Why Your Performance Max Campaign Isn't Working (And What to Fix First)

Performance Max is the campaign type every account manager loves to blame and every client loves to distrust. When it works, it feels like magic. When it doesn't, it feels like Google is just quietly setting your budget on fire.

In the audits we've run over the past year, the same five issues come up again and again — and none of them are about the bidding algorithm. If your PMax campaign is underperforming, start here before touching anything else.

1. Your conversion data is lying to you

Performance Max is a machine-learning campaign. That means it is entirely as good as the conversion signal you feed it. If your conversion action is counting every "add to cart" as a purchase, or firing twice on refreshes, or missing 30% of real orders because of a GTM race condition, PMax will optimize towards noise.

Before you look at anything else, open Google Ads → Goals → Conversions and check:

• Is the primary conversion the actual business outcome (purchase, qualified lead), not a surrogate (page view, scroll depth)?
• Is Enhanced Conversions turned on and showing green "Recording" status?
• Does the conversion count for the last 30 days roughly match your Shopify / CRM / Stripe number? Within 10% is fine. 40% off means something is broken.

2. You're using one big asset group for everything

This is the single most common PMax mistake. A DTC clothing brand puts t-shirts, dresses, outerwear, and accessories into one asset group, adds a handful of generic headlines, and expects Google to sort it out. It doesn't.

PMax asset groups work best when each one has a coherent audience signal, coherent creative, and coherent intent. Split asset groups by:

• Product category — a dedicated group per major category if you run Shopping feeds
• Margin tier — high-margin products on their own so you can set a different target ROAS
• Funnel intent — separate prospecting and re-engagement groups where the creative can actually be different

3. You haven't told the algorithm what "good" looks like

Audience signals in PMax aren't a hard targeting setting — they're a hint. Google will still serve outside your signal, especially once it thinks it has figured out your customer. But the hint matters enormously during the learning phase.

At minimum, every asset group should have:

1. A custom segment built from search terms your real customers use
2. Your first-party customer list (uploaded via customer match, with Enhanced Conversions for identity matching)
3. One or two interest / in-market segments that match the category

If your asset group has zero audience signals, you're asking a million-dollar ML system to guess. It will — badly, at first.

4. You can't see where the budget is going

The biggest legitimate complaint about PMax is how opaque the reporting is. Google has quietly shipped more visibility in the last 12 months, but you have to go find it.

Three reports we open on every audit:

• Search term insights — Insights tab → Search terms. Tells you what queries are driving your PMax spend. If it's all brand, your incrementality is near zero.
• Asset group report — Campaigns → Asset groups → check conversions and cost per group. The "one group does 90% of the work" pattern is extremely common and worth fixing.
• Channel report — Scripts or Looker Studio. Tells you the Search / Shopping / YouTube / Display split. Display-heavy PMax campaigns usually have inflated view-through conversions and poor incrementality.

5. Your tROAS is fighting against your data

Target ROAS is a constraint, not a goal. If you set tROAS at 400% and your actual blended ROAS is 220%, PMax will shrink your budget spend to death trying to hit an impossible target, and you'll conclude PMax "doesn't work".

Good tROAS hygiene:

• Let a new PMax campaign run at Maximize Conversions or Maximize Conversion Value (no target) for at least 30 days before layering a target
• When you add a target, set it 15–20% below your current trailing 30-day ROAS, not above it
• Move tROAS in 10% steps, not 30% jumps

When to give up on PMax

Sometimes PMax genuinely isn't the right campaign. If you're a B2B lead-gen business with fewer than 20 conversions a month, standard Search will almost always outperform — you don't have enough conversion volume to train an ML campaign. If your product has a 6-month sales cycle, PMax can't wait that long for the feedback loop.

But for most e-commerce brands doing $30K+ a month, PMax should be your workhorse. When it isn't, the problem is almost always in this list — not in the algorithm.