Automated campaigns take 12% of US ad spend, and Google's targets no longer beat themselves

Automated and AI-driven campaign types will take roughly 12% of United States advertising spend this year, against 2% in 2023, on estimates from the research consultancy Madison & Wall published by Digiday on September 10. Madison & Wall sizes this year's US market at $479 billion excluding political advertising, and expects the automated slice to reach $158 billion, or 27% of the market, by 2030. The Interactive Advertising Bureau lifted its own US growth estimate for the year to 12.3%, from 9.5%.

Adoption is already past the experimental stage. Digiday reports that almost a third of Google's search budget now flows through AI Max or Performance Max.

We don't think automation in media stops at 20% or 30% — we think it gets to 90%

That is John Dawson, vice-president of strategy at the agency Jellyfish, quoted by Digiday. For anyone marketing software, the useful question is not whether the share climbs. It is what the machine aims at, because that definition just moved.

Google's targets stopped beating themselves

Target CPA and Target ROAS used to work as ceilings. Set a $10 cost-per-acquisition target and the system could keep delivering conversions at $5 if the auction allowed it, so the number behaved like a limit you hoped to undercut. Under Google's revised approach the target is read literally: a $10 setting means the system aims for conversions costing about $10, not less. That account comes from a session led by Reva Minkoff of Digital4Startups, written up by Search Engine Land on September 9. Search Engine Land is owned by Semrush, which sells tools into this category.

The trade is predictability for upside. Budget increases become easier to forecast when the platform holds performance near a stated efficiency level, and campaigns that quietly outran their targets for years lose that hidden margin. Minkoff's remedy is unglamorous: open close to what the account already achieves, then walk the target down 10% to 20% at a time and let each change sit for a full conversion cycle. She cites a transportation client whose CPA fell 75% in two weeks as the target stepped from $10 to $7.50 to $5.

Two weeks is not a B2B software sales cycle. A trial that converts to paid in 45 days, or an enterprise deal that runs two quarters, will not produce a mature read inside a fortnight, and moving the target before conversions settle means tuning on data that has not finished arriving. The cadence matters more than the direction.

The machine optimises whatever you call a conversion

Here is the part that should worry a SaaS team more than the bidding change itself. If low-quality leads are reported back to the platform as successful conversions, the system will go and find more of them, and it will be behaving correctly while it does. Every form fill counted as a win teaches the auction that people who fill in forms are the customers you want. In B2B software, where a demo request from a student, a competitor or a bot costs the same to acquire as one from a qualified buyer, it compounds against you for months before anyone in revenue notices the pipeline is thinner than the dashboard.

Your dashboard is more confident than your data

The third piece is measurement. Writing in Search Engine Journal on September 8, Bengu Sarica Dincer, a SaaS SEO manager at Designmodo, argues that a growing share of what appears in attribution reports is modelled or statistically reconstructed rather than observed, and that platforms present estimated and measured outcomes with the same visual confidence. Her illustration is familiar to anyone who has sat through a pipeline review: three systems report 150 conversions, 180 conversions and 120 new customers, and none is wrong. They count different events over different windows.

She recommends anchoring on whichever system sits nearest the real commercial outcome, usually the CRM or the subscription record, and treating analytics and ad platforms as evidence about the journey around it rather than as rivals for the truth. When upper-funnel work looks like it contributes nothing, she warns, teams defund it, and the decision feels data-driven even though it may only reflect what the tracking could see.

Put the three together and the shape is clear. A rising share of budget goes to systems that optimise toward a signal, the signal now sets the performance level rather than a ceiling, and the reports used to check the result are partly estimated. Automation is not the risk. Automation pointed at a conversion definition nobody has audited since the last website rebuild is the risk.

• Audit what you send back to Google and Meta as a primary conversion. If it is a form submission rather than a qualified opportunity, the auction is being trained on the wrong buyer.
• Feed quality signals, not just events. Offline conversion imports or value rules that reflect pipeline stage give the system something worth optimising toward.
• Split branded from non-branded before you set any target. Cheap brand conversions flatter every blended number they share with cold demand.
• Judge a target change only after a full conversion cycle has closed, not on the platform's prompt.