An AI PPC agency sells you two things: management of automated campaign types like Performance Max and Advantage+, and the creative and measurement work those systems need to function. The first part is now mostly done by the platform. The second part is where the money is genuinely earned. And the awkward number for anyone shopping for one is this: in a survey of 1,306 PPC professionals, 73% of in-house teams said they are committed to keeping paid search internal, up from 44% two years earlier, and 20% of clients were considering replacing agency work with AI tools outright.
That is the market moving away from agencies at precisely the moment agencies are repositioning themselves as AI-powered. Worth understanding why before you sign anything.
Automation did not make this easier
The pitch on most agency pages is that AI has made paid search more efficient. The practitioner data says something close to the opposite.
In that same survey, 53% of PPC professionals said managing paid search is harder than it was two years ago. Only 16% said easier. When the ones finding it harder were asked why, 65% pointed to black-box platform technology, 50% to signal loss from privacy changes, and 43% to increased competition.
Performance Max specifically: 50% use it often or always, 56% are frustrated by its black-box targeting, 41% by opaque attribution, and 32% flagged keyword cannibalization inside their own accounts. Notably, 38% still run Standard Shopping alongside it rather than trusting PMax with the whole job.
So the honest version of the value proposition is not that automation made campaign management simple. It is that automation moved the difficulty somewhere new, and the new difficulty is harder to explain on a pricing page.
What you should expect to pay
Pricing has not changed much even though the work has. The prevailing structures:
- Percentage of ad spend: typically 10% to 20% of monthly spend, with hybrid B2B arrangements around 12% to 15% of media budget
- Flat retainer by spend tier: under $5,000 spend runs about $750 to $1,500 a month; $5,000 to $15,000 runs $1,500 to $3,000; $15,000 to $50,000 runs $3,000 to $7,500; $50,000 to $100,000 runs $7,500 to $12,000; $100,000 to $250,000 runs $10,000 to $20,000
- Hourly consulting: around $100 to $150 an hour for experienced media buyers
- Retainer minimums: commonly $1,000 to $1,500
Regional note from the same survey: 25% of European agencies bill hourly compared with 12% in North America.
The percentage-of-spend problem nobody wants to discuss
Here is the structural issue with the most common pricing model, and I have not seen an agency page address it.
When a human was choosing bids, keywords, and placements, paying them a share of spend was defensible. Their decisions determined the spend. Now the platform’s algorithm determines the spend, and it is optimising toward conversion volume, which generally means spending more. An agency on 15% of spend earns more when the algorithm scales the budget, whether or not that scaling was incremental.
That is not an accusation of bad faith. Most agency operators are not consciously reasoning this way. It is just that the incentive and the outcome you want have quietly decoupled, because the party making the spending decisions changed and the fee structure did not.
If you take one thing from this article: ask for flat-fee or performance-based pricing, or at minimum a fee that caps at a spend ceiling. An agency that resists that conversation entirely is telling you something.
What the work actually is now
Strip out bid management and audience building, because the platform does those, and four things remain. These are what you are actually buying.
Creative volume. Automated systems need variation to test against. Accounts producing 20 or more ads monthly report meaningfully higher returns than accounts producing a handful, not because more ads are inherently better but because a starved system optimises inside a tiny space. Creative production is now the bottleneck in most paid accounts, which is a different skill from the one agencies historically sold. Our roundup of AI content marketing tools covers that production layer.
Incrementality measurement. Holdout testing, where a randomised share of your audience is withheld from advertising entirely so you can compare conversion rates. This is the only method that tells you whether reported conversions were earned or harvested. It costs reach and it is the single most valuable thing a good agency does above roughly $10,000 monthly spend.
Account structure and feed work. Product feed quality, conversion tracking accuracy, exclusion lists, and asset group segmentation. Unglamorous, and it determines whether the automation has anything useful to work with.
Knowing when not to automate. In B2B accounts, Search campaigns have been shown to outperform Performance Max, roughly 553% against 436%. For considered purchases with long cycles and small qualified audiences, keyword intent still beats audience modelling. An agency that puts everything into PMax because it is easier to manage is optimising for their workload.
Six questions that separate real from repackaged
- “Show me a holdout test you ran for a client at my spend level.” Not a case study with a ROAS number. A test with a control group. Most cannot produce one.
- “What is your fee if my spend doubles?” You are testing whether the model is aligned, not the number.
- “How many creative variants do you produce monthly, and who produces them?” If the answer is vague or the work is subcontracted invisibly, creative volume is not actually part of the service.
- “Which of my campaigns would you keep manual?” Anyone who says none has not thought about B2B or high-intent segments.
- “How do you report new-customer acquisition cost separately from blended ROAS?” This is the question that matters most, because blended ROAS can hold steady while new-customer costs climb sharply. If they do not separate the two, they cannot tell you whether you are growing.
- “What do you use for click fraud prevention?” In that survey, 71% of practitioners used no dedicated software at all. A specific answer here is a signal of operational seriousness.
One more thing worth knowing: 50% of PPC professionals still track budgets manually in Google Sheets or Excel. An agency describing an “AI-powered proprietary platform” may well be describing a Looker Studio dashboard over a spreadsheet. That is not necessarily bad. It is worth knowing before you pay a premium for it.
When to keep it in-house instead
The in-housing shift is happening for reasons that apply to a lot of businesses:
- Your spend is under roughly $10,000 a month, where agency fees eat a large share of a small budget and holdout testing is not yet viable
- Your product or sales cycle requires context an external team cannot absorb quickly
- You already have someone who can produce creative at volume, which is now the scarcer skill
- Your account is stable and mature, where the platform’s automation does more of the work with less oversight
The counter-argument is real too, and it is the strongest case for an agency in 2026: 90% of in-house teams report difficulty finding qualified talent, and 88% of in-house teams have five or fewer people. If you cannot hire, an agency is a hiring workaround. Just price it as one.
Note also that a study of enterprise AI deployment found external partnerships roughly twice as successful as internal builds, which cuts against the in-housing enthusiasm. The honest read is that both routes work and the deciding factor is whether you have the people.
The wider shift to plan for
Two structural changes are worth building into any agency conversation right now.
Meta is removing the older creation paths for Advantage+ Shopping and App campaigns around 19 May 2026, folding them into a single automated setup. Manual alternatives are being retired rather than maintained, so an agency’s manual expertise has a shelf life.
Separately, roughly 78% of Google Ads spend now flows through Smart Bidding or Performance Max, and the organic top-of-funnel that used to feed retargeting audiences is shrinking as AI-generated search answers absorb clicks. That pushes more acquisition burden onto paid and makes attribution murkier, which is covered in more depth in our comparison of GEO vs SEO.
For the tooling side of this, our overview of AI marketing tools is a reasonable starting point, and AI marketing strategy takes the wider cross-channel view rather than paid acquisition alone.
The short version
An AI PPC agency is worth hiring if it sells you creative volume, incrementality testing, and account hygiene. It is not worth hiring if it sells you bid management, because you are paying for something the platform now does for free.
Ask for the holdout test. Everything else is negotiable.
