JIA JINGQIU / METHODS & TOOLS
Paid user growth.
Channels, creative, customer journeys, payment and retention for Web, App and SaaS. Define the user you need and an affordable acquisition cost before testing and allocating budget. Commerce methods are organized separately by platform.
PAID USER GROWTH / WEB · APP · SAAS
From acquisition to continued use and payment.
Channels find prospective users; the website and App help them get value. Check the promise, first experience, conversion records and payback period together before changing content, the product or the budget.
Define the acquisition goal, affordable cost and observation window before choosing channels, creative and destinations. Calculations and tests below are adaptable operating methods, not claims of achieved campaign results.
OPERATOR NOTES / METHODS & IMPLEMENTATION
Web, App and SaaS need different outcomes.
For Search, inspect intent, the destination and qualified conversions. For feeds, test whether the audience understands the problem and benefit. For apps, separate installs, first use, trials and payment. Review each path over its own payback window before changing creative, product experience or budget.
01Google Search → Web
Check intent, geography, destination and qualified outcomes. Low impression share may reflect budget, rank, eligibility or conversion—not always a need to spend more.
02Meta → Web / SaaS
Explain the problem and evidence to new prospects; offer choices to informed ones. Test distinct concepts before fine-tuning the opening.
03App install → Subscription
Check store page, deep link, first task, trial completion and charges. Cheap installs or trials do not establish profitable scale.
04Web2App and sales-assisted SaaS
A web payment still needs app activation; a lead needs opportunity and won-deal tracking. Keep the paths and their costs separate.
Implementation, inputs and local checks
Ad exports + page/app versions + event dictionary + orders/subscriptions/CRM. Align country, OS, source, windows and currency. Retain differences between Apple, platforms, MMP and billing. Code calculates realized contribution; the model drafts explanations and tests.
python3 scripts/review.py examples.json --example paidThis runs synthetic input checks, not a live account integration or an experiment proving lift.
Set budget steps and observation windows from sample size, conversion delays and affordable loss. For sales-led SaaS, follow opportunities and closed revenue rather than treating registrations or trials as returns.
01 / WEB · APP · SAAS
User growth.
Start with the audience and need. Connect ads to registration, a trial or a qualified lead, then follow first value, payment, retention and payback. SaaS lead quality and closed sales need the actual sales cycle to mature.
Web / SaaS: Google, Microsoft, Meta, LinkedIn and Reddit; TikTok, Pinterest and Snapchat can supplement the plan where the consumer audience, content and product fit.
App: Google App Campaigns, Apple Ads, Meta, TikTok, Snapchat, Reddit, AppLovin and Unity Ads. Select in-app inventory according to its fit for a game or another product.
02 / PLATFORM METHODS
Commerce, by platform.
TikTok, Amazon, Shopify, AliExpress (AE), Mercado Libre and Shopee. Explore product, content, advertising and store operations separately for each platform.
The channel comparisons, calculations and execution references remain on this page. Platform-specific methods, Skills and tools are organized in the platform directory.
This is a selection of channels by purpose, not a checklist to activate in full. Acquisition costs and payback targets depend on the project.
Channel method / Meta: creative, destinations and result reviews
META ADVERTISING / CHANNEL METHOD
From creative research to a supported recommendation.
Test different use cases and benefits before refining an opening, demonstration or offer. Connect every version to its ad and destination so the results can inform a specific change.
01 / CHANNEL & DESTINATION
Where the ad appears is one question.
Where the customer buys is another.
An ad seen inside a mobile app may lead to a website purchase. Acquiring users for an App adds a store page, installation, first use and payment. Placement, destination and optimization goal need separate decisions.
Website campaigns: test demand, products and pages together.
