,000-subscriber brand, this table fits comfortably inside Google Sheets. For larger volumes, Recharge Data Exports and a weekly load into BigQuery or Snowflake handles it cleanly. Subscription cohort retention from Saras Analytics covers the RFM-style segmentation that sits underneath this join if you need deeper cohort cuts.

Day 15 to 22: compute blended month-three contribution margin per channel. The formula is straightforward. Sum Initial, Committed, Reactivated, and Referred revenue for a cohort through month three. Subtract cost of goods sold, variable shipping costs, Australia Post insert costs, payment processing, and the original acquisition spend. Divide by the subscriber count in the cohort. The output is a single AUD number per channel. For Australian brands running GST-inclusive pricing, do the calculation on ex-GST revenue, not the gross line. Brands that get this step wrong usually end up double-counting the ATO cut as margin.

Day 23 to 30: build the weekly attribution report. The report has one row per acquisition channel and columns for each revenue bucket plus blended M3 contribution margin per subscriber. The cadence is weekly, not monthly, because the cohort-level number takes sixty to ninety days to stabilise and catching a drift early matters. Hand the weekly report to the paid media lead, the CFO, and the subscription ops lead. If your team runs on a standard Australian fortnightly pay cycle, time the review to the Tuesday after pay-day so the budget-approval conversation sits on real numbers, not platform-reported ROAS. Download our Subscription Cohort Calculator template if you want a ready-made spreadsheet shape that matches this output.

Phase 2: Reallocating Budget to Month-Three Contribution Margin (Month 2 to 6)

Phase 1 produced a four-column scoreboard. Phase 2 is where the scoreboard starts driving dollar decisions. The trap most operators fall into at this point is running two scoreboards in parallel, the platform-reported first-order ROAS dashboard and the Decomposition Model weekly report, and letting the platform one win whenever there is disagreement. Pick one. The Decomposition Model is the one.

Month 2: set a channel-by-channel target for blended month-three contribution margin per subscriber. For most physical-product subscription brands I work with, the breakeven target lands between AUD $18 and AUD $42 per subscriber by month three, depending on AOV and gross margin. Subscription LTV CAC analysis from Ecommerce CFO sets the minimum viable ratio for acquisition channels at 3:1 LTV to CAC using a twelve-month LTV horizon. Back-solve that ratio into a month-three contribution margin floor, and you have your target. Channels that clear the floor get maintained or increased spend. Channels that fail the floor get frozen for a full cohort cycle and re-measured.

Month 3: begin the budget reallocation. Take the bottom-quartile channels by month-three contribution margin and cap them at sixty percent of their prior-quarter spend. Redirect the freed budget to the top-quartile channels by the same metric. The reallocation rule is not proportional to current spend. It is proportional to proven recurring revenue. Subscription performance marketing guidance from DollarPocket covers the same principle from the media-buying angle. The shift typically moves 20 to 35 percent of paid budget within the first two cycles.

Month 4: stress-test the reallocation. Some channels that look cheap on Initial revenue collapse quickly once you price in the replacement cost of churned subscribers. Others have Committed revenue that strengthens over time because the subscribers acquired were primed for longer retention. Subscription model comparison from TechRepublic has a useful side-by-side of curation versus replenishment versus access economics, which matters here because the reallocation logic has to account for which of those three models your subscription sits inside.

Month 5 to 6: formalise the monthly subscription attribution review. It replaces the Meta and Google standalone reviews. The agenda is always the same: Decomposition Model scoreboard by channel, month-three contribution margin versus target by channel, delta versus prior month, and one reallocation decision. The review should take forty-five minutes. If it takes longer, the dataset is not clean enough and the work is back in Phase 1. Subscription retention playbook from Chargebee is a useful complement here because the reallocation decisions that come out of Phase 2 frequently surface retention problems inside specific channels that need their own operational fix.

One Australian-specific point worth building in early. If you are shipping subscription products across state borders in Australia, Australia Post insert costs vary by zone and weight band. Curation-box brands shipping across all six states and two territories often find that per-unit insert costs swing by AUD $2.80 between best and worst zones, which changes the blended margin floor for channels that happen to skew to one state. Build the freight component of the M3 contribution margin at the state level, not the national average. Brands that do not make this adjustment often over-credit channels that happen to attract buyers from Victoria and New South Wales and under-credit channels that pull customers from Queensland, Tasmania, or Western Australia.

The New North Star: Blended M3 Contribution Margin per Channel

First-order ROAS was the right metric for a simpler market. It told you whether a single order covered the cost of acquiring the customer. That worked when the customer bought once, did not churn, and your only job was to get the checkout to clear. Subscription businesses do not operate in that model. They operate in a model where the customer either stays or leaves over a sequence of monthly decisions, and the acquisition channel shapes which of those outcomes occurs.

