YouTube ads for ecommerce: the minimum viable setup

We get the same conversation with ecommerce brands every few months. They tried YouTube, ran something for a few weeks, watched the last-click ROAS come back at half what they were getting on Search, and concluded YouTube does not work for performance. The conclusion is mostly right about what they ran and mostly wrong about the channel.

The bigger story is what that conclusion reveals about how Google has restructured its inventory and how most brands’ measurement has not kept up. Performance Max routes spend through YouTube and Display as part of its cross-channel model. Demand Gen serves on YouTube, Discover, and Gmail with creative-led formats. AI Max extends the same logic into Search. The proportion of paid search budget flowing through video and visual placements keeps rising, and Google’s reps are actively pushing brands toward YouTube at every account review. Creative has become the primary performance lever on the placements Google is now investing in, not one input among many, and the brands still treating Search and Shopping as the whole picture are reading degraded results from accounts that have started shifting underneath them.

What follows covers what that picture looks like on YouTube specifically: why creative is the lever, why most measurement is wrong, and what the minimum viable setup for an ecommerce brand testing the channel actually looks like. It is built on test data from accounts where YouTube produced material revenue lift, and from accounts where it did not.

Why creative is the lever now

A short orientation, because the rest only makes sense once this part is clear.

Google has spent the past three years steadily moving its ad inventory toward video, visual, and AI-driven placement. Performance Max collapsed multiple campaign types into one and routes spend across YouTube and Display by default. Demand Gen launched as a creative-led format across YouTube, Discover, and Gmail. AI Max for Search lets the algorithm match queries against assets it generates from your landing pages. Every one of these moves shifts the operator’s leverage away from keyword targeting, bid management, and structural campaign decisions, and toward the creative that actually populates the placements the algorithm is now choosing.

This is the same shift that has already played out on Meta, where creative volume and diversity are the primary performance variables and have been for some time. On Google, the shift has been slower and less visible, because the bulk of spend has historically run through Search and Shopping placements where the ad surface is text and product imagery, not video. That is changing. The proportion of Google budget flowing through creative-led placements is rising, and brands that build creative production capacity for YouTube and Demand Gen now will be positioned for that shift.

The implication for YouTube specifically is that the campaign is, mechanically, a creative campaign in the way a Search campaign is mechanically a keyword campaign. The setup decisions below (budget, audience, bidding) are how you give creative the conditions to do its job. None of them substitute for the creative work itself.

Creative: minimum viable production

The working minimum for an ecommerce brand testing YouTube is one creative concept rendered into multiple formats. Two pieces of creative in different lengths or with different opening hooks gives you enough to start. The absolute floor is one video, one concept, in different formats. You can run that for two weeks and refresh, or four weeks if you have the right frequency cap and volume.

This is materially less than a Meta production cycle requires, and the reason matters. At top of funnel on YouTube, diverse creative tends to matter more than sheer volume. The high-volume model that works on Meta, where feeding the algorithm with large quantities of creative variants is the proven approach, does not transfer directly. YouTube responds better to fewer, stronger, more deliberate assets, with creative refreshed every two to four weeks depending on fatigue.

What those two concepts should be depends on the brand and the campaign objective. For a brand testing YouTube cold, a common starting point is one awareness piece that introduces who the brand is and one product-led concept that shows the catalogue or hero category in action. But the important thing is having two distinct variants for testing, not one concept running unchallenged.

Each concept needs format variants. Skippable in-stream ads in the 15 to 30 second range. Six-second bumpers for reach top-up and frequency support. Vertical for Shorts where it applies. The format mix is where most brands fall short.

We ran a brand awareness Christmas test on a UK footwear account in late 2024 covering devices, demographics, audience targeting, and creative length. Twenty-second creative outperformed six-second creative on the metrics that mattered for the campaign. TV placements had CPMs too high to be efficient. Third-party and custom audiences outperformed first-party retargeting at top of funnel. We exclude TV placements by default in subsequent setups for that account, and we treat the 6s versus 20s finding as directional, not universal.

The broader principle on length: 6-second bumpers have a real role in reach top-up and frequency support, and longer formats (15s, 20s, 30s) tend to do the work of building awareness and recall. Running 6-second creative as your only YouTube format is a common mistake, and it is often what produces the “YouTube does not work” conclusion brands arrive at.

The same creative that works on Meta at brand-awareness level can typically work on YouTube. You do not need a parallel production cycle. Vertical Shorts placements behave more like TikTok and benefit from native vertical formatting, but the broader top-of-funnel asset library can travel.

The supporting decisions: budget, audience, bidding

Creative is the lever. The other setup decisions are how you give it the conditions to work, and getting them wrong is the easiest way to make good creative look like bad creative in the dashboard.

Budget. The minimum requirement is £100 per day for the UK market. In the US that figure needs to scale with audience size: if you are targeting the area around a few major cities with a total population similar to the UK, £100 per day still works as a floor; if you are targeting the whole US, budgets need to increase significantly. Spread the budget too thin and frequency caps prevent the campaign from producing a result you can read either way. Geographic targeting and budget have to be calibrated together; you cannot leave one of them on default. For brands without ongoing budget, concentrated bursts of two to four weeks in a defined geography work better than diluted always-on at a thinner rate. This is the approach we used with The Folio Society in the US: state-by-state bursts rather than national always-on, with each burst validating before the next one started.