For commerce, prioritize valid payments and order value. For lead generation, follow through to qualified leads and sales. Views, carts and form submissions help explain the journey; they are not interchangeable measures of success.
| Channel's job | Creative and destination | What to investigate first | What automation can handle |
|---|---|---|---|
| Google Search Respond to expressed demand | Organize ads and pages around search intent. Review brand searches, category needs and competitor comparisons separately. Continue the ad's price, use case and promise on the landing page. | Are actual search terms relevant? Which page receives the click? Is budget limiting valuable sales? Separate poor intent, a poor destination and insufficient budget. | Organize search terms, flag irrelevant queries and check pages. Proposed negatives need a reason and an impact review before approval, so useful demand is not blocked. |
| Shopping / Performance Max Use product data and creative to find buyers | Check titles, images, prices, inventory and variants. For PMax, also check asset themes, product selection, URL expansion and brand controls. It is not another keyword campaign. | Look at SKUs, margins, new versus returning customers and branded demand. Strong overall ROAS does not mean every product is profitable or every sale is incremental. | Sync approved product data, flag price or stock mismatches and draft assets. Keep approval records for product exclusions, goal changes and budget changes. |
| Meta Make the case with a use case and evidence | Test different situations, problems and benefits before refining openings, demonstrations, proof and offers. Preview Feed, Reels and catalog placements separately; keep the landing page focused on the same benefit. | Follow each creative version through outbound click → page arrival → purchase → refund. Strong clicks with weak landing-page results call for a promise and destination check, not just an audience change. | Draft scripts and placement versions from approved product facts. Check captions, safe zones and links. Generated visuals still require product-accuracy and rights checks. |
| TikTok Demonstrate a use within the content | Build content around a real task and its use in practice. External creator, product and video samples inform topics; the product page must make specifications, price and delivery easy to verify. | Separate views, clicks, arrivals and sales. Trending content and estimated creator sales are not your account's returns; popularity alone cannot justify a budget increase. | Classify external samples, organize topics and scripts, and record rights and asset versions. Generation is not permission to invent customer experiences or copy competitors' work. |
| YouTube / Demand Gen Explain the product and the reason to buy | Prepare lengths, product evidence and next steps for the actual placement. A finished film alone is not a complete campaign plan for every format. | Record click-based results, post-view conversions and brand goals separately. Align measurement before comparing with Search; use a separate experiment to assess additional sales. | Prepare edit lists, captions and delivery checks by placement. Keep viewing and conversion definitions distinct rather than counting every view-attributed sale as new revenue. |
| Remarketing Address an unfinished purchase | Use different proof, reminders and products for viewers who did not buy, abandoned carts and repeat customers. Exclude people who have completed the goal or should not be contacted again. | Consider audience size, frequency, buying cycle and new-customer share together. Following people who would buy anyway can produce attractive returns without adding business. | Update segments and exclusions using permissioned data. Monitor frequency and shrinking audiences; retain a comparison group to assess actual contribution. |
How should website measurement be checked before launch?
Complete an authorized test journey and verify arrival, product views, carts, payment and value. When browser and server send the same event, use a matching deduplication identifier. Reconcile orders, refunds and reports against transaction records. Server-side reporting still has to meet consent and platform data requirements.
In Google, inspect the conversion actions actually used for bidding, not just their “primary” or “secondary” label: a custom goal can also use secondary actions for bidding. Deduct refunds in the business ledger. If platform conversion values need adjustment, use the method that platform supports; do not assume every channel automatically reverses them.
Sources: Google conversion goals · Shopping · PMax · Refunds and conversion adjustments
App campaigns: an install has to lead to lasting use.
Define what successful use means: completing a core task, exporting a result or making a first transaction. Move optimization from installs toward deeper actions only when those events are reliable and their volume and reporting delay support learning.
