Google Ads for Subscription Products: Bid on Paid Subscriptions, Not Free-Trial Signups

Subscription businesses almost always configure Google Ads the same way: the conversion is a signup or trial start, the strategy gets a target CPA, and the system then does exactly what it was asked to do — buy as many cheap signups as possible. A month later signups are up, CPA is down, and the number of paying subscribers is flat or lower. On paper the campaign improved. In reality you paid for people who will never convert to paid — and that is the single most common way to set up Google Ads for subscription products wrong.

Smart Bidding is not the problem. The problem is that in a subscription business a delayed event sits between the click and the money, with trial-to-paid conversion somewhere between 10% and 40% — and if that event never reaches the ad account, the bidding model cannot see it. Google Ads for subscription products needs a different conversion architecture than an ecommerce store where payment happens five minutes after the click.

What follows is a concrete setup: which events to send and with what status, how to handle the lag between click and payment, what value to assign a subscription, how to deal with cancellations and refunds, and how to switch over without destroying bid strategy learning. It applies to SaaS, subscription services, recurring-billing info products and subscription boxes alike.

Why optimising for signups breaks the economics

Take two traffic sources with an identical $20 signup CPA.

Metric	Source A	Source B
Signup CPA	$20	$20
Trial-to-paid rate	8%	28%
Cost per paying customer	$250	$71
Average 12-month revenue	$180	$180
Outcome	$70 loss per customer	$109 profit per customer

To a strategy optimising signup CPA, the two sources are identical. Worse: source A typically supplies more volume at the same CPA, so the algorithm will scale it. That is not an algorithm failure — it is precise execution of the wrong objective.

Subscriptions have a specific trap on top of this: the no-card free trial. It maximises signups and minimises purchase intent, and because it produces a lot of events, the system learns from it fastest. The more junk signal you supply, the more confidently the model moves in the wrong direction.

The governing rule: a bid strategy optimises for exactly the event you designated as primary. If the money arrives after that event, you are not optimising revenue — you are optimising an early and unreliable proxy for it.

The three events you need to send

Event	Status in Google Ads	Why
Signup / trial start	Secondary (observation only)	See top-of-funnel volume and diagnose drops without letting bidding chase it
First successful payment	Primary, used for bidding	This is the event that represents money and belongs as the strategy’s target
Retention at 2–3 months	Value adjustment on the existing conversion	Separate subscribers who stay from those who churn after one payment

Primary versus secondary conversion statuses are a standard mechanism and they solve the “observe but do not optimise” problem directly. Setup details are in our guide to primary and secondary conversions and lead quality. The common error is keeping signups primary “for more data”: there is more data, but it is data about the wrong thing.

Click-to-payment lag is the central technical problem

An ecommerce store measures lag in minutes. A subscription with a 14-day trial sees payment two weeks later, and annual plans with long consideration cycles can take a month or more. Three practical requirements follow.

The conversion window must cover the cycle

If the window is shorter than the real lag, some payments never attach to their click and you optimise against a distorted picture. Measure the actual lag distribution in your own data: on which day do you reach 50%, 80% and 95% of payments? Set the window at the 90–95% level, not at the mean. Mechanics and limits are in our breakdown of conversion windows and conversion lag.

Recent weeks are always incomplete

With a 14-day lag, the last fortnight of data is by definition under-reported. You cannot make decisions on it: a campaign that looks like it dropped may simply not have reported yet. The working rule is to read a period closed off by at least the lag length plus a few days, and to compare cohorts by click date rather than conversion date.

Bid strategy learning runs slower

Smart Bidding needs a sufficient volume of events, and lag delays their arrival. That means changing settings less often, leaving the campaign alone through learning, and not drawing conclusions before lag plus learning period has elapsed. See our notes on the Smart Bidding learning period.

How to get the payment into the account

Payment happens in your billing system, not on a thank-you page, so a plain conversion tag will not catch it. Three workable routes:

1. Enhanced conversions. If checkout does complete on your site, send hashed first-party data alongside the conversion — it measurably improves coverage in browsers without third-party cookies. Setup is in our guide to enhanced conversions.
2. Offline conversion import via GCLID. Capture the GCLID at signup and store it next to the user ID in your database. When billing records a payment, upload the triple of GCLID, actual conversion time and value back into Google Ads. This is the primary route for trial-based subscriptions — step by step in offline conversion import.
3. Feeding events through Data Manager. If you already have a CRM or warehouse that can publish events on a schedule, build the pipeline on Data Manager and first-party data rather than manual uploads: fewer manual steps means fewer gaps.

Three things to verify on implementation: the GCLID survives redirects and mobile app handoffs; the conversion time sent is the actual payment time, not the upload time; and uploads run on a regular schedule rather than monthly, because batched data arriving in bursts keeps pushing the strategy back into relearning.

