Hot PMax questions: best campaign setup, reporting, incrementality & more

So, I’ve taken those questions, and I’m answering most of them for you right now.

And as for the meta-question, “Why does PMax raise so many questions for advertisers and partners?”, I’ll answer that one last.

Note: Some questions have been lightly edited for readability. The answers are accurate as of October 2023.

How to choose the right
campaign type

How do you choose PMax or Standard Shopping?

In the bluntest of terms, I’d state it like this: If you want to pursue revenue in the most aggressive way possible and raw by-the-numbers performance is your top priority, PMax is probably for you. If you want or need specific reporting and controls, or a more customized implementation, Standard Shopping might be the way.

Having said that, there are about a thousand ways of answering this and you need to test if you are not sure. You can read more on the topic in detail here.

Does running a Search campaign still make sense next to Performance Max?

Yes. PMax is a keywordless technology, and its Search capabilities are most comparable to DSAs. There is still space in Google Ads for well-organized keyword campaigns, brand search campaigns, RSAs, etc. Even Google describes PMax and Search and the new “Ads Power Pairing.”

What is the best strategy for B2B? PMax or Standard Shopping?

There is no reason why you can’t succeed in B2B with PMax. I’d simply recommend adding as many signals as you can to help inform the algorithm on what a B2B shopper looks like compared to a B2C shopper. You might miss some controls you had with Standard Shopping such as negative keywords, or you might have additional topic exclusions or placement exclusions in mind. These can typically be handled through a Google rep, you’ll just have to be pushy.

Would you trust PMax to pick up on short-term signals, e.g. Black Friday, or would you push key product lines through Standard Shopping?

I do trust Google’s technology to understand clear, scheduled seasonal events like Black Friday. That said, it’s better to be safe than sorry. I would support Google with seasonality adjustments. Standard Shopping offers additional promotional controls such as scheduling, priorities, modifiers, and negative keywords that could of course be beneficial to you. But assuming whichever option you’re currently using is working, I would adapt that approach for the holiday rather than changing approaches entirely.

Here are a few tips about seasonality adjustments:

• intended for short periods of 1 to 7 days
• Google says they will automatically ramp down with no need for negative adjustments – but keep an eye on this
• your adjustment should be based on past data:
  • how did comparable periods or promos lift conv. rate?
  • how much of that lift was noise? (calculate your CVR’s standard deviation and subtract 1 standard deviation from the lift)
  • did this effect vary per brand or category?
• but don’t succumb to analysis paralysis – your adjustment doesn’t need to be perfect

PMax feed-only vs
full-build approach

What are your thoughts of doing a feed-only PMax campaign and then using Demand Gen for all the rest of the assets?

I think this could become a popular option. Here are three reasons why:

1. Although Google wants the market to embrace full-funnel campaigns, ultimately advertisers are rarely enthusiastic about this. Finding ways to segment budget and reporting for different funnel stages remains attractive to many.

2. The display portion of PMax is too spammy for some advertisers. Running feed-only PMax and Demand Gen in parallel lets you cover premium ad inventory while leaving behind less attractive placements.

3. Full-build PMax and Demand Gen have some overlapping scope, where it’s not 100% clear how priority is solved (presumably ad rank). For example, PMax and Demand Gen can both serve YouTube & YouTube Shorts, Discovery, and Gmail. Using feed-only PMax could help reduce those overlapping scopes.

So, it looks promising, but needs to be tested. Thanks for the question.

What are the best practices to be sure that a feed-only PMax campaign brings the most incremental sales?

It is hard both to verify and to enforce PMax in terms of incrementality. This is also true of feed-only campaigns since they will still include warm traffic in the form of dynamic remarketing. Google’s options here are:

• New Customer Value mode: encourage the algorithm to bid for incremental traffic by artificially boosting the conversion value of new customers
• New Customer Only mode: force the algorithm to bid exclusively for incremental traffic using your customer lists and Google’s own matching system

I don’t recommend NCV mode for this because it will affect your reporting and will still market existing customers anyway. NCO mode is more to the point, but it will only be as successful as the lists you can provide (Google does some lifting as well, but they need your lists). NCO also has a tendency to completely kill volume.

You can take other measures like excluding brand traffic, boosting or suppressing products based on their propensity to acquire new customers, etc. But it’s a tough spot. You can run normal and NCO campaigns in parallel, so that you at least have the clarity of “this budget is incremental, this other budget is muddy”.

I have a PMax feed-only campaign that has a £500/day budget that just continually spends £100/day. Is there anything I can do to get it to spend more?

A tricky question that depends a lot on your website, products, brand(s), and account structure.

Long-term answer:

Performance Max is fundamentally a demand capture campaign. Yes, there are potentially branding and demand generation placements in there, but they will typically account for a minority of spend and will often be in service of remarketing (more capture!) rather than new customer acquisition.

