Start here

This is how Tigrowth runs Meta and Google Ads for ecommerce brands on Shopify. Meta creates demand, Google captures it, creative is the lever, the structure stays as simple as the business allows, and every budget step is earned by results the store's own sales data agrees with.

It is the playbook our media buyers work from every day: the defaults, the gates a change must pass, the checklists that stop expensive mistakes, and the standards every build is checked against. Use it to onboard a new buyer, to audit an account you have inherited, or to check your own.

Written for ecommerce brands on Shopify spending on Meta and Google. Lead-gen businesses need different conversion rules.

Seven principles

These override everything else in the playbook, on both channels.

1. Every change solves a written problem. Name it, show it in at least 7 days of data, then change one thing. If unsure, wait: both platforms punish impatience.
2. Fix tracking before touching budget or bids. The platforms learn from what the tag reports. A broken tag trains them on the wrong customers.
3. The store's numbers decide, not the platform's. Judge on Shopify new-customer revenue and MER. Platform ROAS is a claim, read as an upper bound.
4. Simple structure by default. Every campaign or ad set needs a business reason to exist. More campaigns split budget and shrink the data each decision rests on.
5. Creative and offer are the main lever. Targeting is mostly automated now. The ad, the offer and the landing page are where an account is won.
6. Earn every budget step. Raise in gradual steps, let each settle, re-check. Cuts can happen any day. Never change budget and bidding together.
7. No number without its comparison. Every metric in every report carries the prior period and last year.

How Meta and Google fit together

Meta puts the brand in front of people who were not looking; some of them search for it later, and Google catches that search. So Meta is judged on new customers and blended MER, not its own ROAS, and Google's brand campaigns are reported apart so they cannot hide a falling non-brand line.

The shared measurement basis

• The business number leads every report: MER from Shopify, with net sales beside it, then NC MER and new customers.
• Platform numbers sit beside Shopify with the gap stated. Meta reads at 1-day click and incremental; Google states its attribution model on every figure.
• Year-on-year compares like with like: sale days to sale days, with creator-code spikes, bot traffic and dead tracking weeks removed first.
• In markets where consent is opt-in, pixels go quiet before consent. Judge those markets on Shopify, not platform ROAS.
• Where budget allows, test incrementality before a large bet: a Meta conversion lift study, a Google geo test. Until then, treat claimed revenue as an upper bound.

The rhythm

Every account gets the same loop on both channels. Most of the work is reading, not changing.

The weekly read starts from data the tools have already pulled (breakdowns on Meta, script Sheets on Google), so the hour goes on judgement, not exports.

Glossary

Meta Ads

Meta creates demand. The default account is one campaign with one ad set, optimised for purchases, on Incremental attribution, with broad delivery and no audience inputs. Creative is the lever, and every budget step is earned by New Audience results at 1-day click that Shopify agrees with.

1. Account shape

Every extra campaign or ad set splits budget, overlaps in the auction and shrinks the data each decision rests on. £100 a day in one ad set beats £100 across five. In one account we inherited, every purchase-optimised ad set was stuck in Learning Limited because the budget was spread too thin for any of them to learn.

Business requirements that earn a split

Each split still needs enough budget to drive results on its own: roughly 25 to 50 purchases a week is a guide. Combine similarly priced markets unless splitting adds real volume.

Not reasons to split

• Separate testing campaigns or ad sets. They take budget from the main goal and produce low-volume data. New creative goes into the live ad set.
• Ad sets split by audience (broad vs lookalike vs interests). They reach the same people and compete.
• Standalone retargeting or catalogue retargeting campaigns. Meta remarkets inside broad delivery already (section 7).
• Separate top, middle and bottom of funnel campaigns. Meta sequences delivery itself.
• An ad set built to force spend onto an ad Meta ignores. Use Push Delivery instead.

Catalogue ads sit as ads inside the main ad set. Advantage+ Sales is fine as the one campaign.

What a standard account looks like

Persona ads, catalogue ads and creator ads all sit in the same ad set and compete in the same auction, so Meta spends on whichever finds buyers. A promo campaign switches on for a sale and off again. A second always-on campaign appears only for a business requirement from the table above.

2. Signal and conversion tracking

No spend decision is trusted until browser and server events are flowing, deduplicated, on the right dataset, and Meta purchases sit within an agreed tolerance of Shopify orders.

Setup, week 1 of any account

1. Confirm which dataset the site actually fires, with server events in the last 7 days. Old pixels die quietly, and audiences built on them shrink to nothing.
2. Pixel plus Conversions API (the Shopify app or Meta's one-click setup), deduplicated. Check the connected status, the Integration column reading "Multiple", and server events in the breakdown.
3. Domain verified. Event Match Quality: Purchase 8+ (9+ for premium brands), Add to Cart 8+.
4. Optimise on standard events only. Custom conversions are reporting columns (a product, a collection, a price band).
5. Consent: in opt-in markets, pixels and CAPI are silent before consent. Judge those markets on Shopify.
6. UTMs in the creative's URL parameters field, not appended to the link.
7. Place a test purchase end to end and check Events Manager for duplicates.

Monthly, and after any site or app change: Meta purchases vs Shopify orders by day, EMQ, Events Manager diagnostics, catalogue health. A gap freezes budget steps until it is explained.