| Channel / scenario | From ad to use | How to judge quality | Automation priorities |
|---|---|---|---|
| Apple Ads Search intent inside the store | Separate brand, category, competitor and discovery terms. Match the keyword's use case to a custom product page. Add validated terms as exact-match keywords to the relevant ad groups, and add exact-match negatives in the discovery campaign. | Distinguish downloads, redownloads, first opens, core tasks and payment. A cheap download is not necessarily a cheap paying customer. | Organize search terms, check store-page relevance and propose terms to graduate from discovery. Verify match types and exclusions before making changes. |
| Google App campaigns Distribution across Google placements | Build assets around an install or in-app goal. Each text, image and video must make sense on its own because the platform combines them. This is not keyword-by-keyword Search management. | Compare install cost, core-task cost, payment rate and payback by operating system and acquisition date. Asset ratings are not randomized experiments. | Check asset coverage, app-event reporting and goal settings. The platform handles bidding and distribution; the team owns event quality and the chosen goal. |
| Meta / TikTok App acquisition Show what the product helps someone do | Align the ad demonstration, store screenshots and the task after first open. New users follow the store or designated acquisition path; returning users reach the relevant feature, not always the home screen. | After installation, follow task completion, trial-to-paid conversion, renewals and refunds. Compare iOS / Android, markets, app versions and users observed for the same length of time. | Draft assets for specific uses and audiences; check the store page, app version and events. Wait for sufficiently complete data before judging performance. |
| Website ads → App Not another install-buying campaign | Send people who already have the App to the relevant product or feature. Provide a working web or store fallback for others. Check login, deep links, payment and return navigation. | Reconcile equivalent Web and App conversions so the same business action is neither counted twice nor lost after the user enters the App. | Check deep links and web fallbacks regularly. In supported Google scenarios, use App Connect to pair measurement and bidding goals. |
| In-app native / interstitial / rewarded A placement, not a business goal | Use only where the product and publisher are a fit. Make closing, clicking and reward conditions clear. Observe incentivized users separately from people who chose to engage without a reward. | Inspect invalid clicks, post-install quality, retention and payment by publisher and placement. Investigate sources and anomalies when install volume suddenly rises. | Flag placement anomalies, adapt creative and group users by quality. A high click-through rate must not automatically increase spend on a suspicious source. |
Why can't Web and App share one “today's ROI” table?
The events differ. Apple Ads counts store downloads; a measurement provider generally starts with the first open. Align downloads, redownloads, first opens and attribution windows before explaining a discrepancy.
The reporting delays differ. iOS privacy-preserving attribution, platform modeling and in-app events may arrive late and may not resolve to an individual. Label missing data, zero results and incomplete reporting separately.
The payback periods differ. Group users by acquisition date and compare equal 7-day or 30-day observation periods. For subscriptions, track trials, first payments, renewals and refunds. Predicted lifetime value is not cash received.
Sources: Apple / MMP differences · Privacy-preserving attribution · App conversions · App Connect
Amazon: product operations set the limits of scale.
Check inventory, price, Featured Offer status and the detail page before diagnosing traffic. Ads may appear on websites or in apps, but when the destination is an Amazon purchase, install and subscription metrics are the wrong yardstick.
| Ad type | How to organize it | How to choose an action | Automation output |
|---|---|---|---|
| Sponsored Products | Use automatic and exploratory targeting to discover demand. Manual keywords and product targeting pursue validated directions; the relevant ASIN's page must be ready for purchase. | Connect search-term, targeting, placement and product reports. Separate no impressions, irrelevant queries and clicks without sales. Check product status before adding negatives or raising bids. | Candidate keywords, exclusions and bid changes supported by original report fields. State the observation period; do not bulk-exclude terms for a short run without orders. |
| Sponsored Brands / Video | Organize headlines, visuals and product selections around a brand or category theme, leading to the relevant Store subpage or detail page. | Separate brand-building and direct-sales goals. Review the product selection, new customers and destination instead of copying SP thresholds. | Themed creative, destination checks and report summaries. Confirm formats and eligibility in the account. |
| Display / DSP · when applicable | Use distinct messages for viewers who did not buy, replenishment customers and new shoppers in related categories. Match lookback windows to the buying cycle. | Check on- and off-Amazon placements, reach, frequency and click/view attribution. Do not rank different ad types by attributed sales without accounting for their differences. | Alerts for audience, frequency and attribution anomalies. Budget proposals include margin and inventory constraints. |
| My Listing × Ads method | Product evidence → reviews → keywords → Listing → paused ad draft → single-report analysis. | Resolve product-fact and page issues first. ACOS uses ad-attributed sales; TACOS uses total sales on a consistent basis. They are not interchangeable. | Every proposal retains its source, rationale, fields to change and reviewer. Drafts enter delivery only after approval. |
Align attribution windows, seller or vendor status and the scope of related products before comparing channel effectiveness.
02 / UNIT ECONOMICS & BUDGET
A return target needs actual arithmetic.
