This lesson follows a Shopify store selling a 20oz leak-proof commuter tumbler. Products, orders, and spend are classroom cases; they do not establish that you have launched ads, passed tracking QA, or been authorised to change an account. The useful output is a decision log someone can review next week: the evidence, one variable, what stays fixed, the review date, and the condition for pausing or restoring.
12 parts
Show all 12 parts
- 01Understand the window, variable, rollback, and log
- 02Write six evidence jobs into the log
- 03Review in order before changing the account
- 04Separate conversion columns from the order calendar
- 05Read each field from its real evidence source
- 06Find the no-conversion layer
- 07Choose an allowed first-cycle move
- 08Separate urgent repair from an ordinary experiment
- 09Write the variable, window, and rollback line
- 10Check competing explanations in three mixed-edit cases
- 11Read the full seven-day tumbler drill
- 12Use the meeting to agree on a review
Start with one separate case: a 7-day readout has 80 clicks and 6 carts, yet Google Ads records 3 Purchases while Shopify has 1 order; 2 clicks came from “free tumbler” and “used insulated bottle.” “CVR got worse” is not a diagnosis. Duplicate measurement, query intent, page fit, and a thin sample remain possible explanations. Turn the symptom into a falsifiable question before changing the account. The later $29/$180 drill uses different data.
A falsifiable question allows the evidence to show that your first explanation was wrong. For example, ask whether the same order received duplicate credit within the same conversion action, then trace identifiers, trigger records and credited results. Different platform totals alone do not establish duplication. That check helps separate a tracking repair from a date or attribution difference.
Series start
From a metric readout to a reviewable action
You can start here without the metrics lesson. Align impressions, clicks, CTR (click-through rate), CPC (cost per click), CVR (conversion rate), CPA (cost per conversion), and ROAS (attributed revenue/ad spend) in one window, then add GA4 behavior, Shopify orders, refunds, discounts, and costs. Day 7 is a review date. It is neither automatic permission to edit nor proof that the sample supports a conclusion.
- How long should I wait before optimizing Google Ads after launch?
- Should I review search terms, budget, or ROAS first in week one?
- When should I add negative keywords?
- How can I tell what worked after changing several ad settings in one day?
- Can I change budget using current ROAS before conversion lag ends?
- When should I roll back instead of continuing to optimize?
- What evidence should remain in a seven-day review record?
Understand the window, variable, rollback, and log
These four terms determine when to wait, pause, change, or restore. Use them to make a review executable.
- observation window
Observation window is the time and sample range you wait after launch or a change before judging performance. Where you see it It lives in review sheets, change logs, and team calendars rather than a single Google Ads button. If misread Without it, teams read learning noise, daily luck, or low sample as trend. Example: If a new Search campaign has only 18 clicks after 3 days, mark low sample instead of rebuilding.
- primary variable
Primary variable is the one main factor allowed to change in this cycle, such as negatives, budget, copy, or hero section. Where you see it It belongs in the one allowed change field of the first optimization decision log. If misread Changing too many variables makes next week’s review unable to identify the cause. Example: If this cycle only adds negatives, do not also change budget, copy, and page.
- rollback condition
Rollback condition is the standard for pausing, restoring the previous version, or reducing risk if the change fails. Where you see it It should be written with the change record: how long to watch, which metric decides, and who acts. If misread Without it, teams explain bad results after the fact instead of acting by rule. Example: If a new negative reduces high-intent impressions sharply, revert it and review match type.
- change log
Change log records who changed what, when, based on which proof, and when it will be reviewed. Where you see it It can be a spreadsheet, Notion page, or task, but next week’s review must see it. If misread Without it, neither wins nor losses become reusable account knowledge. Example: Record: May 20 add free/jobs negatives; review search terms and CPA on May 27.
Write six evidence jobs into the log
Write the evidence before the action. These six rows are questions for the log, not verified account states; name the source, date, and missing proof for each.
What the log must answer
Tracking status
- Write
- Whether Purchase, value, currency, and transaction ID are trusted.
- Why
- Untrusted tracking means CPA / ROAS should not guide optimization.
Sample status
- Write
- Spend, clicks, conversions, and days observed.
- Why
- Low sample turns many conclusions into noise.
Search term evidence
- Write
- Which terms are high-intent, suspicious, or irrelevant.
- Why
- The first cycle starts with what the system actually bought.
Page path evidence
- Write
- Browse, cart, checkout, payment failure, quick bounce.
- Why
- If terms are right but the page fails, budget alone will not fix it.
Primary variable
- Write
- Pick only one: negatives, budget, copy, page, or bidding.
- Why
- Next week must be able to read whether this action worked.
Frozen items and rollback
- Write
- What stays unchanged and what triggers rollback or pause.
- Why
- Prevents optimization from becoming unreviewable churn.