Events, done right

• Standard events to optimise; custom conversions as reporting columns only (optimising on one throws away sales of every other product); custom events only for things with no standard equivalent, such as a 15-second visit.
• Traffic quality check: the 15-second-visit event divided by landing page views, at 1-day click and first conversions, compared across ads.
• Results that look inflated are a setup problem first: events firing too early or on the wrong page, team refreshes of the confirmation page, missing deduplication.

3. New account launch

A new account launches as one campaign, one ad set, Purchase, Incremental attribution, no bid caps, and stays there until it reaches around 50 purchases a week for several stable weeks.

• Budget: a conservative cost per purchase and a spend the brand is comfortable with. With no history, assume a £100 product costs £100 to sell. Sanity check: cost per purchase x 50 is the minimum weekly budget for the main campaign. Below it the ad set sits in Learning Limited, which does not matter when results are good. Say so before launch.
• Season the account first: low spend on approved ads so Meta raises the spend limit before real budget lands.
• Expect a CPM pit: the first few thousand pounds feel expensive while Meta learns who buys.
• Creative: 1 to 3 distinct persona ads, each with 9:16 and 4:5, a static and a video, five texts and five headlines. Around half the Reels as creator or partnership ads where the brand has creators.
• Audiences: broad, Advantage+ audience on, no inputs.
• Catalogue: fashion brands add catalogue ads once the catalogue is clean (section 7).
• Structure is added only for a business requirement, never to graduate by funnel stage.

4. Audiences and exclusions

Go hands-off: Advantage+ audience on, no age, gender, interest or lookalike inputs, and controls only for legal or proven reasons. Read where spend lands through audience segments.

Audience segments, every account. Define them in Advertising Settings as broadly as possible. Engaged = all website visitors 180 days, the email list and other engagement audiences. Existing Customers = purchasers 730 days plus customer lists. Every report and every scaling call reads the New Audience row. Narrow definitions push buyers into New Audience and flatter prospecting.

Existing customers are not a problem by default. Broad delivery typically spends 20 to 25% on engaged people and existing customers, higher at first and falling as that pool is used up. Act only on a confirmed issue in the segments breakdown, such as existing customers taking a large share of a promo at lower margin. Fixes, in order:

1. A value rule bidding less for the Existing Customers audience.
2. Customer Lifecycle Strategy. Check its pre-filled exclusions by hand; they are often wrong.
3. Full exclusion, as a last resort.

Value rules before restrictions. Before any value rule, answer two questions: do we know something Meta does not (margin, lifetime value), and which breakdown proves the problem? Value rules raise costs by design, so one per proven problem, logged like a bid change. They are also how Meta now replaces placement exclusions.

Keep audience and custom conversion names neutral. Meta disables audiences with health terms in their names.

5. Creative

The ad, the offer and the landing page are the main lever. Build diversity in phases: start with 1 to 3 strong persona ads, judge in aggregate after two weeks, and add a completely different batch only when results call for it.

Phases, not volume

1. Start with 1 to 3 ads, each built on one persona and their problem. One ad can hold up to 10 images and videos.
2. Run at least two weeks and judge the ad set in aggregate. Meta serves warm audiences first, so early results flatter.
3. Results good: leave it alone.
4. Results not good: build a batch that looks completely different (new persona, angle, visuals, text). Near-duplicates are not diversity.
5. Still poor after several rounds: look at the landing page and the offer, one change at a time.

Launching 20 to 40 ads at once mostly means Meta shows two. In one account, 21 of 37 live ads spent under £50 in 14 days. If results are poor and many ads barely deliver, pause the low-delivery ads (judged over two weeks, never one to three days) so delivery can focus.

How many ads to run

An account can end up with 30 or 40 ads. That is fine when each batch was added because the results asked for it.

Fatigue and frequency. Most "fatigue" is Meta using up the warm audience it served first. Do not refresh on a calendar or a frequency number. Check the audience segments breakdown, and add a batch when New Audience results at 1-day click fall across two 7-day windows. High-AOV products need repeated exposure, so higher frequency can be normal.

Let Meta pick the ad. One ad taking 50% or even 99% of spend is normal. Meta tests on small samples, then moves budget to the ad it predicts will win, so that ad can show a higher average CPA and still be the right call (the breakdown effect). Do not turn off the favoured ad on averages. If you doubt it, use Push Delivery on the under-served ad for a few days; it usually confirms Meta was right.

Every ad

• Built in the new creative workflow: up to 10 media per ad, three aspect ratios per creative, placements and URLs set per creative.
• 9:16 plus 4:5 or 1:1 so every placement is covered; at least one static and one video in each launch; videos 5 seconds or longer.
• All five primary texts and headlines at varied lengths, built on the customer's problem. Leave Meta's text and creative variations in the mix.
• Customise enhancements before turning any off; site links, promotions and product cards off unless wanted.
• The CTA button matches the goal; no CTA split tests.
• Video: hook in the first 3 seconds, captions on. Aesthetic brands lead with collections, colourways and styling.

Where ideas come from

• Revive past winners only on this year's numbers. Last year's results include engagement clicks Meta no longer counts (section 8), so they overstate what a relaunch will return.
• Model long-running competitor ads from the Ad Library; adapt, never copy.
• Organic posts that earned attention go into the sales campaign as creative. Never the Boost button.
• Creator and partnership content (section 12).
• Before pulling an ad that carries a creator discount code, check the code is still redeeming in Shopify.