You might need to support PMax by generating more demand first. Otherwise PMax has no demand to capture.

Short-term answer:

• change (lower) your ROAS target
• feed hygiene and optimization
• adding audience signals
• increasing or decreasing segmentation
• on-site improvements

Is it still possible to set up feed-only PMax?

As of early October 2023 – yes. Despite rumors to the contrary, it is still possible. The workflow can be tricky (since it’s a bit of a “hack”), and it’s prone to user error, which likely accounts for reports that it doesn’t work anymore. It’s also possible that Google will close this loophole in the future.

I see both positive and negative signals, for example:

• Google issued bug fixes in Ads UI, API, and Editor to support feed-only creation
• Google is trying to serve more placements based on product feed images 

Time will tell!

Audience signals in PMax

How can I test audience signals in PMax? Is that a thing?

Yes, it’s a thing, and a great question btw – audience signals are a key strategic and tactical element of Performance Max campaigns, yet the performance impact is hard to know. A post-analysis or sequential test would be the simplest way, but is subject to seasonality and other variables. Geo split testing is therefore a better approach.
Bear in mind that signals in PMax are start signals rather than targeting. In an ideal case, Google will learn from your signal to improve performance and expand on it to find more (new) customers. In a bad case, Google might misinterpret what was valuable to you about that signal, or might “drift” over time. It’s worth reviewing and refreshing your signals periodically.

When setting up PMax campaigns, should we be adding audience signals from the get-go or should we not apply audience signals so PMax can work its magic and learn?

To me, nothing speaks urgently against waiting – it arguably comes with the benefit of getting a feel for PMax’s baseline performance as you implied. However, personally I feel there are enough unknowns already at work, where I’d like to influence PMax from the get-go. I can’t support the claim with data, but I’d expect a shorter and more efficient learning period.

See how we take back control of
Performance Max campaigns

Learn more

Product segmentation for PMax

Would you recommend segmenting products by performance in different PMax campaigns or a single PMax with multiple asset groups (e.g. top sellers, low performers, opportunities, low volume SKUs)

You need to create multiple campaigns to do this. This way, the product segments are receiving different budget and efficiency goals, which will control how they are paced. Asset groups, in turn, are used for conveying thematic information and audience signals. Read more here.

Reporting with PMax

Do you have any recommendations on reporting for PMax. There doesn’t seem to be in-depth reporting for PMax currently. Do you know if/when more updates are coming to Google?

It’s true, reporting is different – and often limited – for PMax. However, Google is improving this over time and (in my opinion) it’s a better situation than with Smart Shopping.

Here are a few of my favorite reporting options you might not know about.

Search Term Insights – this report is not new but has been enriched with a detailed view that offers historical data, selectable time ranges, and download options. You can use this to better understand your key search themes and run calculations such as brand revenue share. However, the report is woefully lacking cost data as well as how the branded impressions, clicks, and conversions are changing over time. But this can be solved, and we did it with our PMax Brand Analyzer script. The script shows the evolution over time and visualizes the volume and share of brand vs non-brand traffic in your campaigns. Tip: Schedule the script to run daily for consistent monitoring.

Unofficial placement reports – it’s possible to estimate metrics for placement types including video, display, and search using scripts. I’d say there are “various options” for this, but let’s face it, we’re all talking about the incredible script built by Mike Rhodes based on foundational research from Tobias Hink. Kudos & thanks to those gentlemen.

from Auction Insights:

Retail Category – check your competitors up to 5 levels deep to understand how competition varies across categories.

Campaign Type – see how Standard Shopping and PMax handle competition differently

Impression Share – while no overall impression share was available for PMax, you could see it isolated for search and shopping auctions. Update: PMax impression share is now available as of September 11, 2024.

You can also try our free PMax audit tool, which provides a general overview of your account, as well as information on budget efficiency and under/over delivery.

What is your take on adding micro conversions (add to cart, begin checkout, etc.) to PMax campaigns in situations where there are less than 30 conversions a month?

In general you need to look for micro-conversions that correlate as tightly as possible with your primary conversion goal. Otherwise, if the association is too loose, you run the risk that you will train the algorithm on the wrong behaviors, wrong audience attributes, etc. Ideally it’s a crutch to support the algorithm and push you through a “local maximum” to the point where you have a stable, sufficient conversion volume.

How to set the right ROAS for PMax if you have no specification from the client? Simply leave them with Max Conversion Value?

Before I continue, it’s important to understand if your client has a red-line ROAS that may not be crossed – for the safety of their business. There’s a difference between not having a desired target and not having a necessary target (a safeguard). Out of a sense of fiduciary responsibility you need to be clear with them on that. Force the conversation, help them think about it.

