Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Google Ads › Expert

Steering Bidding Through a Sale and an Outage for a Garden Building Maker

You take an invented garden building maker through the situations where automated bidding learns least well: quotation campaigns with too few conversions to hold a target, a short sale that lifts conversion rate for a few days and a checkout release that stops the purchase tag the week before. You finish with a campaign structure and bidding signal for the quotation lines, a sale plan with its adjustment, budget and frozen changes, an outage response with a recorded window, and the rules for when and against what each period is read.
ExerciseExpert levelModule 1: Bidding through thin data, peaks and outages10 minUpdated

Why this matters

The advanced Google Ads level fed bidding better conversion signals, tested strategy and budget changes with geo splits and separated brand demand from new demand. All of that assumes a strategy with a steady flow of recent conversions to learn from. This exercise works where that assumption fails. Smart Bidding predicts each auction from what the account has recorded recently, so it struggles in three situations an expert has to run rather than avoid: conversions that are rare, a short period that is not like the rest, and a period when the recording itself breaks.

The guide to bidding with few conversions treats rarity as a problem of data, not budget, and works through the remedies in order: consolidate, choose what the strategy bids towards, pool, pick the least demanding strategy and judge over long windows. The guide to bidding through sales and tracking outages covers the other two: a sale raises the real conversion rate for a few days before the model has seen it, an outage makes conversions vanish while sales carry on, and each needs its own tool, with the changes that start learning held back around both.

The garden building maker and its spring

Skouvelt Garden Buildings is an invented maker of timber garden buildings. Sheds and summerhouses are sold online through a Shopping and a Search campaign on target ROAS. Garden offices, garden studios and log cabins are built to order and sold by quotation, each through its own Search campaign on target CPA. The site also lets visitors book a free design consultation, which is not yet a conversion action. A spring sale on sheds and summerhouses is planned, and in the week before it a checkout release goes out.

Illustrative example — not a real result

Skouvelt Garden Buildings: the account and the order records for last quarter

CampaignSellsBid strategyPrimary conversions recordedStatus
Garden offices (Search)By quotationTarget CPA £1207 quote requestsLimited by low conversion data
Garden studios (Search)By quotationTarget CPA £1205 quote requestsLimited by low conversion data
Log cabins (Search)By quotationTarget CPA £906 quote requestsLimited by low conversion data
Sheds and summerhouses (Shopping)Online ordersTarget ROAS 500%412 purchasesEligible
Sheds and summerhouses (Search)Online ordersTarget ROAS 500%158 purchasesEligible

Design consultations and the orders that followed, from the order records, by the campaign of the first click:

CampaignConsultations bookedBecame ordersAverage margin per order
Garden offices349£6,400
Garden studios195£6,100
Log cabins111£2,300

Offices and studios are built by the same workshop team and the owner sets one monthly budget for both. Quotation campaign spend last quarter: £4,500 on offices and studios, £900 on log cabins.

Last year's spring sale on the two shed campaigns (15% off, Thursday to Sunday):

PeriodDaily budgetSpendConversion ratePurchasesAverage order valueROAS
Two weeks before£350£4,7601.9%27£890505%
Sale, Thursday to Sunday£350£1,4003.2%13£760706%
Week after£350£2,3801.7%12£890449%

On the sale Saturday both shed campaigns were limited by budget from early afternoon. This year's sale is 20% off, Thursday to Sunday.

The week before this year's sale, shed and summerhouse orders against purchases recorded in Google Ads:

DayShop ordersPurchases recordedNote
Monday3130Normal
Tuesday2819Checkout release at 17:10
Wednesday301
Thursday, 08:30The daily check of orders against recorded purchases flags the gap

On Thursday morning the agency proposes lowering the target ROAS to 350% so spend keeps flowing, or switching both shed campaigns to maximise conversion value until the fix; the sales manager wants Shopping paused; the developer can release a fix by 14:00. Quote requests and consultation bookings are recorded by separate tags and are unaffected.

Thin data and unusual days

The two guides share a discipline. For thin data, the order of work matters: the account structure guide tests every campaign split against a control need, such as its own budget, target, location or conversion goal, and anything split by theme alone belongs at ad group level. Only then do you decide what counts as a conversion, which strategy the data can support and how long a judgement must wait.

For unusual days, the type of period decides the tool. A short, predictable rise in conversion rate is the case Google documents seasonality adjustments for, and an adjustment changes bids, not budget. A tracking outage needs the reverse: hold everything, and where Google documents it for the strategy, keep the window out of learning with a data exclusion. In both, a change to targets, strategy or conversion actions adds a second disturbance the model then has to absorb, and makes the period harder to read afterwards.

Task

  1. Decide which quotation campaigns merge and which stay separate, using the control-need test, and keep the themes that merge as ad groups.
  2. Choose what the quotation strategy bids towards, check the choice against the order records campaign by campaign, and say which outcome stays imported and in view.
  3. Choose a bid strategy for each quotation campaign, and when and how a target is added.
  4. Plan the spring sale: whether a Seasonality adjustment fits and how its expected change is estimated, the budget for the sale days and afterwards, the changes held back and the edits allowed during it.
  5. Respond to the outage on Thursday morning: what is done first, which proposals are accepted or refused and why, the window recorded and what is reversed once tracking is verified.
  6. Say when and against what each period is read: the outage weeks, the sale and the quotation restructure, and in what order the changes are made so each one can be read on its own.