Compliance: UK ASA / CAP Code on every claim, Meta policy (no personal-attributes framing), and the brand's own guidelines.

Benchmarks: build each brand's good and great lines (ROAS, CTR, CPM, revenue per day, split by full-price vs promo, concept and format) once there are six months of ad-level data. State the attribution basis on every line; lines built on 7-day click cannot be read against Incremental ad sets.

6. Budgets, bidding and scaling

Scale only what works on New Audience at 1-day click, in gradual steps you let settle and re-check. There is no magic percentage, and a budget change does not reset learning.

Setting a budget. Start from the goal and a conservative cost per purchase. About 25 to 50 purchases a week per ad set is a guide, not a rule: if results are good, ignore Learning Limited. More structure makes any threshold harder to reach.

Is it scalable? All yes:

1. New Audience at 1-day click is good enough on its own over the trailing 7 days, reading the audience segments and attribution breakdowns together. Results propped up by remarketing, view-through or engage-through collapse when budget rises.
2. Stable across 1-day click, New Audience and the incremental column, on meaningful volume: 10+ purchases at 1-day click in 7 days as our floor.
3. Shopify NC MER on target for the same 7 days, and Meta within tolerance of Shopify orders.
4. Inside the brand's agreed monthly budget.

Expect costs somewhat worse than the New Audience baseline: Meta has to reach past the likeliest buyers.

How to raise. Increase, let it settle for several days, check New Audience at 1-day click, repeat. Steps of 10 to 20% with a few days between are the safe default; bigger jumps are fine on a short runway if the brand accepts the risk. Find the point where it stays profitable and stop there.

Cuts: any size, any day. Lowering budget does not restart learning. Cut when NC MER misses target across a 7-day window, or the same day for a leak. Every promo plan carries post-sale step-down dates: budgets left at sale level after a sale are one of the most common leaks we find.

Peaks and sales. Use scheduled budget increases on campaigns already proven in full-price trading, not new campaigns. They take an amount or a percentage, up to 50 periods, and revert on their own. Run a daily budget you can sustain for two weeks or more before the first peak day, without stopping and starting.

Bidding. Highest volume only. No cost caps, ROAS goals or bid caps: if something should not scale, a cap just stops Meta spending. To shape delivery, use value rules.

Structure change. All yes:

1. The problem is written down and shows in 7+ days of aggregate data.
2. It cannot be fixed inside the existing ad set (new ads, a value rule, a catalogue filter).
3. Settings that cannot be edited later are decided once: attribution, pixel and event, promoted product set.
4. The brand has approved. Build paused, run the pre-launch checklist, then enable.

Margin targets. CPA below new-customer contribution margin. Example: AOV £100, contribution margin £50, target CPA £40 gives a 2.5x ROAS floor.

7. Retargeting and catalogue

Remarketing happens inside broad delivery, typically 20 to 25% of spend, so we do not run standalone retargeting or catalogue retargeting campaigns by default.

A separate retargeting campaign chases the same people twice, overlaps in the auction and fatigues fast on a small pool. Its strong numbers are mostly view-through and engage-through credit for people who were buying anyway.

The two exceptions

1. A high-ticket offer on a limited budget.
2. A tiered model: low-ticket buyers nurtured to a high-ticket upgrade.

When one applies: a combined website and customer list audience, Advantage+ audience off, Awareness objective with Maximise Reach, frequency cap 1 to 2 a day, feeds, Reels and Stories only. Report it on 1-day click only, and make the ads echo the emails doing the real work.

Catalogue ads (fashion and multi-product brands). Catalogue ads go inside the main ad set, so Meta shows product sets to whoever it judges likely to buy.

• Promote bestseller product sets filtered on custom labels, not on product-name text. A "name does not contain Womens" filter once let a single womens polo take a third of a mens catalogue's spend.
• One gender's set never serves the other; no sale-only items in full-price sets.
• Check the product breakdown on every catalogue ad weekly.
• Make catalogue creative look like the brand: branded backgrounds, overlays, lifestyle frames, not product-on-white for premium brands.

8. Reporting and measurement

Never take Meta's default number at face value. We optimise on Incremental attribution, judge on 1-day click and incremental conversions, and report Meta's contribution to Shopify NC MER and blended MER.

Attribution setting

• Default: Incremental. Meta cannot pad results with view-through credit, which also stops it leaning into remarketing. Results look lower because view-through, engage-through and many 2 to 7 day click conversions fall away; most of those were closed by another channel.
• Standard (7-day click, 1-day engage-through, 1-day view) only where a brand needs continuity with past reporting. Then judge on 1-day click.
• Attribution cannot be changed after an ad set is created, so set it at build and put it in the ad set name.

The 2026 click redefinition. Click-through now counts only link clicks. Likes, shares, saves, comments, profile taps and video plays moved to a new 1-day engage-through window in 2026, and the engaged-view threshold fell from 10 seconds to 5. Meta-reported year-on-year is not like for like across the change, because last year's 7-day click included engagement clicks. Use Shopify or GA4 for year-on-year. Creator ads that read 10x or more last year are the most affected.

Breakdowns read every week

• Audience segments: spend and results for Engaged, Existing Customers and New Audience. Read first in any account.
• Attribution setting (Standard only): 1-day click, 2 to 7 day click, 1-day engagement, 1-day view. You want a heavy share in 1-day click.
• Conversion count: first conversions vs all. 15 purchases is not 15 customers.
• Compare Attribution Settings: add the incremental column, 1-day click and 28-day click.
• Custom conversions as reporting columns to see what actually sold.