Review in order before changing the account
Work through the evidence in this order and leave account changes until last. Observation does not mean ignoring a demonstrated fault: duplicate Purchases, approval or eligibility failures, price/availability conflicts, or failed payment and checkout promises need an authorised owner to contain or repair the affected scope. Preserve the readout first and start a clean window after repair.
From measurement to action
Verify tracking
Check whether the conversion action, value, currency, test order, and backend order reconcile.
Caution
Do not move budget first.Judge sample
Check whether spend, clicks, conversions, and days support a conclusion.
Caution
Do not read two no-conversion days as structural failure.Read search terms
Identify whether spend went to high-intent, suspicious, or clearly irrelevant queries.
Caution
Do not only read the keyword list.Read page path
Read what happened after the click: browsing, carts, checkout, or a quick bounce.
Caution
Do not blame page problems only on ads.Pick one action
Choose one change: budget, keywords, negatives, copy, page, or bidding.
Caution
Do not change them all on one day.
Four states and their next steps
Observe
- When
- Low sample, tracking just fixed, no obvious direction drift.
- Next
- Write the observation window and next review date.
Pause
- When
- Clearly irrelevant terms, wrong location/language, or page promise mismatch.
- Next
- Pause the waste source and record why.
Modify
- When
- The problem is clear and can be handled by one variable.
- Next
- Change one main variable and freeze the rest.
Rollback
- When
- After change, traffic worsens, cost breaches, or conversion quality falls.
- Next
- Restore the previous version and write the trigger.
Does changing one variable at a time block an urgent fix?

As text: the whole walk-through
- Tracking reconciled Verify tracking first: the test order, value and currency reconcile with the backend order, then go on.
- Window running Change one variable, here one set of negative keywords for clearly irrelevant terms, freeze the rest and open the window.
- Direction only The budget goes up before the window ends: you cannot tell which change did what, so this reading is only a direction.
- Window running Drop the extra change and restart the window today; budget waits for the next round.
- Urgent fault A duplicate Purchase turns up in the window: it is a fault, not an experiment, so keep the readout and Change history first.
- Window running Once someone with access has fixed it, the window restarts from the fix day and the earlier readings are only a direction.
- Rolled back Impressions on high-intent terms drop past the rollback line written in advance: roll back that one change as agreed.
- Read the numbers After the rollback go back to the numbers: write down the trigger, then see whether the negative set can be narrower.
- Window running Change one variable, here one set of negative keywords for clearly irrelevant terms, freeze the rest and open the window.
- Kept The window runs out clean and never crosses the line: keep the change, log it, and start the next round from the numbers.
Illustrative: the lesson's own examples; no durations are drawn, set the window by your own review plan.
Separate conversion columns from the order calendar
An observation window is your comparison plan. A conversion window is an action setting that determines how long after an ad interaction a conversion can be counted. Conversion lag is the delay from interaction to conversion and usable reporting. Do not change the conversion window merely to improve this week’s result. Align time zone, currency, conversion actions, attribution scope, and refund definitions before comparing normal and conversion-time columns.
What each readout answers
Conversions
- Use
- Check whether attributed learning signals have started to accumulate; useful for tracking status and optimization input.
- Do not infer
- Do not read it as this week’s Shopify order calendar; it may be written back by click time.
Conversions by conv. time
- Use
- Read when conversions actually happened to explain why Ads conversions and the order calendar differ in week one.
- Do not infer
- It is short-window readout evidence, not a replacement bidding basis.
Conversion value
- Use
- Use aggregate value to identify reconciliation needs, then check amount, currency, and transaction_id in actual requests and order parameters; the total alone does not validate the payload.
- Do not infer
- Do not treat platform value as profit; reconcile it with Shopify refund, discount, and net sales.
Conversion value by conv. time
- Use
- Read value in the actual current conversion window to explain short-term ROAS versus Shopify sales gaps.
- Do not infer
- Do not use it alone to release budget; still write break-even ROAS, Max CPA, and contribution profit.
Cost / conv.
- Use
- After tracking is trusted, compare CPA to Max CPA; if tracking is not trusted, fix conversion action first.
- Do not infer
- Low CPA does not prove order quality, post-refund profit, or scale readiness.
Search terms / Change history
- Use
- Keep waste terms, negative level, change time, and next review in one record.
- Do not infer
- Do not turn this into full negative governance; this lesson only decides whether terms are the one first-cycle variable.
Account information is evidence, not an automatic conclusion
Search terms report
- Platform information
- The search terms report shows actual queries, keyword, and match type; low-volume or privacy-limited queries may not appear as individual rows.
- Use in this review
- In the first cycle, use it to confirm visible waste terms, promotable terms, and page-content needs, not as the complete list of long-tail demand.
- Does not establish
- Do not assume missing queries mean no waste, and do not rebuild a campaign from match type alone.
Conversion tracking data
- Platform information
- Conversion, conversion value, conversion rate, and related columns are useful only when actions, value, currency, dedupe, and attribution logic are trusted.