Platform ROAS helps manage campaigns within that platform. The business ledger uses actual revenue and costs to decide whether further spending makes sense. Both are useful; sales claimed by several platforms cannot simply be added together.
Automated bidding: what changes when you optimize volume, value or return?
Volume goals. If the aim is more installs, qualified leads or purchases, first verify that the action represents business value. Counting conversions of very different value as equal can buy many cheap results with little contribution.
Value goals. Google's Maximize conversion value pursues more value within a budget; it does not come with a profit or ROAS floor. It needs reliable, ongoing value reporting and a budget the business can afford.
Return constraints. Target ROAS seeks a balance between value and efficiency. A target disconnected from historical performance can restrict volume and does not guarantee every transaction is profitable. Match the platform target to the values actually reported, then use the cost ledger to decide whether the result is acceptable.
App goals need their own checks. Available install, in-app action and payment-value goals, and their learning requirements, vary by platform and account. Choose an event that is accurate, consistently reported and close to revenue. High event volume alone does not make page views or Credit usage a success.
ILLUSTRATIVE / USD, ONE ORDER PER NEW CUSTOMER
Acquisition cost ceiling. To retain $10 per order for fixed costs and profit, this example allows up to $30 to acquire a new customer. The corresponding ROAS on the same net-revenue basis is at least 100 ÷ 30 ≈ 3.33.
Return on advertising contribution. This page defines it as (net revenue − variable costs − ad spend) ÷ ad spend. Here, (100 − 60 − 30) ÷ 30 ≈ 33.3%. This is neither the company's net profit margin nor ROAS.
Compare like with like. First establish whether the platform reports gross or net sales. Discounts, refunds, tax, shipping and new/returning-customer definitions change the ceiling. A 3.33 target is not universal.
ILLUSTRATIVE / 100 NEW PAYING USERS, OBSERVED FOR 30 DAYS
Set the payback period first. Each paying user contributes $14 within 30 days in this example. To retain another $4, the acquisition cost ceiling is $10. That cost is spend ÷ new paying users, not spend ÷ installs.
Find where the economics break. Track install cost, first-task completion, trial-to-paid conversion and renewals together. If installs are cheap but trials do not convert, improve the promise, onboarding or payment experience rather than only adding install budget.
Keep forecasts separate. Longer-term revenue can be estimated, but state the range and assumptions, then update them with renewals and refunds. Users observed for less than 30 days cannot prove this example has paid back.
Keep three budget buckets. Validated combinations support ongoing sales; new directions have a separate test budget; reserve funds are released only when results and delivery capacity allow. Set the split by business stage and affordable loss, not a fixed formula.
Meet the conditions before adding spend. Reporting is complete, equally mature results meet the agreed cost/payback requirement, the journey has no serious faults, and inventory, fulfillment or compute capacity can support more demand. “Limited by budget” is a clue, not an instruction.
Bound each change. Record the increase, account-wide ceiling, next review time and stop conditions. Do not make major changes to goals, audiences, creative and budget together. Keep a comparison and assess what the extra spend contributes, not just the historical average return.
A daily budget is not a hard stop-loss. For example, most Google campaigns can bill up to twice their average daily budget on a given day. Track allocated budget, actual spend and billing limits separately. Periodic pause checks cannot guarantee stopping at an exact amount.
Four common situations: wait, repair, keep testing or pause?
| What you see | Check first | Next action |
|---|---|---|
| Low App returns yesterday | Have users been observed long enough? Are events, privacy-preserving reports or modeled results delayed? | Mark the data incomplete; do not automatically cut budget for low ROI yet. Keep the approved spending limits in force. |
| More clicks, no more sales | Check arrival losses, the page's promise, checkout, deep links and the core task. | Reproduce the real journey and give the product or page owner a specific issue. Keep the original creative as a comparison. |
| High CTR, weak payment and retention | Is the ad attracting the wrong people or promising more than the product delivers? | Revise the benefit, use case and demonstration, then test again. High CTR alone does not make a winning creative. |
| Broken links, wrong prices, unusual spend | Confirm affected ads, products and times. Rule out false alarms from reporting delays. | Pause the affected scope under pre-approved emergency rules and notify the owner. Resume only after repair and verification. |
03 / THE ADVERTISING LOOP
From data and creative to budget changes.