What the monthly report says

1. NC MER, blended MER, new customers and nCPA from Shopify, against prior period, last year and target.
2. Meta's 1-day click and incremental purchases beside Shopify, with the gap stated. GA4 lines up closest with 1-day click; the rest is impact GA4 cannot see.
3. A creative read: which concepts and personas bring new customers in cheapest.

9. Premium and luxury brands

High-AOV brands run the same simple account. What changes is the creative, and how patiently results are read across a 60 to 180 day consideration cycle.

• Measure on Incremental, NC MER and blended MER, plus 28-day click in Compare Attribution Settings. Higher frequency can be normal.
• Standalone remarketing only under the high-ticket, limited-budget exception.
• Widen the on-brand range without cheapening it: more locations, textures, palette shades and styling formats. A locked studio look gives Meta too little variety to match ads to people.
• Persona-led batches. The fashion funnel shapes the creative, not the campaigns: collection ads to collection pages, single-product ads to product pages, justification ads (materials, longevity, warranty) to product and story pages, all in the same ad set.
• Partnership ads inside the main ad set; a separate ad set only when a creator contract fixes a spend and dates.
• Merchandising is the offer: new collections, drops and collaborations, not discounts.

10. Testing

Testing is optional, and there are no separate testing campaigns or ad sets. New creative goes into the live ad set and is judged in aggregate over two weeks or more.

Do not optimise a test for link clicks to find high-CTR ads, and do not hunt for one winning combination: a good ad wins differently for different people. Log every test and its verdict in the change log.

11. Account ownership and compliance

The brand owns every asset and adds the agency as a partner. The agency never logs in as the brand.

• The brand's own business portfolio owns the Page, Instagram account, ad account, dataset, catalogue and payment method. The agency's portfolio is added as a partner, asset by asset, never as a person.
• Contractors get the least access that does the job. 2FA for everyone; an account spending limit; a primary and a backup payment method.
• Ignore partner requests from unknown businesses, even through genuine Meta emails.
• Customer lists use consented data only (GDPR and PECR). Landing pages match the ad, work on mobile and carry a privacy policy.
• AI tools and connectors that read ad accounts get the brand's approval first, view-only by default, with write access only where the brand has agreed.
• Monthly audit: the ACTIVE list matches the plan; disapprovals low; users and 2FA current; payment valid; CAPI firing and deduplicated; catalogue healthy; audience segments defined.
• Offboarding: remove the partner access, hand over the change log and naming key. The brand keeps every asset.

12. Creator and partnership ads

Creator content is often the strongest creative an ecommerce account has, because a real person who is not the brand does the talking. Partnership ads run it with both names on the ad: the creator's handle and the brand's.

Where they sit. Partnership ads are ads like any other: inside the main ad set, judged in aggregate with everything else (section 1). A separate ad set only when a creator contract fixes a spend and dates.

How much. On a new account, aim for around half the Reels to be creator or partnership ads. Premium brands with smaller audiences lean on them harder, often 20 to 50% of spend as they scale. Let results set the share: Meta moves budget to the creator ads when they win.

Choosing creators

• Pick creators who look and sound like the brand's most profitable customers, not the ones with the biggest following.
• Organic skill is not ad skill. Ask for raw footage and see how the creator handles a hook before a long deal.
• Start with 3 to 5 creators so one poor fit does not decide the channel.
• Agree usage rights that cover the whole ad run, and a fee or revenue share that protects margin.

Setup

• The creator grants partnership permission from their account. Check it covers the dates you plan to run, and renew it before it lapses.
• Build in the new creative workflow like any ad: 9:16 first, 4:5 or 1:1 as well, five texts and five headlines.
• Put the creator's handle in the ad name so results join up later.
• Find creators and their existing content in Meta's creator marketplace, now merging into Meta Creator Marketing Hub.

Creative brief. Give creators the persona and the problem, not a script. Hook in the first three seconds, product on screen early, the creator's own words. Ask for several hooks on the same body so one shoot gives several ads.

Creator codes

• Give each creator a unique discount code, so Shopify shows their sales whatever the attribution window.
• Check redemptions in Shopify before pulling a code ad: codes nobody is watching are often still selling.
• Strip code spikes before reading year-on-year trends.

Measuring them. Creator ads are where the 2026 click change bites hardest: likes, saves and comments no longer count as clicks, so last year's creator results look better than they were. Judge them on New Audience at 1-day click, code redemptions and Shopify NC MER (section 8).

Compliance. The paid partnership label covers the ad. Any post a creator makes for the brand on their own feed needs clear #ad disclosure under the UK ASA / CAP Code.

Google Ads

Google captures demand. Audit before changing anything, give every campaign a lever reason to exist, change bidding slowly on 30 to 90 days of data, and judge Google against Shopify, not against itself.

1. Taking over an account

Never stop an inherited account and rebuild it. Audit first, keep what converts, and change in stages over 3 to 6 months.

The three-minute triage. Run it on products (Shopping, PMax) or keyword themes (Search) before the full audit:

The audit ends in a traffic light plan, shared with the brand:

• Red: urgent fixes. Tracking, disapprovals, spend leaking with no return.
• Amber: important, not urgent. Structure, feed, bidding changes on the gate.
• Green: working. More budget, or leave alone.