- Use in this review
- The first cycle starts with tracking status: when tracking is not trusted, the allowed move is QA or reconciliation, not budget, bid, or copy changes.
- Does not establish
- Do not read zero conversions as no demand, or a few conversions as proof that the structure is correct.
Change history
- Platform information
- Change history can show account edits, timing, user, and change type, but it records the edit, not the business reason.
- Use in this review
- The decision log adds what change history does not: why you changed, the primary variable, frozen items, and rollback timing.
- Does not establish
- Do not treat change history as the review itself; seeing a budget or negative edit does not mean the edit had evidence.
Budget changes
- Platform information
- Budget changes affect serving pace and spend limits; Google Ads may spend more than the average daily budget on a day while monthly limits still apply. Read charge limits against the current budget type and edit rules.
- Use in this review
- In the first cycle, test budget in small steps only after tracking, search terms, and page path pass, with a written check window.
- Does not establish
- Do not cut budget just because one day spends fast, and do not raise budget repeatedly because one day ROAS looks good.
Negative keywords
- Platform information
- Negative keywords can keep ads away from irrelevant queries, but overly broad scope or the wrong level can block real demand.
- Use in this review
- In the first cycle, handle only clearly irrelevant terms and record negative level, match type, reason, and check window.
- Does not establish
- Do not treat every non-converting search term as waste; a short window can simply mean low sample or conversion delay.
Conversion windows
- Platform information
- Conversion windows decide how long after an ad interaction conversions can be counted, so short-window reviews are affected by conversion lag.
- Use in this review
- First-cycle review should write the observation window and conversion lag before deciding whether to extend, fix tracking, or change one variable.
- Does not establish
- Do not rebuild the campaign just because ordinary conversion columns look weak on day 3 or day 7.
Read each field from its real evidence source
Pull each field from its proper system rather than infer correctness from one ROAS total. Treat the paths below as navigation clues, subject to the current interface, permission, and visible fields. Inspect transaction_id in actual requests, event parameters, and implementation evidence: Google Ads does not report it as a column. Change history records account edits; keep business reasons, page versions, and review plans in your own log instead of claiming the platform recorded those judgments.
Four admin evidence routes
Tracking and order reconciliation
- Path
- Google Ads > Goals > Conversions > Summary; GA4 purchase; Shopify Orders.
- Fields
- conversion action, primary / secondary, status; transaction_id, value, currency; order id, net sales, discount, refund
- Supports
- Whether CPA / ROAS can be used to judge first-cycle action.
- Does not prove
- It does not prove budget, structure, or bidding should change now.
- Allowed action
- If it does not reconcile, only fix tracking and reconciliation; pause optimization conclusions.
- Freeze
- Freeze budget, bidding, keywords, page, and copy.
Search terms and waste control
- Path
- Google Ads > Insights and reports > Search terms; Keywords; Negative keywords; Change history.
- Fields
- search term, matched keyword, match type; clicks, cost, conversions, Conv. value / cost; negative level, reason, change time
- Supports
- Whether spend went to clear waste terms or promotable high-intent queries.
- Does not prove
- It does not prove every non-converting term should be negated.
- Allowed action
- Add only obvious negatives, with negative level and reason written.
- Freeze
- Freeze budget, page, bidding, and ad copy.
Page path and checkout friction
- Path
- GA4 Landing page / events; Shopify checkout / orders; CRO page evidence.
- Fields
- landing page, source / medium, Final URL; view_item, add_to_cart, begin_checkout, purchase; checkout error, mobile device, page version
- Supports
- Whether traffic promise, page fit, or checkout path has clear friction.
- Does not prove
- It does not prove campaign structure has failed.
- Allowed action
- Fix one page or checkout variable, such as shipping promise, price explanation, or payment error.
- Freeze
- Freeze search terms, budget, bidding, and account structure.
Budget and profit guardrail
- Path
- Google Ads Budget / Change history; ROAS calculator; Pricing calculator; Shopify order economics.
- Fields
- current average daily budget, previous budget, change time, spend pace; CPA, ROAS, break-even ROAS, Max CPA; contribution profit, refund reserve, inventory cover, rollback trigger
- Supports
- Whether this small budget test has profit room and a stop line.
- Does not prove
- It does not prove scale readiness or stable cost at higher budget.
- Allowed action
- After tracking, terms, and page pass, make only a small budget change and write the rollback trigger.
- Freeze
- Freeze bid strategy, page, keywords, and copy.
In the page-path row, view_item means a product view, add_to_cart a cart addition, begin_checkout a checkout start, and purchase a purchase event. These locate recorded actions; they do not automatically describe the same people moving through every step. Align the page, source and observation scope, then check orders and error records for the break rather than subtracting four event totals.
ROAS Calculator
Bring price, unit costs, ad spend, target ROAS, and the fee/discount assumptions actually used. Record modelled break-even ROAS and affordable CPA. The tool models revenue/orders/profit from inputs and a target; it does not fetch an account, and its scenario ROAS is not actual ad ROAS.