How each step gets done.
The platform handles bidding and distribution. The team owns profit targets, data, product facts and spending decisions. AI helps organize research, break down creative and draft analysis; calculations, checks and execution need explicit rules.
Turn the business goal into campaign conditions.
- BUSINESS INPUTS
- Market, product or App use case, new/returning customers, price, variable costs, stock or service capacity, and how long cash can wait.
- DECISIONS REQUIRED
- Primary conversion action, affordable acquisition cost, payback period and maximum test loss. If these are undecided, the system cannot invent a budget.
- TOOL SUPPORT
- Organize inputs, check the cost sheet, calculate scenarios with explicit formulas and flag facts awaiting confirmation.
- HANDOFF
- A one-page test plan: goal, audience, channel, destination, budget, comparison, review date and owner.
Prove the records describe the same action.
- WEBSITE
- Deduplicate the same browser and server event. Reconcile purchase values and transaction IDs with orders; record refunds separately. Follow qualified leads through to sales.
- APP
- Define first open, core task, first payment and renewal. Check the analytics SDK or measurement provider integration, distinguishing directly observed, aggregate and modeled results.
- AUTOMATED CHECKS
- Delays, duplicates, currency, timezone, sudden event-count changes and unsourced records. Retain raw data; fetching it again must not add it twice.
- WHEN CHECKS FAIL
- Separate measurement failure from campaign failure. Stop new adjustments based on bad data, keep emergency spend monitoring and assign the repair to the right owner.
Turn research into creative that can run and be tested.
- RESEARCH INPUTS
- Purchased samples, search demand, reviews and first-party customer feedback. Give every observation a source; estimated competitor performance is not a validated answer.
- PRODUCTION ORDER
- Choose the audience and use first, then the benefit, evidence, demonstration, offer and next action. Test different creative directions before refining an opening or shot within one direction.
- WHAT AI DOES
- Organize feedback, propose creative directions and draft scripts, storyboards and placement versions. People verify products, characters, features, prices and rights.
- VERSION RECORD
- Connect the creative ID to the script, visuals, captions, ad and page version. Give revisions new versions so results can inform a specific production decision.
Platform selection is not the same as testing a question.
- EVERYDAY DELIVERY
- The platform can give creatives unequal exposure. Performance rankings help shortlist candidates; they do not prove that a particular shot caused more conversions.
- CONTROLLED TEST
- Define one question, a primary metric, a budget and a complete observation period. Use supported split tests to allocate audiences and keep other conditions as consistent as possible.
- INCREMENTALITY
- To learn whether extra spend adds sales, compare outcomes with and without that investment. Platform lift studies or geographic controls require eligibility, enough data and comparable conditions.
- NO EARLY WINNERS
- Do not stop because one version leads for a day, or change goals, creative and budget together during a test. Without enough evidence, keep observing or redesign the test.
Approve the change before the system makes it.
- BEFORE THE CHANGE
- List account and ad IDs, old and new values, evidence, budget impact, expiry and reviewer. Read the current values immediately before execution to avoid overwriting someone else's recent adjustment.
- PERMITTED SCOPE
- Start with read-only alerts, then a limited execution trial. Define per-change amounts, total budget, cooldowns, protected items and emergency stop rules in advance.
- AFTER EXECUTION
- Save the receipt and read back the state. Mark each batch item as successful, failed or unexecuted. One successful request does not mean every change took effect.
- IF IT FAILS
- Stop new automated adjustments, retain the record and let the owner decide whether to restore the previous settings. Restoring settings cannot recover money already spent.
A review should change the next round of work.
- DAILY CHECKS
- Is reporting complete? Is spending unusual? Do the website and App work? Is stock available? Notify owners only about issues that need action.
- REVIEW MATURE RESULTS
- Compare acquisition cost, contribution, retention and refunds across comparable channels, markets, operating systems, creative ages and user ages. Keep “insufficient evidence” as a valid conclusion.
- CLEAR OWNERSHIP
- The media owner approves budgets. The creative owner delivers new assets. Product and engineering fix journeys and events. Data/finance reconcile receipts, refunds and costs.