If add-to-cart, page views or other soft actions are set as primary conversions, the account cannot scale: smart bidding is optimising for the wrong thing.

Rules for the first changes

1. Fix tracking before you change bidding. Smart bidding learns from what the tag reports.
2. Take the easy wins first: negatives, feed errors, disapprovals, obvious villain products.
3. Change mid-tier campaigns and products first. Leave the best performers alone until the new structure has proved itself.
4. One structural change per campaign at a time, then wait two weeks before judging it.
5. Every campaign needs a lever reason to exist (section 2). If nobody can say why it exists, it goes on the list to merge or pause.

The full week 1 audit checklist is on the Checklists tab.

2. Account structure

Every campaign exists to give us a lever: a way to push spend up or down on one thing without touching the rest. More campaigns is not more success.

Before launching any campaign, answer three questions in the build doc: what is it for, what will it do to the campaigns already live, and could an optimisation in an existing campaign do the same job?

What a grown account looks like

Each line has one lever reason. A new account starts with the BOF campaigns and adds the rest stage by stage, as below.

Duplicate targeting. Only one ad from the account enters any auction, so the same product in two campaigns is not cannibalisation. The problem is the same products, to the same audiences, in the same places, with different bidding: any optimisation in one campaign is undone by the other. When several PMax campaigns run, add URL exclusions so each reaches only its own category pages.

The funnel and what each stage says

Accounts that run only bottom-of-funnel campaigns are fragile. Marketplaces can outspend you on buying-intent terms, so every account past stage 2 needs some middle and top.

Budget split by stage

PMax-heavy accounts. When PMax takes 80% or more of spend with no funnel around it, more budget will not fix it. Fix the structure before scaling. PMax also supports Search: in one account, pausing PMax pushed Search cost per conversion from $125 to $201 with Search budget unchanged. Never pause PMax without watching Search and Shopping for the next two weeks.

Demand Gen. Use it once bottom-funnel campaigns have scaled to the point where CPC climbs and traffic stops growing. It works more like Meta than Search: audience and creative do the work.

• Budget 5 to 10% of the account while testing.
• One ad group per audience, with every format inside it. Do not split ad groups by placement or format.
• Go cold: customer lookalikes and custom segments on competitor and industry sites beat warm remarketing.
• Never pause high-spend ads or videos people watch to the end.
• Exclude the app network; check placements monthly.
• Judge it on the whole account and the whole business, not its own conversion column.

Search in 2026

• Fewer keywords and fewer ad groups. Broad plus exact match in each ad group; no phrase match.
• Broad match catches new and AI Mode queries but needs clean conversion tracking. Avoid it in tight niches, premium products and restricted categories.
• AI Max: never switch it on inside an existing Search campaign; it changes targeting enough to break the current bid strategy. Test it in its own campaign and check brand share in its search terms.

Build specs

3. Feed and product tiers

For Shopping and PMax the feed is the targeting. Titles decide which searches you enter, tiers decide how much you spend on each product, and AI attributes decide whether you qualify for Google's new AI Shopping placements.

Product tiers. Tier every product on 90 days of product-level ROAS, refreshed monthly, and write the tier into a custom label so campaigns can filter on it.

Agree the label map per brand and leave at least one label free. Check which labels each PMax campaign actually filters on before changing a label's logic.

Titles and images, monthly. Up to 150 characters with the most important details in the first 70. Order: product name, the words people search, brand, product type, then gender, size, colour, material. Pull new title words from the search term report. Test product shots against lifestyle images.

Merchant Center health. Weekly: product warnings and disapprovals. Monthly: site speed under 2 seconds, 6 to 8 images per product, returns and shipping listed, store rating, promotions loaded for every sale.

AI attributes, every brand, now. Google's AI Shopping ads write the ad from feed data. An empty attribute means no answer, and no answer means no ad.

Ship them through your feed tool, or a Google Sheets supplemental feed. Start with heroes and the most profitable products.

4. Bidding and budgets

A bid strategy or target change resets learning, so change slowly, on 30 to 90 days of data, and never budget and bidding together. If unsure, wait.

The change gate. Before switching to smart bidding, or adding or moving a tROAS, every answer must be yes:

1. At least 30 purchases in the last 30 days.
2. Weekly conversions within 20% of each other for the last 4 weeks.
3. No bid strategy or tROAS change in the last 6 weeks (strategy type: no more than once every 60 to 90 days).
4. No budget change in the last 2 weeks.

Results often dip for 1 to 2 weeks afterwards. Judge on conversions and value, not CPC, and make no call on fewer than 60 days.

Setting tROAS. Set it on actual ROAS from the last 30 to 90 days, never on the ROAS you want. Climb it like a ladder (200%, then 250%, then 300%), waiting at each rung. Too high cuts total revenue; too low buys poor traffic. The right target maximises total revenue, not ROAS.

Budgets

• Up 20% every 5 to 7 days, only once conversion metrics have recovered. Under £50 a day, step £10 to £20.
• Ready to scale when conversion metrics are good and Search impression share is under 65%.
• Plan the year as growth and cutting phases. Growth quarters push volume; cutting quarters push ROAS. Do not ask for both at once.

How to scale. Two levers, in this order:

1. Win more of the auctions you already target. If conversion metrics are good and impression share is low, raise budget.
2. Reach colder traffic. Once Search impression share passes 60%, move to benefit-led search terms, PMax on new customers only, Demand Gen or Video.