Product Pricing Calculator
Bring same-currency costs, price, fees, shipping, ad costs, and return assumptions; record pricing and contribution guardrails. Check refund reserve, inventory cover, and actual net profit separately against order/cost records. A calculated result does not authorise scale.
Here, break-even ROAS is the modelled revenue/ad-spend ratio needed to break even. Max CPA is the affordable cost line taken from actual unit economics, not the target CPA (tCPA) set in the account. Observed marginal CPA compares additional comparable spend with additional accepted conversions; do not invent a number when the latter is zero or uncertain, or treat a before/after difference as proof of budget-driven lift.
Find the no-conversion layer
Separate a no-conversion week by layer. Selecting a question opens both conditional references; it does not answer the question for you. Write the evidence, missing proof, and observe/pause/modify/rollback state in your log. No universal multiple of target CPA establishes sample sufficiency on its own; clicks, buying cycle, and conversion lag matter too.
Choose the no-conversion checkpoint
- If yes
- Move into search-term and page-path checks.
- If no
- Check approval, budget, location, bidding, keyword coverage, and conversion settings.
- If yes
- Add negatives or tighten structure before rebuilding.
- If no
- Continue to sample size and page path.
- If yes
- Move into primary-variable decision.
- If no
- Mark sample as low and keep the window.
- If yes
- Inspect checkout, payment, price, shipping, and trust.
- If no
- Inspect page promise, speed, mobile, and search-term fit.
- If yes
- First explain attribution source, conversion lag, test marking, and reporting definitions; then inspect transaction ID, value, currency, and event firing, repairing any actual defect.
- If no
- Treat as real no-conversion and choose one main variable.
Choose an allowed first-cycle move
Choose the case closest to the evidence and read the state, wrong move, permitted move, frozen items, and rollback condition. These cards are references; they neither inspect nor change an account.
Choose the closest first-cycle case
- Evidence
- Ads reports conversions, but Shopify orders, value, currency, or transaction ID do not reconcile.
- Wrong move
- Using CPA / ROAS to change budget or bidding strategy.
- State
- Pause optimization conclusion
- Allowed move
- This cycle only fixes measurement and adds test-order proof plus reconciliation notes.
- Frozen items
- Freeze budget, bidding, keywords, and page.
- Rollback condition
- If the fix changes historical readout, restart the observation window and do not mix old data.
- Log sentence
- Cycle state: tracking fix; primary variable: conversion measurement; review: read CPA only after test order and Shopify / Ads reconciliation pass.
- Evidence
- After 3 days there are few clicks, and spend has not reached a readable share of affordable CPA.
- Wrong move
- Rebuilding the campaign or changing budget heavily because there are no conversions.
- State
- Observe
- Allowed move
- Write the observation window and only check approval, impressions, clicks, and obvious query mismatch.
- Frozen items
- Freeze page, bid strategy, budget, and keyword structure.
- Rollback condition
- If the window ends with no useful clicks or obvious query drift, move into primary-variable decision.
- Log sentence
- Cycle state: observe; primary variable: none; review after the planned click/spend window.
- Evidence
- The search terms report shows free, jobs, repair, wrong-location, or off-product queries.
- Wrong move
- Negating every non-converting term at once or rebuilding the structure.
- State
- Modify
- Allowed move
- This cycle only handles clearly irrelevant terms and records the negative level plus reason.
- Frozen items
- Freeze budget, copy, page, and bidding.
- Rollback condition
- If high-intent impressions are blocked, revert the negative and use a narrower scope.
- Log sentence
- Cycle state: modify; primary variable: negative keywords; review query quality and retained high-intent impressions.
- Evidence
- Search intent is right, but browsing is shallow, carts are low, or checkout/payment path is weak.
- Wrong move
- Adding terms or budget and treating page friction as a traffic shortage.
- State
- Modify
- Allowed move
- This cycle changes only one of first-screen proof, price explanation, shipping/returns, or mobile friction.
- Frozen items
- Freeze keywords, budget, and bidding.
- Rollback condition
- If engagement and carts do not improve after the page change, restore the previous version and test the next proof item.
- Log sentence
- Cycle state: modify; primary variable: landing page proof; review engagement, carts, and checkout path.
- Evidence
- A few orders or better ROAS appear, but sample is small, and budget changes affect serving frequency and spending limits.
- Wrong move
- Changing budget up and down several times on the same day.
- State
- Small change or continued observation
- Allowed move
- Only when tracking, terms, and page path pass, adjust budget lightly and write the check window.
- Frozen items
- Freeze bid strategy, page, keywords, and copy.
- Rollback condition
- If spend pace worsens or CPA crosses the stop line, restore the previous budget.
- Log sentence
- Cycle state: small budget test; primary variable: average daily budget; review spend pace, CPA, and conversion quality.