- NEXT HANDOFF
- Retained combinations, reasons to stop, new test questions, asset and page versions, budgets and owners. The next round should not restart from zero after a long report.
What does automation replace, and what still needs an owner's sign-off?
| Work | Automated or assisted | Human decision |
|---|---|---|
| Data collection and reconciliation | Scheduled authorized retrieval, deduplication, calculations, delay and discrepancy checks. | Event meanings, accounting definitions, permissions and interpretation of anomalies. |
| Market and creative work | Organize purchased data and feedback; propose directions; draft copy, storyboards and placement versions. | Product facts, usage rights, brand expression and versions admitted to testing. |
| Platform bidding | The platform's automated bids and impression allocation. | Conversion goals, values, budgets and limits. Do not have another script constantly change the same bids in parallel. |
| Campaign adjustments | Produce candidates and change previews under explicit rules; execute approved changes within limits. | New campaigns, budget expansion, goal changes, major pauses and restarts. |
| Incident handling | Alerts; scoped pauses only under pre-approved emergency rules. | Diagnosis, verification, resumption and changes to the rules. |
Meta pre-launch checks. Compare the audience, placements, budget allocation and automatic creative adjustments actually enabled in the account with the plan. Check the product catalog, links, crops, captions and licensed assets; also review the actual previews after generation or automatic adjustments. Save a settings checklist before assessing delivery results. Asset scores are not experiment conclusions.
TikTok Smart+ pre-launch checks. Confirm availability in the account, then decide whether budget is allocated automatically at campaign level or set separately for each ad group. Check the combinations of audiences, placements, creative and destination URLs. After duplicating an older campaign, recheck budget, bidding, optimization goals and test settings rather than assuming they all carry over. Only campaigns that pass these checks enter the approved delivery scope.
Smart+ controls · Settings that change after duplication
Use platform reports, rules and existing tools first. Add code only when they cannot support cross-source reconciliation, a specific decision or a batch handoff. Even preview modes need a scope check: Google Scripts preview isolates AdsApp operations, but does not prevent writes through other services such as email or spreadsheets.
Google Scripts preview boundaries · Handling partial batch failures
Existing code: creative review and experiment records
const DEFAULTS: Required<ReviewOptions> = {
min_impressions: 1_000,
min_clicks: 20,
min_conversions_to_scale: 3,
min_baseline_rows: 3,
};
// Compare only eligible rows within the same account.
// Output: SCALE | ITERATE | PAUSE | INSUFFICIENT_EVIDENCEThe public module uses the median of eligible ads in the same account for directional screening. These are configurable defaults, not significance thresholds. SCALE / PAUSE flags candidates for review; it does not change budgets or prove incremental lift. The project retains the upstream Meta Ads MCP and adds a read-only offline evaluation module.
Preflight checks and platform settings
Before execution, verify the account, conversion definitions, usage rights, budget limit and available features. Propose exact changes to spend, publication or account settings; execute only after approval and read back the result.
Preflight checks
Define the demand, audience and outcome being tested. Measure brand awareness separately from purchase returns, and preserve comparable creative and observation windows.
Check aspect ratio, captions, sound, safe zones and calls to action for the actual placement. Verify products, people and prices, then review automatically cropped versions again.
Match the events and values sent to the platform to the volume, revenue or return goal. Hold scale decisions when conversions are immature, costs are missing or events are duplicated.
Carry the same promise from the ad to the website, store page and in-app destination. Test deep links, first use, trial completion and actual charges separately.
Record old and new values, exact scope and stop conditions. Read back each change; retry only failed items rather than duplicating successful actions.
Official rules / verification
TikTok: browser and server event deduplication
TikTok: split testing · Measuring additional conversions · Automation scope and controls
Google: value-based bidding · Conversion delays · Spending limits
Apple: search terms and campaign structure · Downloads versus first opens
Google: App asset combinations · App Connect
Google: execution previews · Partial successes and failures
Cost calculations, budget allocation, change approvals and daily handoffs are operating methods developed from these mechanisms. They are not universal platform standards or promises of fixed growth results.