Audience. Three audience types are worth deploying together, not picking one. Custom audiences built from search behaviour and competitor activity tend to be the best-performing for top-of-funnel YouTube on ecommerce; the footwear Christmas test specifically found custom audiences outperformed broader options. Third-party affinity and in-market audiences vary in quality, but a curated subset (in-market for your specific product category, affinity audiences aligned with your brand positioning) typically produces reasonable cold-audience reach. Lookalikes built from your first-party data give the algorithm a strong direction by anchoring it on your existing high-value customers. For Demand Gen specifically, lookalikes of relevant first-party segments are the recommended approach: use audiences of dress purchasers to inform targeting for a Summer Dress launch campaign, not generic affinity audiences. Splitting audience types into separate ad groups lets you control CPM targets per audience and read which audiences are actually working after the campaign closes.

Bidding. For awareness-objective YouTube activity, target CPM is the default. It gives the campaign a clear cost ceiling per thousand impressions and aligns with the awareness objective directly. View-focused bidding (target CPV) tends to over-optimise for cheap views at the expense of attention quality, and the cheap-view trap is a recognisable pattern in poorly-performing YouTube campaigns. For a deeper look at how smart bidding works and when to override it, we have covered that separately. For mid-funnel and consideration activity, Demand Gen with conversion-focused bidding is the right configuration. Target CPA or maximise conversions, with micro-conversions like add-to-cart enabled alongside purchase, gives the platform enough conversion volume to bid efficiently while still optimising toward the business outcome. For Shorts placements, we have found Demand Gen run as a standalone campaign tends to give better results than Shorts served through PMax.

The measurement problem

This is the part of the picture most brands have got most wrong, and it is the part where the gap between Google’s restructuring and most operators’ decision-making is widest.

The methods operators reach for first are the wrong instruments for the channel. Last-click ROAS over a fortnight tells you almost nothing useful about whether YouTube is working. The conversions YouTube drives mostly happen later, through other channels, and the longer time windows reveal them while the short reporting windows miss them. A YouTube campaign that is generating real demand will show up first as rising brand searches, then as efficiency improvements in your Search and Shopping campaigns over the following weeks, then as direct conversions captured in your own back-end data. None of those will appear in the YouTube campaign’s last-click ROAS, because last-click attribution is structurally incapable of crediting them.

The clean answer is geo-tests. Run YouTube in one region, hold it back in another comparable region, and measure the revenue difference in your own back-end data over a meaningful window.

With The Folio Society, we ranked every US state by purchases per 500,000 people. Not total order volume: per-capita propensity to buy. That completely changes which states you prioritise. California and New York dropped out. States like Vermont, Maine and Colorado came in. Then we concentrated a 10-to-14-day burst behind a bespoke brand film in the highest-intent state and measured the lift before moving to the next. Revenue grew 58% in the targeted state and 45% in surrounding states, against 17% growth across the rest of the country. Brand searches rose. Bottom-of-funnel purchases tracked in the states we hit. Each burst validated before we spent another pound.

The geo-test isolates the channel-level effect at the business level, which is the cleanest measurement available without third-party tools.

Brand Lift Studies and Conversion Lift Studies, native within Google Ads, produce platform-level evidence of YouTube’s effect on awareness and conversion respectively. They have a minimum spend threshold to run reliably but they are useful when the budget supports them.

The search-lift effect is a free supplementary signal. Watch brand search volume in Google Search Console during and after YouTube bursts. Category-level search volume in your own campaigns will show the same effect. If your YouTube activity is working, both should rise meaningfully, and the lift is consistent enough across our managed accounts that search amplification is an expected effect of YouTube activity, not a side benefit.

For brands with the budget for third-party measurement, Phosphor or equivalent multi-touch attribution tools produce a multi-channel view that reflects YouTube’s actual contribution more accurately than Google’s last-click default.

The mistake to avoid is judging YouTube on last-click ROAS without any of the above in place. The result will look bad, the campaign will get killed, and the brand will join the queue of brands telling us six months later that YouTube does not work for them.

What 90 days actually looks like

If the setup is right, the picture in the first 90 days follows a recognisable pattern. Last-click ROAS will look weak, often half of Search or worse, and that is expected, not a verdict. Brand search volume should rise during and after burst periods. Search and Shopping campaigns should see efficiency improvements as YouTube generates demand that converts through other channels. Geo-test results, if they were designed in from the start, become the headline answer on whether YouTube is incremental for your business. Creative will need refreshing at least once; a campaign flat-lining by week three is creative fatigue, not channel failure.

What usually separates brands that conclude YouTube works from brands that conclude it does not is whether the measurement was set up to match the channel before the campaign launched. Brands that defaulted to the dashboard they already had open ended up reading the wrong instrument.

What this means for the next two years

Google’s strategic emphasis on YouTube is not subtle. The platform is investing in YouTube as a performance channel, not just an awareness one, Demand Gen and PMax are increasingly routing creative-led spend through YouTube placements, and AI Max for Search is extending the same algorithmic logic across the rest of the platform. For brands that have been running paid Search, Shopping, and Meta as the core stack, YouTube is the most likely additional channel to compound performance through 2026 and 2027.

The setup work is real but it is teachable. The test methodology is robust enough to give a clear answer at relatively modest spend. The harder shift is recognising that the measurement framework most brands have used for the past decade was calibrated for a paid search world that is becoming a paid video world, and that running 2026 campaigns through 2018 measurement produces predictable misreadings.

Brands that get the creative work and the measurement work right have a structural advantage over the rest of the field for as long as YouTube remains undervalued relative to its actual contribution to performance, and that window is closing faster than most ecommerce teams realise.

This is also where the loop between creative production and live performance data matters most. YouTube creative that performs well feeds insight back into the next round of production: which concepts drove brand lift, which formats held attention, which audience segments responded. That insight sharpens the next brief, and the cycle compounds. Brands running YouTube as a disconnected burst miss the compounding entirely. Brands running it inside a connected creative-to-performance loop, where what is learned from YouTube also informs what runs on Meta and Search, get more out of every channel as a result.

FAQ