If an account will not scale, it is usually the wrong structure (spend pooled in overlapping campaigns) or the wrong data (soft conversions as primary). Fix those before adding budget.

5. Optimisation cadence

Every account gets a weekly hygiene pass, a monthly optimisation pass and a 90-day structural review. Start each pass from the script Sheets (Operating model), so the time goes on judgement.

Weekly, 30 to 60 minutes per account

• Campaigns, ad groups and asset groups with bad conversion metrics noted, fixes queued
• Budget pacing against the monthly plan
• Search term audit: negatives added; converting terms added as exact match
• Keywords with outlier CPC or quality score under 5
• Product warnings and disapprovals in Shopping and PMax
• Disapproved or limited ads and assets

Negative keyword rule. Add a negative only for a top 10 to 20 spending search term with no conversions for 2 to 4 weeks. Exact match negatives; be very careful with broad.

Monthly

• Campaigns ready to scale; good-metric, low-spend groups broken out
• Product tiers refreshed
• Location, device and audience performance
• Auction insights when CPC rises
• PMax search term insights, brand exclusions, placement exclusions
• Bottom assets replaced; video retention read and new hooks briefed
• Is tROAS limiting performance? Falling value, spend well under budget, or impressions falling without an explanation
• All conversion actions recording; no auto-apply recommendations switched on

Every 90 days

• Bid strategy gate for campaigns still on Max Clicks
• Day-of-week and hour-of-day; age exclusions in PMax and Demand Gen
• Landing pages: CVR by page, PageSpeed, price, delivery and returns visible
• Structure: every campaign still has its lever; no duplicate targeting; funnel stage and budget split
• Benchmarks: Search CTR 7% and CVR 3%; Shopping CTR 0.9% and CVR 3%; Display CTR 0.6% and CVR 1%. Below these, fix ads or pages before adding budget

6. Peak and promo playbook

Peaks are won in the three months before them, and protected by splitting the account into always-on and promotional campaigns.

The offer comes first. Run every offer through three tests: it makes financial sense (no discount past margin), it drives more total revenue than not discounting, and it attracts new customers whose lifetime value is profitable. A discount is one option among bundles, gifts with purchase, free delivery and gift cards.

Black Friday timeline

Promo build checklist

• Offer in every live campaign's copy, images and video
• Promotion and price assets; countdown in headlines
• Promotional PMax asset group
• Sale price set so the cross-out shows
• Merchant Center promotion approved before launch day
• No bid target capping spend; budget-limited status checked daily during the sale
• Guardrail written down: the ROAS floor, and how many days under it before you pull back

7. Reporting and measurement

Google's ROAS is a platform claim, not the business result. Report Google against Shopify, split brand from non-brand, and never show a number without its comparison.

• Every metric carries the prior period and last year.
• MER from Shopify leads, with net sales beside it.
• Brand and non-brand reported separately for Search, Shopping and PMax. Brand ROAS hides non-brand decline.
• State the attribution basis on every Google number.

Incrementality. Treat Google-claimed revenue as an upper bound. Value the next pound at the marginal rate, not the account average. Before a large upper-funnel bet, run a geo test: the campaign in selected regions only, all sources compared 30 days before and after.

Session economics. Revenue per session should be at least 2x cost per session. Below that, the fix is on site or in the offer, not in the ad account.

8. Conversion tracking

One purchase conversion, counted once per order, with value, is the only primary goal for ecommerce.

1. Purchase tracking from one source only: the Shopify Google app or a GTM tag, not both.
2. Enhanced conversions on.
3. Consent mode v2 where consent banners run; note the opt-in markets.
4. Add to cart, checkout and page views as secondary only. GA4 purchases imported as secondary, never a second primary.
5. Customer lists uploaded monthly for new-customer goals.
6. Test: Google Ads conversions and value against Shopify for the same 7 days, within 10%. Rerun within 48 hours of any theme, checkout app or consent banner change.

9. Experiments

Test one thing at a time, outside peaks, with a written hypothesis and a pass mark set before the start.

A test without a pass mark set in advance does not count.

10. Ads and creative standards

Ads are where the account says something no competitor can say.

How many ads, by campaign type

More ad groups or asset groups do not mean more ads served. Split only where a theme, category or budget needs its own lever.

Keyword research. Write down the phrases a buyer would type before opening any tool. Check CPC against margin: if the likely CPA at a 3% conversion rate is above break-even, leave it out. One theme per ad group, fewer and bigger ad groups, 1 to 3 long-tail broad keywords to start, exact match added only from search term audits.

Responsive search ads

Copy rules: say something competitors cannot; real numbers ("4.8 from 2,300 reviews") over claims ("We're #1"); no gimmicky capitals or symbols; final URLs live.

Search assets, every campaign: 6 sitelinks, 6 callouts, 2 structured snippet headers, 4 images, price assets where prices compete, promotion assets in every sale, verified business name and logo.

PMax asset groups: 15 headlines, 5 long headlines, 5 descriptions, 5+ landscape and square images, 3+ portrait, logos, and 3 of your own videos (horizontal, square, vertical). Replace every Low-rated asset monthly.

Video: hook in the first 2 seconds, product on screen by 5, brand said and shown, captions burned in, one call to action.