Separate urgent repair from an ordinary experiment
One variable is a rule for ordinary experiments that can wait and compare, not a reason to leave a fault running. Choose the campaign actually carrying traffic and read its evidence, urgent or eligibility route, normal experiment, and limits. Do not apply Search ad-group/negative advice directly to Shopping or PMax. A practice note does not verify a real item, Market, inventory, order, or publishing state.
Choose the campaign carrying the traffic
- Evidence
- Write Search terms, matched keyword, match type, location/language, ad approval, and Final URL. When low-volume or privacy-limited queries are not shown row by row, record “not visible” as a gap rather than proof of no demand or no waste.
- Urgent or eligibility route
- When approval, location/language, duplicate Purchases, or a checkout promise is clearly wrong, this is not a seven-day experiment. Preserve the current readout and Change history, have the authorized account lead contain or repair that fault, then start a fresh observation window from the repair date.
- Normal-experiment route
- When tracking and page path are readable and a clearly irrelevant query is spending, this cycle may add one scoped negative set; budget, RSA, page, and bidding remain frozen.
- What to record
- Write campaign_surface=Search, query/keyword/match, approval status, one action, old/new value, observation window, rollback trigger, and next readout.
- What it cannot prove
- This is not full negative-keyword governance and does not prove an unseen query is absent, a keyword has no value, or the account is ready to scale.
- Evidence
- Write Merchant Center item status, product ID, target Market, data-source price and availability, landing-page visible price and availability, and the item URL. Do not read campaign spend alone; the same item can have different issues in different Markets.
- Urgent or eligibility route
- When price or availability conflicts across product data, landing page, structured data, or checkout, or an item is disapproved, record an offer-fulfillment or eligibility problem first. Do not hide it with budget, bidding, or query tests; record the affected item and Market for an authorized account lead to fix and read back.
- Normal-experiment route
- Only after item status, price, availability, and page have been read back as consistent and tracking is trusted may one budget or page change become a normal experiment; deeper feed diagnosis belongs in the Merchant Center/Feed lesson.
- What to record
- Write campaign_surface=Shopping, item_id, Market, item status, price/availability read-back time, page URL, one variable, responsible lead, and recovery condition.
- What it cannot prove
- The practice record does not prove every item, Market, tax, shipping, review, or Merchant Center state has passed, and it is not repair or publishing permission.
- Evidence
- Write the asset group, listing-group product scope, connected Merchant Center data, conversion goal, and current Final URL/URL-expansion scope. If traffic lands on an unexpected URL, record the actual URL and date first; do not call it a bidding failure by default.
- Urgent or eligibility route
- When product set, price/availability, URL scope, or conversion measurement conflicts with the real offer, pause this cycle’s optimization conclusion. Preserve the current scope and Change history, have an authorized account lead verify product, URL, and measurement, then establish a new observation window.
- Normal-experiment route
- Only after product scope, URL scope, and measurement are read back and the condition maps to one clear variable may you test that variable; do not change budget, bid goal, product scope, and URL scope together.
- What to record
- Write campaign_surface=Performance Max, asset group, listing group, product/URL scope, conversion goal, one variable, review window, and rollback/escalation condition.
- What it cannot prove
- This does not recommend enabling or disabling automation, Final URL expansion, or bidding features; read current settings, reporting visibility, and authority back in the actual account.
Write the variable, window, and rollback line
Limit the candidate change to one factor. Negatives need demonstrably irrelevant intent; budget needs trusted measurement, healthy queries, and a usable page path; copy needs the right intent with an unclear promise; page changes address post-click friction. Free, jobs, repair, and wrong-fit examples are not a universal blacklist: confirm that they are irrelevant to this product and scope.
Action prerequisites and frozen items
Negative keywords
- Prerequisite
- Clearly irrelevant, free, jobs, repair, wrong location.
- Freeze
- Freeze budget, page, and bid strategy.
Budget
- Prerequisite
- Trusted tracking, healthy terms, page path has valid signals.
- Freeze
- Do not also change keywords, copy, page, or bidding.
Ad copy
- Prerequisite
- Low CTR with correct intent and unclear ad promise.
- Freeze
- Do not also change page and keywords in the same cycle.
Page
- Prerequisite
- Terms are right but CVR, carts, or checkout path is weak.
- Freeze
- Keep ad structure and budget unchanged.
The following are templates for a separate log. The 3–7-day, 7-day, and 3–4-day plans are classroom examples, not Google requirements or automatic edit deadlines. Adapt them to comparable clicks/spend, actual conversion lag, seasonality, and intervening changes. Lowering budget does not erase spend or instantly reset that day’s limit; read the rules for the current budget type.
Choose a change-control reference
- Current pressure
- Search terms show free, jobs, repair, wrong-fit, or clearly irrelevant queries, and spend is concentrating there.
- Choose when
- Choose this when tracking is trusted, page path has no obvious break, and the waste term does not overlap real product demand.
- Only primary variable
- Add one negative keyword set only, scoped to ad group or campaign first; do not touch budget or page.