What value to assign a subscription

Sending the payment without a value is half the job. A $9 plan and a $99 plan look identical to the strategy. Three approaches, from simple to mature:

Approach	What you send	Fits	Risk
First payment	The actual first charge amount	Early stage, one or two plans	Undervalues annual plans and long-lived subscriptions
Expected value by plan	Average revenue over N months for that plan × retention probability	Most subscription businesses	Requires honest cohort data
Margin-based value	Expected revenue minus variable cost to serve	Mature products with different cost profiles per plan	Harder to compute, sensitive to assumption errors

Practical advice: start with the second approach over a 6–12 month horizon. The formula is simple — value = average revenue per plan over the horizon × share of cohort surviving to the end of it. Take the numbers from your own cohorts, not from industry benchmarks. If the product is brand new and you have no cohorts yet, start with the first payment and move up once you have two or three complete cohort cycles.

From there it is worth layering on conversion value rules to adjust value by geography, device or audience without rebuilding the data pipeline. And if you run the business on profit rather than revenue, the logical next step is POAS instead of ROAS.

Cancellations, refunds and chargebacks

In a subscription business a share of first payments comes back: cancellation inside a guarantee window, a card dispute, a failed rebill. If you never return those events to Google Ads, the model keeps treating refund-heavy segments as valuable.

The mechanism is conversion adjustments: a retraction for a full refund, a restatement for a partial one. The operating rule is simple — adjustments upload on the same schedule as conversions, inside the permitted window. Technical detail is in our write-up on conversion adjustments.

One caveat: a wave of cancellations after a deep-discount promotion is an offer problem, not a traffic problem. Before cutting campaigns, check whether you are buying people who came for the discount and leave at the first full-price charge.

Calculating your lag distribution in one evening

The conversion window and the reporting horizon should not be guessed. One simple distribution, computed from your own database, settles both.

1. Export every user with a first payment in the last 6–12 months: signup date, first payment date, plan, source.
2. Compute the difference in days between signup and payment.
3. Build a cumulative distribution: what share of payments has landed by day 7, 14, 21, 30, 45 and 60.
4. Take three numbers: the day you reach 50% (median, for understanding the typical cycle), 90% (for the conversion window) and 95% (for the reporting horizon).

The typical shape for a 14-day trial: a sharp spike on day 14–15 when the trial ends, a smaller bump around day 30 from extended trials and returners, then a long tail. If you sell both monthly and annual plans, compute the distributions separately — annual plans almost always have a longer decision cycle, and a single shared window will be a compromise that costs you data.

The same distribution answers the recurring “why does this month look terrible” question: if 90% of payments land by day 21, the last three weeks of any month will always look weak in-platform, and that is expected rather than alarming.

No-card trials: how to avoid losing the signal entirely

A free trial with no card captured maximises signups and minimises predictability. If your model genuinely requires it, there are intermediate steps that give the strategy a better signal while payment volume is still too thin for learning.

• Treat a qualified trial as its own event. Not “signed up” but “completed a key in-product action” — connected an integration, imported data, invited a teammate. Those users convert to paid at a far higher rate, and the event volume is sufficient for learning.
• Assign different values to qualification tiers. Signup at zero, activated trial at a calculated fraction of expected value, payment at full value. The strategy then gets a gradient instead of a binary flag.
• Plan the handover. As soon as payments support learning, the qualified trial moves to secondary and payment becomes primary. Schedule that step rather than leaving it to good intentions.

The key point: a qualified trial is a bridge, not a destination. Its only job is to carry the campaign to the volume at which you can optimise for money.

Reconciliation: three reports that have to agree

A pipeline without regular reconciliation degrades quietly — a GCLID lost on a new landing page, a broken upload, a changed field format. The minimum routine is a monthly comparison of three sources:

Source	What to take	Acceptable variance
Billing	Count and value of first payments by signup-date cohort	the reference
Google Ads	“Subscription payment” conversions by click date for the same period	10–15% below billing is normal — some payments are not attributable to ads at all
Analytics (GA4 or in-house)	Paying users with google / cpc as the source	Close to Google Ads, but a different attribution model — differences are expected

How to read deviations: a sudden collapse in Google Ads against billing in one month is almost always a technical break rather than a drop in traffic quality. A gap that widens gradually month over month usually means GCLID loss on some devices or browsers. A gap specific to one landing page is almost certainly a redirect stripping parameters. The general mechanics of cross-platform gaps are covered in our piece on Google Ads and GA4 conversion discrepancies.

Campaign structure for subscriptions

• Separate new from existing. If you do not exclude current subscribers from acquisition campaigns, you are paying for people you already have. Upload the customer list through Customer Match and exclude it on acquisition.
• Use customer lifecycle goals. Customer type labelling and lifecycle goals let you tell the system that a new subscriber is worth more than a returning one without building separate campaigns for it.
• Keep brand terms separate. In subscriptions, brand traffic contains a lot of existing customers looking for the login page. Blending it with acquisition guarantees inflated ROAS and wrong conclusions.
• Do not shard into micro-campaigns. With a two-week lag, a campaign generating five conversions a month will never exit learning. Consolidate until the volume is there.

