A keyword can produce cheap installs and still be a poor place to spend. It can also look unprofitable while its subscribers are still in a trial. The budget decision depends on what happened after acquisition and how much time those customers have had to pay.
Frederick James describes a method built around exact-match competitor keywords and a feedback loop between Apple Ads spending and subscription transactions. The useful idea is the loop: make acquisition decisions using downstream economics at a level where you can take action.
The available account does not establish that competitor keywords reliably outperform category keywords. Treat that as a hypothesis. The workflow below is an operating recommendation for testing it.
Define the decision before building the report
Start with one question: which keyword and country combinations deserve another bounded allocation of spend?
Choose a limited set of competitor keywords and a category-keyword comparison group. Record the intended match behavior, country, product page, bid approach, and budget for each group. Keep conditions comparable where practical, and document differences that could explain performance.
Competitor searches imply interest in another product. They do not establish interest in yours. Your page needs to explain why your app is a relevant alternative, with accurate product claims and a distinct identity.
Write down the observation window and spending limit before launch. A 30-day window is one possible starting point, but it may capture little renewal behavior for your subscription model. Choose a horizon that answers a real cash or payback question, and name what it leaves out.
Build two datasets with explicit units
The first dataset records acquisition spending. A useful proposed reporting unit is acquisition date, campaign, keyword, and country. Retain stable identifiers where available; names can change.
The second records subscription activity. It should preserve transaction identifiers, customer or subscription identifiers, event dates, revenue, refunds, and the acquisition information actually available for those customers.
Do not assume a payment export contains keyword-level attribution. Establish the supported path from acquisition attribution to customer identity to transactions. If that path only supports campaign-level reporting, make campaign-level decisions until finer attribution is reliable.
Aggregate transactions into acquisition cohorts before combining them with spend. Joining a daily spend row directly to several transaction rows can repeat the same spend for every transaction and corrupt the result.
Before interpreting performance, check:
- Spend totals reconcile with the acquisition export.
- Revenue and refunds reconcile with the transaction export.
- Repeated transaction records are removed without deleting legitimate renewals.
- Date boundaries, time zones, and currencies are consistent.
- Missing or ambiguous attribution remains visible as an unresolved bucket.
Resolve discrepancies before letting the report influence bids.
Compare customers at the same age
Calendar revenue and acquisition-cohort revenue answer different questions.
Revenue collected this month may include renewals from customers acquired months ago. Dividing that total by this month’s keyword spend does not tell you whether the new acquisition worked.
Instead, assign customers to an acquisition cohort and measure their subsequent transactions over a fixed age window. For a 30-day view, compare cohorts that have each completed 30 days of observation. Show younger cohorts separately.
Useful columns include spend, attributed installs, trials, paying customers, realized revenue, refunds, variable costs, and cohort age. Early trial conversion can help diagnose a problem, but it should not quietly substitute for mature economics.
Preserve two clocks: when the customer arrived and when the transaction occurred. The first supports acquisition decisions; the second supports financial reconciliation.
Make the economics explicit
Define contribution before acquisition spend as realized revenue minus refunds, platform fees, and relevant variable delivery costs. If the revenue field already represents net proceeds, avoid subtracting the same fees twice.
Then calculate:
Contribution-to-spend ratio = contribution before acquisition spend ÷ acquisition spend.
Contribution after acquisition spend = contribution before acquisition spend − acquisition spend.
A ratio of one means the measured contribution covered acquisition spend within that window. It does not mean the business covered salaries, development, or other fixed costs.
Use realized revenue as the baseline. If you also forecast renewals, display that forecast separately and state the assumptions. Otherwise, an optimistic retention assumption can turn a weak cohort into an apparent winner.
The right hurdle depends on cash constraints and the acceptable payback period. Set it before reading the winners table.
Turn the report into hold, cut, or scale decisions
Each decision row should include the economics, cohort age, evidence volume, attribution gaps, proposed action, and reason.
Hold when the cohort is immature or the evidence remains too thin. Keep a spending cap and a next review date. Holding should not become an indefinite license to collect more data.
Cut or reduce when mature performance falls below the predefined hurdle and a measurement error is unlikely to explain it. An exposure cap can also force a stop before the evidence becomes conclusive. Record whether the action reflects poor economics or limited tolerance for uncertainty.
Scale cautiously when mature performance clears the hurdle and is not dominated by a few transactions. Increase exposure in bounded steps, then evaluate the new cohorts separately. The historical average does not tell you what the next unit of spend will earn.
The source does not establish a universal minimum conversion count. Set minimum evidence requirements around transaction variability, the size of the budget decision, and the loss you can tolerate. If a keyword-country cell stays sparse, consider a broader grouping with a clear economic rationale.
Inspect how dependent the result is on the largest transactions. This is a sensitivity check, not permission to delete inconvenient customers from the official result.
Separate channel comparison from incrementality
A competitor-keyword group may show stronger attributed contribution than a category-keyword group. That is useful for allocation, but the groups can attract different people with different intent.
The comparison does not establish how many customers advertising caused to subscribe. Some attributed customers might have found the app anyway.
Where feasible, test incremental impact separately using a planned holdout or randomized allocation across suitable markets or time blocks. Define the allocation, observation window, and total business outcome in advance. Account for differences between markets, timing effects, and exposure spilling across groups.
Until such evidence exists, describe the result as attributed cohort economics. That wording preserves the distinction between a useful accounting view and a causal conclusion.
Keep analysis assistance downstream of validation
An AI assistant can summarize a validated table, flag unusual changes, and draft reasons for candidate actions. Calculations and reconciliation should remain reproducible outside the model.
Give the analyst enough context to distinguish an immature cohort from a deteriorating one. Require every proposed action to identify the relevant rows and assumptions. Keep budget changes under operator review while the workflow is being established.
The common failures are predictable: duplicated spend, missing attribution hidden from the report, young cohorts compared with mature ones, and a few large payments mistaken for repeatable performance. More frequent analysis will not fix those problems. Clear definitions will.
Evidence and limitations
This playbook draws on Frederick James’s public account of competitor-keyword acquisition and connecting Apple Ads spending to subscription transactions. The available account was retrieved partly through third-party metadata; the complete linked article was unavailable.
The reported revenue growth and ROAS example lack underlying spending, cohort, retention, refund, and counterfactual data. They do not establish expected returns or support universal conversion benchmarks.
The reconciliation checks, cohort rules, and decision framework here are recommendations derived from the measurement mechanism, not a validated implementation or a report of Lucas’s results. Attribution availability, current targeting options, and applicable platform requirements must be confirmed for the intended setup.
Source basis
- Frederick James’s public account of exact-match competitor-keyword acquisition and subscription-revenue attribution.
- Critical synthesis of cohort maturity, transaction reconciliation, contribution economics, and the limits of attributed performance.