- Separate change-log template
- Change history: 2026-__-__, added negative ____, level ____, reason: Search terms spent ____ with no purchase intent.
- Example observation plan
- Classroom plan: Watch 3-7 days or one comparable spend window; check whether high-intent impressions, clicks, and CPA are preserved.
- Rollback trigger
- If high-intent queries or core keyword impressions drop sharply, revert the negative and narrow match type or level.
- Do not co-change
- Do not also change budget, RSA, Final URL, bid strategy, or page hero.
- Bring to the next review
- Next review brings Search terms, Change history, keyword impression change, and Shopify order quality.
- Current pressure
- Queries have purchase intent and CTR is acceptable, but add_to_cart / begin_checkout is weak, often from mobile or first-screen proof gaps.
- Choose when
- Choose this when search term, keyword, and product match, but GA4 page events show the post-click break happens on the page.
- Only primary variable
- Change one page proof point only: hero promise, price explanation, shipping/returns, review proof, or mobile CTA.
- Separate change-log template
- Change history / page record: 2026-__-__, Final URL ____, changed only ____, reason: GA4 ____ event was weak.
- Example observation plan
- Classroom plan: Watch 7 days or at least the prior click sample; read scroll / add_to_cart / begin_checkout before CPA.
- Rollback trigger
- If engagement and carts do not improve while CPA worsens, restore the previous page and inspect the next proof item.
- Do not co-change
- Do not add negatives, change budget, or switch bidding at the same time, or the page result becomes unreadable.
- Bring to the next review
- Next review brings Final URL version, GA4 landing page events, checkout error, and order SKU.
- Current pressure
- One day ROAS looks good or two orders appear, and the team wants to raise budget before conversion lag, search terms, and profit are read.
- Choose when
- Choose this when tracking is trusted, search terms show no obvious waste, page path is acceptable, but sample or profit proof is still thin.
- Only primary variable
- Either make one small average daily budget change or keep budget unchanged and extend the window.
- Separate change-log template
- Change history: 2026-__-__, budget from ____ to ____, change ____%, reason: ____ windows met ____.
- Example observation plan
- Classroom plan: Watch at least 3-4 days and do not stack budget, bid strategy, keyword, or page edits on different days.
- Rollback trigger
- If marginal CPA from added spend exceeds Max CPA, or contribution profit turns negative, return to the previous budget.
- Do not co-change
- Do not change tROAS, negatives, page, and budget together; do not raise budget repeatedly from one high-AOV order.
- Bring to the next review
- Bring average daily budget, spend, marginal CPA, break-even ROAS, Max CPA, Shopify order data, and separately calculated profit to the next review.
- Current pressure
- Ads reports conversions, but GA4 revenue and Shopify Orders do not reconcile, or transaction_id / value / currency is unstable.
- Choose when
- Choose this when every performance move depends on CPA / ROAS but conversion and order facts have not passed acceptance.
- Only primary variable
- Fix measurement only: conversion action, GA4 import, Shopify purchase, value, currency, or dedupe.
- Separate change-log template
- Change history / QA record: 2026-__-__, fixed ____, test order ____, Ads value ____, Shopify net sales ____.
- Example observation plan
- Classroom plan: Restart a clean observation window after the fix; pre-fix data is directional, not a budget or bidding basis.
- Rollback trigger
- If the fix creates duplicate purchase, wrong currency, or abnormal value, roll back tag/import settings and pause optimization conclusions.
- Do not co-change
- Do not raise budget, change tROAS, add negatives, or edit page before measurement passes.
- Bring to the next review
- Next review brings conversion action, GA4 DebugView, Shopify order, transaction_id, value, and currency.
Check competing explanations in three mixed-edit cases
One primary variable reduces ambiguity; a before/after comparison alone does not prove causality. Auctions, seasonality, weekday mix, traffic composition, and lag can still change. Use the three concrete mixed-edit cases below: identify the competing explanations, then state the remaining variable, frozen items, and next readout. Record the actual elements changed within a page version as well.
Inspect a mixed-change impulse
- Mixed-change case
- The 20oz tumbler gets 2 orders on day 7. The team raises daily budget from $30 to $60 and adds free, replacement, and bulk as negatives on the same day.
- What becomes unclear
- If CPA drops next week, you cannot tell whether budget found better traffic or negatives removed waste. If orders drop, you cannot tell whether budget learning moved or negatives blocked real demand.
- Keep only
- This cycle only adds one clearly irrelevant negative set. Keep budget at $30 and discuss budget next cycle.
- Frozen items
- Freeze budget, RSA, Final URL, landing page, and bid strategy.
- Next week asks
- Next week asks only: did high-intent impressions, clicks, CPA, and purchase quality improve because negatives cleaned traffic?
- Mixed-change case
- Search terms are relevant, but CVR is low. The team changes product-page hero, shipping copy, RSA headlines, and descriptions together.