Landing pages. Before any campaign goes live: a clear message in the first screen; the page matches the ad (product, offer, price); reviews and proof on the page; delivery, returns and size guide visible without scrolling on mobile; one clear add-to-cart; under 2 seconds on mobile; the final URL returns 200 with no redirect chain.

Operating model

The analyst reads, diagnoses and prepares. The account lead approves and makes the change. The brand approves anything that moves its money or its brand. Every change goes in the change log before the next pass.

Who decides what

Everything new is built paused. Live changes (enable, pause, budget) are made by the account lead, then read back.

The change gates

A change goes ahead only when every answer is yes.

Meta budget step

1. New Audience results at 1-day click good enough on their own, trailing 7 days.
2. Stable across 1-day click, New Audience and incremental; 10+ purchases at 1-day click.
3. Shopify NC MER on target; Meta within tolerance of Shopify orders.
4. Inside the agreed monthly budget, and the last step has a change log row.

Then raise 10 to 20%, let it settle several days, re-check.

Google bid or target change

1. 30+ purchases in the last 30 days.
2. Weekly conversions within 20% for 4 weeks.
3. No bid or target change in 6 weeks.
4. No budget change in 2 weeks.

Google budget step: up 20% every 5 to 7 days, only once metrics have recovered from the last step.

Structure change, both channels: the problem is written down and proven in the data; it cannot be fixed inside what already exists; settings that cannot be edited later are decided; the brand has approved.

Cuts: any size, any day, on both channels.

The change log

One sheet for every brand and both channels, one row per change. Filter by brand.

Review on: 7 days for a Meta budget step; 14 days for Google budgets, bids, value rules and creative batches; 28 days for structure and attribution. The Thursday follow-up reads this column. A change with no row cannot pass the next gate, so no row means no next change.

Alerts and first moves

Automated rules notify only. Rules that pause or move budget on their own bypass the gate and the log.

Urgent message template

Quick heads-up: your [Meta / Google] ads stopped delivering at around [time] today. The cause is [declined card / account restriction / ads rejected for X]. [What we have done]. What we need from you: [update the card / complete verification / approve the revised ad]. We will confirm once ads are serving again.

Monitoring scripts (Google)

Every brand runs the same small set of read-only Google Ads scripts. Monitoring scripts run once from the agency's manager account across all brands; analysis scripts write to Sheets that stay in the brand's Drive folder.

Script rules

• Scripts that change the account (auto-pause, bid changes, label writes) stay off. They bypass the change gate and break the change log.
• Read every script before installing it. Scripts run with full account access: look for anything that writes, pauses, or sends data outside Google. Some popular third-party scripts share output Sheets as "anyone with the link can edit" or post account names to the author's own log; patch those lines out first.
• Authorise from a shared agency login, never a personal one, so scripts survive staff changes. Alerts go to a shared inbox and the account owner.
• Monthly: every script ran without errors, and every live brand appears in the output. A brand that unlinks from the manager account drops out of monitoring silently.
• On handover, install single-account versions inside the brand's own account so they keep running.

Scheduled reviews

Naming conventions

Names say what a line is for. Logs and reports go by ID, because names get edited.

Meta campaign: BRAND | PURPOSE | SCOPE | MARKET | BUDGET | PERIOD, e.g. ACME | BAU | All | GB | CBO | 2026 Q4.

• PURPOSE: BAU, PROMO_ plus the event (PROMO_BF26), Sales, Traffic, Leads.
• SCOPE: All, Mens, Womens, a category, or ASC when it is Advantage+ Sales.

Meta ad set: SEGMENT | AUDIENCE | MARKET | ATTRIBUTION | EXTRA, e.g. WW | Broad | GB | INC | Catalog. Attribution sits in the name because it cannot be edited later.

Meta ad: {PREFIX}_{MON}_{phase}_{category}_{gender}_{descriptor}_{format}, where PREFIX is whoever produced the creative and phase is bau, sale, offer or test. Revived ads add this year's actual return.

Google campaign: BRAND_MARKET_TYPE_FUNNEL_FOCUS_MODE, in capitals with underscores, e.g. ACME_UK_PMAX_MOF_NewCust_AO.

• TYPE: SRCH, SHOP, PMAX, DGEN, VID, DISP, AIMAX. FUNNEL: TOF, MOF, BOF, RMK.
• FOCUS: Brand, Comp, Generic, Hero, Sidekick, Zombie, Villain, NewCust, or a category.
• MODE: AO (always-on) or PROMO_ plus the event code.

Keep full-price selections clean: when picking "best ads", exclude names containing sale, promo, clearance, offer or flash, or the list fills with sale creative.

Rename inherited campaigns in stages, with the old name in the change log so year-on-year reports still join.

Intake and offboarding

Intake, week 1

• The brand owns every asset (ad accounts, Page, Instagram, dataset, catalogue, Merchant Center, GA4, payment methods); the agency is added as a partner or manager, never logging in as the brand
• Access: Google Ads via the manager account, Merchant Center admin, GA4 editor, GTM publish, Shopify staff account with reports
• Context file started: brand voice and banned words, AOV, margin by top products, 12-month target, customer persona, conversion action list, best and worst reviews
• Monthly budget per channel agreed in writing
• Brand added to the monitoring scripts and the change log
• Meta audience segments defined; custom audience terms accepted
• Brand approval for any AI tool or connector that reads the accounts, view-only by default

Offboarding

• Final report with the year-on-year view
• Single-account scripts installed in the brand's own account; Sheets transferred
• Change log rows and naming key exported and sent
• Partner and manager access removed on every platform; the brand keeps every asset

Checklists

Copy these into the build doc or the audit. Build paused, tick every box, then enable. A second person checks anything over £100 a day and any promo campaign.