Worked answer

Which campaigns merge. Offices and studios share a goal, a workshop, a margin and the owner's single monthly budget, so nothing but theme separates them: they become one campaign with an ad group for each. Log cabins stay apart, because their margin is a fraction of the others and a target shared with them would be wrong for both, which is the guide's reason for keeping unlike economics separate rather than pooling them in a portfolio.

What the strategy bids towards. Quote requests alone are too rare to learn from, so the design consultation becomes a primary action beside them, counted once per click, and orders from the order records are imported as a secondary action, following the Google Ads setup for lead generation. The order records are the check the guide insists on: consultations from offices and studios turn into orders at a similar rate, so the step predicts the sale for the merged campaign. For log cabins it does not, and a frequent step that rarely becomes a customer teaches the strategy to buy cheap, useless conversions, so cabins keep quote requests only.

Strategy and targets. The merged campaign runs on maximise conversions without a target for a quarter, the least demanding conversion-based strategy that the comparison of maximise conversions and target CPA recommends while conversions are very few. Cabins, with a handful of quote requests a quarter and a low margin that needs a firm price ceiling, move to maximise clicks with a maximum cost per click, a case the comparison of manual and automated bidding lists. A target is added to the merged campaign only after the quarter, near the cost per primary conversion it achieved, inside the ceiling the CPA calculator gives from customer value and conversion rate, and then left alone.

The sale plan. The sale is short, its dates are known and last year gives an estimate, so a seasonality adjustment fits, set for the four sale days only. The estimate is the change in conversion rate during last year's sale against the weeks around it; last year's budget cap cut purchases on the Saturday but not the rate per click, so the rate is still usable, and the deeper discount this year argues for expecting at least the same rise. Because a sale lowers order value while it raises conversion rate, the expected return is checked with the ROAS calculator before the target is left where it is. The adjustment leaves the budget alone, so the daily budget is raised for the four days before the sale starts, lowered on the Monday, and month-to-date spend is checked each day because the monthly limit recalculates with every change. Targets, strategies, conversion actions and structure are frozen from the plan's sign-off until the sale has been read; offer ad copy and negatives may change. The sale promotion readiness checklist holds these decisions beside the offer, feed and site checks.

The outage. The tracking outage response SOP runs it. The daily check has already confirmed the outage against a source outside the tags, so the analytics owner opens the incident and the paid media manager protects the two shed campaigns: budgets and targets held, a data exclusion applied for the window where Google documents one for target ROAS. Both agency proposals are refused, because a target change or a strategy switch during an outage teaches the model the false collapse as normal. Pausing is refused too: it is a last resort for campaigns clearly overspending on bad signals, and these are not. The fix is released, verified with real purchases rather than test events, and the window is recorded from the release time to the point where recorded purchases are back in their usual relationship to orders. Once recovery is confirmed, the exclusion is closed, the window is annotated in reports and the SOP's last step uploads the missing orders where the platform supports it.

Order and reading. The sale is never judged on its last days, because Conversion Lag keeps crediting purchases back to its clicks after it ends. It is read once delayed conversions have arrived, against the same weekdays without a promotion and last year's sale, at account level, with the weeks after watched for overbidding and for orders the sale pulled forward. Whether the sale paid, net of the discount and the borrowed orders, is the question the guide to measuring a sale or promotion answers. The quotation merge waits until the sale has been read, so neither change muddies the other, and it is judged over a full quarter against the quarter before and the same quarter a year earlier, because the counts are small and Sample Size makes a few weeks noise. Recorded conversions are confirmed back in step with orders before any of those comparisons start.

Illustrative example — not a real result

Skouvelt Garden Buildings: what the periods showed once they could be read

PeriodRead onResult
Outage windowTuesday 17:10 to Thursday 15:40, excluded49 orders uploaded after the fix; recorded purchases matched orders again from Thursday evening
Sale, Thursday to SundayTen days after it endedDaily budget £900; spend £3,210 of £3,600 planned; conversion rate 3.4%; 31 purchases at £715; ROAS 690%; never limited by budget
Week after the saleThe following fortnightConversion rate 1.8%; ROAS 472%, in line with the weeks before; no sign of overbidding
Quotation restructureA full quarterSee below
MeasureQuarter beforeQuarter afterSame quarter last year
Spend, offices and studios£4,500£4,600£4,300
Primary conversions that campaign's strategy saw12 quote requests13 quote requests and 49 consultations10 quote requests
Office and studio orders, from order records141712
Log cabin orders121

The merged campaign achieved £74 per primary conversion, so the next quarter adds a target CPA of £80 and holds it. With orders this few, the team reports the result as encouraging against both baselines, not as proof.

Common mistakes

  • Automating before consolidating. Three thin campaigns learn less than one campaign with their combined data.
  • Bidding cabins on consultations. A frequent step that seldom becomes an order steers money the wrong way.
  • Raising the target for the sale by hand. It starts learning and is easily left at the sale level afterwards.
  • Setting the adjustment and forgetting the budget. The model bids harder and the budget runs out by midday.
  • Changing the target or strategy during the outage. Two disturbances, one unreadable result.
  • Reading the sale on the Monday after. Conversion lag fills in its last days over the following week.
  • Restructuring the quotation campaigns in sale week. Every comparison then mixes two changes.

The checkpoint that follows in this module tests the same judgements and adds the guide to scaling Google Ads spend: where headroom sits once the data is sound, and how each step of growth is judged at the margin.

Pages this lesson applies

Terms used on this page

Also related

Related