- What becomes unclear
- If CVR improves next week, you cannot tell whether the page explained better or the ad promise filtered poor clicks. CTR movement also pollutes the page readout.
- Keep only
- This cycle only changes first-screen proof plus shipping/return clarity. Freeze RSA and keywords.
- Frozen items
- Freeze RSA, keywords, negatives, budget, and bid strategy.
- Next week asks
- Next week asks only: with the same high-intent click stream, did add_to_cart, begin_checkout, checkout error, and purchase improve?
- Mixed-change case
- One-day ROAS looks good, so the team wants to double budget and switch from Maximize clicks to Target ROAS.
- What becomes unclear
- If performance worsens next week, you cannot tell whether the budget jump was too large, bidding relearned, value signals were weak, or sample was thin.
- Keep only
- This cycle only makes a small budget action. Freeze bid strategy and discuss Target ROAS after value reconciliation and sample are ready.
- Frozen items
- Freeze bid strategy, conversion action, value source, Final URL, and negative structure.
- Next week asks
- Next week asks only: after the small budget move, are marginal CPA / ROAS, Search terms, Shopify net sales, and contribution profit still inside guardrails?
Read the full seven-day tumbler drill
Now use the full drill: a Search campaign sells a $29, 20oz leak-proof tumbler. In 7 days it spends $180 for 96 clicks. Ads reports 2 Purchases and Shopify shows 3 orders, including 1 from organic search. The page has 8 add_to_cart events, 4 begin_checkout events, and 1 payment failure. Keep this separate from the opening 80-click case.
First reconcile the two attributed Ads purchases with actual sources, order fields, value, currency, and dedupe. Ads 2 versus Shopify 3 is not inherently contradictory, especially with an organic order; explain attribution, time, and test status before diagnosing a defect. Nor do $180 and 96 clicks decide failure. At an affordable CPA of $45, spend is about 4 times CPA and supports a closer layered review; at $120, the evidence is thinner. These classroom comparisons are not a significance test or universal pause formula, nor do they establish that this $29 single cup can afford either CPA; check actual order value, costs, and contribution separately.
Read every query: leak proof tumbler and 20 oz travel mug show purchase intent; custom tumbler bulk may require wholesale/customisation you do not offer; free coffee cup conflicts with this paid offer; cup holder bottle depends on whether the page actually supports that promise. Eight carts show some page engagement, while 4 checkouts and 1 failed payment point toward shipping, price explanation, trust, or payment friction. Select one evidenced break instead of changing negatives and checkout together.
Do not jump from a readout to budget
Ads and order totals differ
- Readout
- Ads reports 2 purchases, Shopify shows 3 orders.
- Wrong reaction
- Change budget from platform ROAS.
- Read first
- Reconcile transaction ID, value, currency, and source first.
- Allowed this cycle
- Fix tracking or mark the readout; do not move budget.
A free query is spending
- Readout
- Search terms include free coffee cup.
- Wrong reaction
- Negate every term with no conversion.
- Read first
- Handle clearly irrelevant terms only; keep high-intent low-sample terms.
- Allowed this cycle
- Add clear negatives, freeze budget and page.
Checkout friction after carts
- Readout
- 8 carts, 4 checkouts, 1 payment failure.
- Wrong reaction
- Expand terms and budget.
- Read first
- The page and checkout already show friction.
- Allowed this cycle
- Fix one of shipping, returns, or payment path.
CPA exceeds the affordable line (here $45)
- Readout
- CPA is above the affordable line.
- Wrong reaction
- Switch to tCPA or cut budget hard.
- Read first
- Split sample, term quality, page CVR, and AOV / margin.
- Allowed this cycle
- Choose one primary variable and write rollback conditions.
Three week-one problems need different diagnoses
Low delivery
- Bad impulse
- Rebuild the campaign or double the budget.
- Diagnosis
- This is not yet a conversion problem; check search volume, rank, and ad approval first.
- Limit the move
- This cycle may only add high-intent terms or fix approval/rank problems; page and bidding stay frozen.
Relevant clicks, no orders
- Bad impulse
- Immediately negate all non-converting terms.
- Diagnosis
- High-intent non-conversion is not always a keyword problem; price, trust, shipping, mobile, or payment may be the blocker.
- Limit the move
- This cycle changes only first-screen proof or shipping/return promise; keyword structure stays frozen.
First order, high CPA
- Bad impulse
- Cut budget hard or switch to a tight tCPA immediately.
- Diagnosis
- First decide whether high CPA comes from low sample, poor terms, weak CVR, or insufficient AOV/margin.
- Limit the move
- Choose one variable only: control waste terms, fix page, adjust budget lightly, or wait for more sample.
High CPA may come from sample, query quality, page CVR, AOV (average order value), margin, or measurement. It does not automatically mean cut budget.
High-intent queries with few clicks, a higher-priced product, or a longer buying cycle are not all waste because they failed to convert this week. Keep watching them, or use the search terms lesson to decide whether their intent warrants a split.