A migration plan that does not reset your results

1. Week 0. Implement GCLID capture, create the “Subscription payment” conversion, verify delivery on test orders. Change nothing live.
2. Weeks 1–3. Accumulate data. Signup stays primary for now; payment runs alongside as an observed conversion. Reconcile payment counts in the ad account against billing — a gap above 10–15% means a hole in the pipeline.
3. Week 4. Flip the statuses: payment primary, signup secondary. Recalculate the target CPA: it must rise by the inverse of your trial-to-paid rate. At 25% conversion, the target roughly quadruples.
4. Weeks 4–7. Learning period. Do not touch budgets, targeting, new campaigns or targets.
5. Week 8+. Introduce conversion value and move to value-based bidding if volume supports it.

The most common failure in this migration is leaving the old target CPA in place. The system sees an unreachable target, throttles impressions hard, the advertiser concludes “the new setup does not work” and rolls back. Recalculating the target is mandatory, not optional.

Checklist: Google Ads for subscription products

1. Subscription payment exists as its own conversion action and is set to primary.
2. Signup and trial start are moved to secondary.
3. GCLID is captured at signup and stored with the user ID.
4. Conversion window covers 90–95% of your real lag.
5. Uploads run on a regular schedule, not monthly.
6. Value is sent, not just the fact of payment.
7. Adjustments are configured for cancellations and refunds.
8. Existing subscribers are excluded from acquisition campaigns.
9. Brand traffic sits in its own campaign.
10. Target CPA recalculated for the new event.
11. Reporting is built on click-date cohorts, not conversion dates.

Common mistakes

Six ways to lose the result even when Google Ads for subscription products is technically configured correctly:

• Optimising for signups “until we have data”. That is precisely how you never get the right data.
• Making micro-conversions primary. Pricing-page views and button clicks are useful diagnostic signals, but as a bidding target they produce the same failure as signups. Their proper role is covered in micro-conversions and funnel optimization.
• Judging the last two weeks. With lag, that data is always incomplete.
• Valuing on revenue when plan margins differ. A cheap plan with heavy support load can be loss-making.
• Never returning cancellations. The model keeps buying an audience that churns.
• Changing everything at once. New conversion, new target, new campaigns and new creative on the same day means no result can be attributed to anything.

If you sell subscriptions to businesses rather than consumers, layer the long-cycle specifics from our B2B PPC guide on top, and take the economic framework from unit economics and LTV in media buying.

Where PPC Rebels fits

Migrating to payment-based optimisation takes six to eight weeks, and the campaigns have to run without interruption throughout: any downtime resets learning and stretches the cycle. If you need a stable account with history and responsive support, look at Google Ads agency account rental and the rest of the PPC Rebels services. The conversion architecture itself is account-agnostic and can be built anywhere.

FAQ

Should I optimise for trials or for payments?

For payments. Keep the trial as a secondary conversion for observation. Optimising for trials grows signup volume while paying subscribers stay flat or decline.

I have too few payments for the strategy to learn. What now?

Consolidate campaigns, remove excess segmentation and consider tCPA before tROAS. If payments are genuinely scarce, an interim step is optimising for a qualified trial with a card on file — but treat that as temporary, not as the destination.

What conversion window should I set?

One that covers 90–95% of your actual lag. For a 14-day trial that usually means 30 days or more. Calculate it from your own distribution rather than accepting the default.

How do I send payments that happen in billing, not on the site?

Capture the GCLID at signup, link it to the user in your database, and upload an offline conversion with the real payment time and value. This is the standard subscription pattern.

Should I send first payment or LTV as the value?

Expected value over a 6–12 month horizon, computed from your cohorts. Multi-year LTV is too sensitive to assumptions and too hard to validate to be a good bidding input.

Do I need to send cancellations and refunds back?

Yes, through conversion adjustments. Without them the model keeps valuing segments that actually generate refunds.

How much should the target CPA rise when I switch?

In proportion to your trial-to-paid rate. At 25% conversion the target roughly quadruples. Leaving the old number in place will collapse your impression volume.

How long should I wait after flipping the statuses?

Plan for the lag length plus the learning period — typically three to four weeks before drawing conclusions. Leave the campaign alone during that time.

What do I do with existing subscribers?

Exclude them from acquisition through Customer Match and address them separately in upgrade and win-back campaigns.

Can I optimise for the second payment instead?

Technically yes, and for products with heavy month-one churn it can be justified. The cost is an even longer lag and even fewer events, so it only makes sense at sufficient volume.

How do I reconcile the ad account against billing?

Compare payment count and amount by click-date cohort over a closed period. A gap above 10–15% points to lost GCLIDs, wrong conversion timestamps or irregular uploads.

Do annual and monthly plans need separate campaigns?

Not usually. Express the value difference through conversion value rather than structural splits, or you simply dilute data volume across campaigns.