A client's AI Max campaign started pulling traffic from a competitor they'd deliberately avoided for years. Performance metrics looked strong. The algorithm was doing exactly what it was designed to do: find more conversions. The problem? That competitor was three times their size, and the advertiser had spent a decade staying out of a bidding war they couldn't win.
This is the scenario Mike Ryan, head of e-commerce insights at Smarter Ecommerce, described on the PPC Live podcast this week. It captures something CFOs and CMOs need to internalize before the next budget cycle: Google's AI can optimize toward a goal, but it cannot understand your business context. The algorithm saw conversions. The advertiser saw a strategic landmine.
The Conversion Data Threshold Nobody Wants to Model
Ryan's research points to a minimum of 30 monthly conversions per campaign for Smart Bidding to work consistently, with 60 or more being the realistic target for stable optimization. That number has implications most account structures ignore.
Consider a mid-market B2B account with 12 product-line campaigns, each averaging 10 conversions per month. Every campaign is technically running Smart Bidding. None of them have enough signal density for the algorithm to exit learning phase reliably. The result is a portfolio of campaigns that look automated but behave erratically, with bid decisions based on insufficient data and performance variance that makes forecasting nearly impossible.
Search Engine Land's analysis of campaign structure confirms the pattern: over-segmented accounts starve individual campaigns of the conversion volume Smart Bidding needs. The fix is consolidation, but consolidation requires rethinking how you report to finance.
CFO Structures vs. Algorithm Requirements
Here's where the tension gets uncomfortable. Finance teams often want campaign structures that map cleanly to P&L lines, regional budgets, or product categories. That organizational logic creates exactly the fragmentation that degrades algorithmic performance.
Ryan's point is that advertisers have moved from tactical segmentation (splitting by match type, device, or geography) toward strategic segmentation, where every additional campaign or split must justify its existence with a clear business reason. The question isn't "can we create a separate campaign for this product line?" It's "does this split give us enough conversion volume to maintain bidding stability, and does the strategic value outweigh the signal dilution?"
For a CMO presenting to the board, this means the campaign structure conversation needs to happen before budget allocation, not after. If your CFO wants 15 separate campaigns to match their reporting taxonomy, you need to model what that does to conversion density per campaign and present the trade-off explicitly.
AI Max: The Controls That Actually Exist
Google's announcement that AI Max is moving out of beta with automatic migration starting September 2026 for campaigns using Dynamic Search Ads, automatically created assets, and campaign-level broad match changes the calculus. This isn't optional experimentation anymore; it's infrastructure.
The good news, as Ryan emphasizes, is that AI Max has guardrails. Advertisers can control expansion through negative keywords, brand inclusions and exclusions, and Search Partner Network settings. The reporting includes AI Max match type and match source, which reveals exactly where query expansion is coming from.
The bad news is that most advertisers don't use these controls proactively. They enable AI Max, see headline performance improve, and assume everything is working. Ryan's "trust but verify" approach means checking search terms from day one, not waiting for a quarterly review to discover the algorithm has been bidding on competitor terms you explicitly avoided.
Ryan's own analysis at SMEC found that AI Max tends to favor competitor terms, Search Partner Network, and cross-language queries. None of these are inherently bad, but all of them require business context the algorithm doesn't have. If your brand strategy involves staying out of competitor auctions, you need to exclude those terms before AI Max finds them for you.
The Viral Claim Problem
Ryan shared something on the podcast that deserves attention beyond the PPC community. He initially concluded from several campaigns that AI Max favored Search Partner Network and posted the finding on LinkedIn. More data later showed that wasn't generally true.
The correction never gets the reach of the original claim. Ryan discussed the responsibility that comes with having an industry audience, particularly when advertisers may use observations to make real campaign decisions. For CMOs and their teams, this is a reminder that early findings from any new Google feature should be treated as hypotheses, not conclusions. Sample sizes matter. Holdout periods matter. The temptation to share a hot take before the data stabilizes is real, and the cost of acting on premature conclusions is measured in budget.
What This Means for Your Next Forecast
If you're building a 2027 marketing plan, here's the model:
First, audit your campaign structure against the 30-conversion minimum. Any campaign below that threshold is a candidate for consolidation or a different bidding strategy. Present this to finance as a signal-density requirement, not a preference.
Second, map your AI Max controls before the September migration. Negative keyword lists, brand exclusions, and Search Partner Network settings should be configured proactively, not reactively. Document the business logic behind each exclusion so the next person who inherits the account understands why competitor terms are blocked.
Third, build a monitoring cadence into your operating rhythm. Search term reports should be reviewed weekly during the first 30 days of any AI Max activation, then monthly thereafter. If you're seeing query expansion into categories you didn't intend, the controls exist to stop it.
Fourth, treat Google's performance claims as directional, not guaranteed. Independent analysis from PPC Live found that while median revenue lift from AI Max was close to Google's 14% claim, median CPA increased 16%, and only 22% of campaigns came close to their original ROAS targets. That's not a reason to avoid AI Max; it's a reason to set expectations correctly and build sensitivity analysis into your forecast.
The algorithm is getting better at finding conversions. It is not getting better at understanding why you might not want certain conversions. That gap is where strategy lives, and it's not something you can automate away.