Ads alone omits order facts, refunds, payment failures, page friction, and full costs. Changing copy, budget, page, and bidding together makes next week harder to explain. A first order alone does not prove the page has won, and a platform recommendation does not replace evidence.
Use the meeting to agree on a review
Use a 30-minute meeting to agree on one action rather than collect one edit from each person. The six phases identify what to discuss, what to produce, and which distraction to avoid. If the team cannot finish the review-and-rollback sentence, it has only exchanged impulses. The schedule is a teaching aid, not a platform requirement.
Meeting discussion, output, and distraction
Tracking
- Minutes
- 0-5
- Discuss
- Confirm tracking and order reconciliation.
- Output
- Pass / fail / direction-only.
- Avoid
- Do not discuss budget first.
Sample
- Minutes
- 5-10
- Discuss
- Confirm sample and observation window.
- Output
- Spend, clicks, conversions, days, affordable CPA comparison.
- Avoid
- Do not decide from one day.
Queries
- Minutes
- 10-17
- Discuss
- Read search terms and obvious waste.
- Output
- High-intent, suspicious, irrelevant terms.
- Avoid
- Do not negate every non-converting term.
Page
- Minutes
- 17-23
- Discuss
- Read page path and checkout signals.
- Output
- Browse, cart, checkout, payment failure, quick bounce.
- Avoid
- Do not blame all page friction on traffic.
Variable
- Minutes
- 23-28
- Discuss
- Choose the only primary variable.
- Output
- Allowed move, frozen items, responsible lead, review date.
- Avoid
- Do not change budget, terms, copy, page, and bidding together.
Rollback
- Minutes
- 28-30
- Discuss
- Write rollback conditions.
- Output
- Continue / pause / rollback trigger metric.
- Avoid
- Do not end with “we will see next week.”
Evidence before these moves
| Stop here | Go |
|---|---|
Changing budget before tracking QA. | Conversion, value, currency, and order reconciliation pass. |
| Evidence Conversion acceptance sheet and test order are complete. | |
Rebuilding campaign after two no-conversion days. | Judge impressions, clicks, terms, page path, and sample first. |
| Evidence No-conversion first-week router is complete. | |
Changing budget, bidding, keywords, copy, and page on one day. | One primary variable, frozen items written. |
| Evidence Decision log has one allowed change and frozen items. | |
Negating every non-converting term at once. | Classify high-intent, suspicious, and irrelevant before acting. |
| Evidence Search-term evidence explains each negative. | |
Copy the current references and complete your log
Common questions
- How long should Google Ads run before the first optimization?
Use about 7 days or one comparable spend window as the first review, but do not edit just because the calendar says so. Check tracking, sample, Search terms, page path, and conversion lag before deciding to extend, fix tracking, add obvious negatives, or change one primary variable.
- Does no conversion in week one mean I should rebuild the campaign?
Not necessarily. Zero conversions can come from low sample, short conversion window, untrusted tracking, weak query fit, poor page fit, or checkout friction. Run the no-conversion route first; rebuild only when evidence points to structural failure.
- Why should the first optimization cycle change only one primary variable?
If budget, negatives, RSA, Final URL, page, and bid strategy all change together, next week cannot explain what caused the result. One primary variable and frozen items make the next review reusable.
- How should I optimize when Search terms show waste?
Handle only clearly irrelevant terms such as free, jobs, repair, or wrong fit. Record negative level, match type, reason, and check window. Do not negate every non-converting term; full match type and negative governance belongs in the search terms lesson.
- When can I change budget after launch?
Only after tracking is trusted, Search terms show no obvious waste, page path is acceptable, and break-even ROAS, Max CPA, or contribution-profit guardrails are written. If proof is thin, extend the observation window instead.
- Can I change the landing page and RSA copy together?
Avoid doing that in the first cycle. If page and RSA / Final URL change on the same day, next week cannot tell whether page proof improved, ad promise narrowed, or query quality changed. Choose one primary variable and move the other into the next cycle.
- What should I do when Ads, GA4, and Shopify do not reconcile?
Pause CPA / ROAS budget moves and fix measurement first. Reconcile conversion action, GA4 purchase, Shopify Orders, transaction_id, value, currency, and duplicate counting. After the fix, restart a clean observation window.
- How should I write a rollback trigger?
Make it executable: what changed, how long to watch, which metric decides, and what gets restored. Example: if a new negative sharply reduces high-intent query or core keyword impressions, revert it and use a narrower match type or level.
Give the next review a clear question
Write which evidence supports the one change, what stays fixed, when to review, and what triggers continue/pause/rollback. For query quality, proceed to search terms, negatives, and match types; for pacing, return to budget and bidding; for page friction, use CRO evidence. Consider scaling only after query quality, measurement, page fit, economics, inventory, and cash are explainable.
Revisit ad metrics for the same-window readout. The later scaling lesson separately checks readiness, inventory, cash, and page fit. Find other routes in the Google Ads course index.