Moving on: this is a very tricky question, because ROAS is a metric that gets pulled in two opposite directions. On the one hand, ROAS is an efficiency metric commonly used as a proxy for profitability. On the other hand, in the post-bidding era epitomized by Performance Max, ROAS is also a primary lever for steering campaigns. It’s a huge red flag that one metric should do double-duty as a rather unchangeable strategic business objective and a highly-changeable tactical pacing tool.

So how to set a target? I would look at historical data, or peers in terms of vertical, budget, etc. For reference, a median ROAS in Performance Max is 5. That does NOT mean it’s the right ROAS for your client, but it could be a starting point in the absence of other facts.

Regarding Max Conversion Value, you might consider testing it to get a sense of the “naturally occurring” ROAS of their campaigns. Many fear that if they switch on Max Conv. Val, then ROAS will just drop to zero. Not true. Google has a hidden backend ROAS, supposedly set at 1. From my observations, most campaigns will not scale to that level though.

I hope this helps.

Tips for measuring PMax incrementality

How would you measure the incrementality of PMax campaigns?

There are multiple approaches here depending on your resources and level of commitment.

A holdout test would be the most conventional choice and should offer a clear answer. Geographic splits can be a simple but effective way of doing this, and matched market tests offer a more valid comparison than random splits. The drawbacks are that these tests take time, require supervision from an analyst or data scientist, and have an opportunity cost in the form of lost revenue (assuming the channel turned out to be incremental). The good news is that there’s a great package from Google that can help you here: Google Matched Markets.

Marketing Mix Modelling (Bayesian) is a “less invasive” option that can offer you insights faster and without the opportunity cost of testing. Moreover, it’s designed to test your entire marketing portfolio and not just one channel, so if you have large investments across multiple channels, it can be more accurate and useful. It basically runs lots and lots of simulations to arrive at reasonable conclusions. You might consider Meta’s Robyn package, Google’s Lightweight MMM package, or a third-party tool.

Lastly, you can consider your Marketing Efficiency Ratio, which holds your total revenue against your total ad costs. This can offer you high-level, directional guidance. Personally I am not a fan of MER and I’ve equated it to reading tea leaves, but there are a lot of smart people out there using it who would disagree, so I’m including it for the sake of completeness.

There are also lift studies, post-test analysis, and other possibilities. Whatever the case, I think it’s important to look beyond the channel reports and at your backend numbers, tests, or other trusted “sources of truth” to verify what the channel is telling you.

Do you recommend turning off the New Customer Acquisition option for BFCM?

If we’re talking about New Customer Value mode, just leave it on, otherwise you’re going to really confuse your reporting. I don’t recommend NCV mode.

If we’re talking about New Customer Only mode, then this decision heavily depends on your promotional strategy. If you’re running promos and you turn off NCO bidding, then you’re more likely remarketing your existing client base with discounts, thus potentially harming CLV (measured as profit, not revenue). It’s up to you if that’s productive or not – maybe you view it as securing market share and it’s supported by the margins.

Personally I would consider running NCO bidding and normal PMax campaigns in parallel so that these two different goals can receive appropriate budgets and efficiency targets – and for clean reporting in general.

Strategies for testing PMax campaigns

What do you think about A/B testing with PMax? How much time do you think is necessary to validate the results taking into account the learning time of smart bidding strategies?

A/B testing for Performance Max is a great idea! It just needs to be taken seriously: a serious test requires time & effort and could potentially limit revenue or present other opportunity costs. It requires commitment.

Regarding the learning period, you cannot test a fresh-baked campaign or account. Your control campaign needs to be rather mature, i.e. out of the learning phase, with ideally 2 or 3 months of baseline performance. This is necessary for the experiment design (for example, creating a fair geo split). Then your treatment campaign will also need to be out of the learning period, but needn’t have any history longer than that.

The test duration depends on the volume in the campaigns, and the magnitude of difference observed. To reach statistical significance on a lower-volume campaign, you’ll need to run the test longer, unless there is an extremely clear winner. Typically tests are planned for 30 to 60 days, but can be shorter depending on results.

It’s important to be clear on your decision criteria, and your reasoning for any allowable changes to those criteria mid- or post-test. Otherwise you risk “moving the goalposts”.

Check out this case study to see how we achieved an incremental uplift of over 13% by A/B testing our approach against a continuously active Google-powered campaign.

Why do you prefer geo split for testing over splitting the feed into equal parts?

It’s far easier to arrive at a statistically valid split using geography, especially with the help of tools like Google Matched Markets. There are many pitfalls of feed-based splits:

• hard to select products/categories for homogeneous performance
• hidden dependencies (complementary products, add-ons, etc.)
• interference from trends and seasonality
• queries matching against multiple products

The ideal split would actually be a user split, not a geo split. However, only Google themselves can deliver that – indeed they’re working on it now. I would still choose a geo split though, because a user-based split test will be hosted in Google’s environment where you won’t have the same level of flexibility and transparency.