Onboarding audit, week 1

Access and tracking come before any judgement on performance.

Tracking, both channels

• Meta: the live dataset confirmed, with browser and server events in the last 7 days, deduplicated
• Meta: Event Match Quality Purchase 8+, Add to Cart 8+; domain verified
• Google: one primary purchase conversion, counted once per order; no duplicate primary purchase actions
• Google: enhanced conversions on; consent mode v2 where banners run
• Both platforms' purchases against Shopify orders for the last 30 days, within 10%
• Opt-in consent markets listed; expect under-reporting there
• GA4: share of revenue in Unassigned; paid traffic not leaking into organic
• Merchant Center: disapprovals, price and stock mismatches, shipping and returns, promotions

Performance read, week 1 to 2

• Spend, revenue and MER for 30, 90 and 365 days, each against the prior period and last year, from Shopify
• Meta: audience segments breakdown. How much spend goes to Engaged and Existing Customers vs New Audience?
• Meta: attribution setting breakdown. How much of the reported result is 1-day click?
• Meta: count of campaigns and ad sets, and how many are Learning Limited
• Meta: age, gender, interest or lookalike inputs; standalone retargeting; default purchaser exclusions
• Meta: off-plan ACTIVE entities, such as boosted posts
• Google: brand vs non-brand split for every campaign type, PMax included
• Google: product-level ROAS for 90 days, sorted into hero, sidekick, zombie, villain
• Google: duplicate targeting; bid strategy and tROAS per campaign with the date last changed
• Google: lost impression share to budget and to rank; funnel coverage against the budget split by stage
• Ends in one traffic light action plan: red, amber, green

Pre-launch QA: Meta

Name and plan

• Names follow the convention
• The problem this build solves is written in the build doc
• No other ACTIVE line targets the same products, market and audience
• Change log row added, with a review date

Settings that cannot be edited later

• Objective Sales; optimisation event a standard event (Purchase unless agreed)
• The right pixel or dataset
• Attribution Incremental (Standard 7-day click only where agreed)
• Catalogue ads: promoted product set correct; filter excludes the other gender and sale-only items

Targeting and placements

• Advantage+ audience on; no age, gender, interest or lookalike inputs; controls only for legal reasons
• Any "acquire only new customers" pre-filled exclusions checked by hand
• Market correct; geo expansion read back as off
• Placements Advantage+; no placement exclusions; 9:16 present so new placements are covered

Budget

• Daily budget matches the plan and the agreed monthly budget
• 1 to 3 ads per new batch (each up to 10 media); no separate testing ad set
• Promo: its own campaign, an end date, scheduled increases dated per the plan
• Account spending limit and backup payment method set

Tracking

• Dataset shows browser and server events, deduplicated, in the last 7 days
• UTMs in the URL parameters field; test URL resolves with the right market and currency
• Audience segments defined in Advertising Settings

Ads and creative

• Every ad has 9:16 and 4:5 or 1:1; a static and a video in the batch
• Five primary texts and five headlines where the format allows
• Enhancements set as planned; site links, promotions and product cards off unless wanted
• Batch checked against the source folder: nothing dropped
• Offer, price, code and dates match the site; creator codes still redeeming
• No sale creative in full-price ad sets
• Correct Page and Instagram identity; partnership permissions granted
• UK ASA and Meta policy; advertiser verification set for EU delivery

Pre-launch QA: Google

Name and plan

• Name follows the convention
• The lever it exists for is written in the build doc
• No duplicate targeting: no other enabled campaign targets the same products, locations and audience with different bidding
• Change log row added, with a review date

Settings

• Locations correct, location option set to Presence
• Language matches the ads; one language per campaign
• Search: Search Partners and Display Network off unless chosen
• Ad schedule and dates set; promo campaigns have an end date
• PMax: brand exclusion on, new customers only where planned, URL expansion and asset automation set as planned
• AI Max off unless this is the AI Max test

Budget and bidding

• Daily budget matches the plan and the monthly budget
• Starting bid strategy right for the campaign type: new Search or Shopping on Max Clicks or manual CPC; new PMax on Max Conversion Value with no tROAS
• Shopping budget at least 10x expected CPC; PMax at least 30x cost per conversion
• Promo campaigns on their own budget, never shared with always-on

Tracking and goals

• Campaign goal is Purchase only
• Auto-tagging on; any tracking template tested
• Purchase tag verified against Shopify in the last 30 days

Targeting and negatives

• Shared negative lists applied: competitors, not our product, jobs and free
• Brand and non-brand campaigns negate each other where planned
• Listing groups include only the intended products; tier label filter correct
• Audience signals on every PMax asset group

Ads, assets and pages

• Asset counts meet the standards; ad strength Good or better
• Every final URL returns 200 with no redirect chain, in the right market and currency
• Landing page checks passed
• Offer, price and dates match the site and Merchant Center promotions
• Nothing disapproved or limited after submit
• Brand is on the monitoring scripts

After launch, both channels

Judge nothing on the first three days. Meta serves warm audiences first, and Google is still learning, so early results flatter or mislead.