The new north star is blended month-three contribution margin per channel, read from The Subscription Revenue Decomposition Model weekly. The reason month three, not month one and not month twelve, is that month three is where curation churn has visibly diverged from replenishment and access churn, the re-bill cycle has settled, and the reactivation tail has started to show up. Extending the window to month twelve produces a theoretically cleaner number but takes a full year to update. Subscription brands operating on single-quarter planning cycles cannot afford that lag, and brands running to EOFY deadlines in June cannot afford to discover in July that the prior twelve months of spend was misallocated.

This is also where the Decomposition Model stops being an attribution tool and starts being a strategy tool. A channel that shows strong Committed and Reactivated revenue but weak Initial revenue tells you to invest in more top-of-funnel placement on that channel, because the subscribers it produces are retaining. A channel showing strong Initial but weak Committed and Reactivated tells you the channel is over-priming single-box churners and the creative needs surgery or the budget needs to move. Treat the Decomposition Model as a category system for channel behaviour, not a finance reporting dashboard.

The shift a subscription operator achieves by running this model for two full quarters is a structural rethink of how the business treats acquisition. Month one, you are a brand tuning first-order ROAS. Month six, you are a brand that routes every acquisition dollar against the recurring revenue it actually produces. The paid media lead stops defending platform-reported numbers and starts defending decomposed ones. The CFO stops asking why Meta keeps getting more budget than performs. The board presentation pivots from ROAS slides to M3 contribution margin slides. That pivot is what separates subscription brands that compound over three years from ones that stay stuck on a treadmill of acquisition spend to replace churned subscribers.

What Operators Get Wrong

Why not just use LTV to CAC at the channel level?
LTV to CAC is the right end-state metric, but the LTV input takes nine to twelve months to stabilise for most subscription brands, and by the time you have it the budget decision is already six cycles old. Blended M3 contribution margin is the proxy that lets you act at week-four cadence with enough resolution to catch the bad channels before they compound.

My attribution window is already 28 days. Isn't that enough?
The attribution window determines what gets credited to the first touchpoint. It does not determine whether the customer stays. A 28-day window against first-order revenue still cannot see the month-three Committed bucket, which is where the Decomposition Model does most of its work. Extending the window solves the wrong problem.

What about subscribers who skip rather than cancel?
Skip-and-return behaviour sits inside the Committed bucket if the skip is a pause of one to two billing cycles, and the Reactivated bucket if it is longer. The practical rule is that any dormant subscriber who returns within 180 days counts as Reactivated and is credited back to the original acquisition channel. Longer-than-180-day returns are treated as a new acquisition and the channel scoring starts over.

Does this model work for hybrid curation-plus-replenishment brands?
Yes, and it is the cleanest way to separate the two revenue streams. Run the Decomposition Model twice, once scoped to the curation SKUs and once scoped to the replenishment SKUs. The channel scoreboards often look very different across the two, because the channel that sells a curation experience is rarely the channel that sells a replenishment routine. Trying to run a single combined scoreboard blurs the signal.

What if my team cannot build a spreadsheet this complex?
A two-person subscription operations team with Google Sheets, a Recharge CSV export, and a Shopify orders export has everything needed to run the first version. More mature brands graduate to a warehouse and a BI tool, but the spreadsheet version captures 80 percent of the value on day thirty. Do not let tool selection block the first pass. Start with the sheet and graduate when data volume forces the move.

How do I handle organic and direct traffic in the scoreboard?
Direct and organic traffic get their own rows, same as paid channels. The common mistake is to discard them as unattributed noise. Organic brand search in particular carries some of the strongest Committed and Reactivated revenue numbers in the matrix, because those subscribers arrived through intent, not interruption. Treat the organic row as a benchmark. Paid channels whose Committed revenue per subscriber falls more than thirty percent below the organic row are almost always over-spending. That gap is one of the most useful signals the Decomposition Model produces.

We run seasonal promotions. How do I stop them distorting the cohort reads?
Tag every cohort by acquisition promotion and read the scoreboard both with and without the promotional cohorts included. Subscribers acquired on a sixty-percent-off launch offer have a materially different churn curve from subscribers acquired at full price, and averaging them into a single channel row hides the skew. An EOFY June sale cohort should be readable as its own line until month three. If the promotion cohort clears the M3 margin floor after promotional discounting is costed in, the promotion worked. If it does not, the channel was subsidising a short-term Initial revenue win at the expense of recurring margin.