Avinash Kaushik spent about 16 years at Google, held client-side leadership roles at companies like Intuit and DirecTV, and now advises brands and agencies as director of strategy at Human Made Machine. So when he tells marketers to renegotiate their agency contracts and expect savings of 25% to 75% starting this month, it’s worth stopping everything else you’re reading.
I knew Avinash for over 20 years. He does not throw out such figures to shock. He builds a model, shows his work, and challenges you to challenge the math. In his last Marketing < > Analytics Intersection newsletter, “Pay Less, Grow More, Agencies in the Age of AI,” is exactly what he does, and the argument goes well beyond the performance and creative agencies he directly names. Anyone managing SEO, content or GEO work through an outside partner, they are expected to make the same calculation on their own contract this week.
Kaushik’s starting point is that three forces converged at the same time. AI has become globally, generally intelligent, not just intelligent for a specific task. It turns out that ad platforms also own the fundamental AI models running underneath the campaigns they sell, so the insights emerged quickly and are already integrated into the tools agencies use every day. And every system in the stack now communicates with every other system in real time, which doesn’t make AI smarter in the abstract; this makes the AI smarter for your specific account. Put these three together and you get what Kaushik calls the “we’re not in Kansas anymore” moment for every type of agency, media, creative, performance, brand, measurement and, yes, SEO.
The practical result is that work that once warranted a monthly fee is now handled by the platform itself, and Kaushik says clients should expect savings of 25 to 75 percent on existing tasks as this shift occurs, as well as 15 to 25 percent growth in fees for truly new work that the old contract never anticipated. He is not advocating for a more restricted relationship with your agency. He argues that the money should be allocated to other tasks.
Where the old contract no longer makes sense
Kaushik divides the scope of the former agency into clusters, and the model in each of them is the same. Account architecture, keyword and audience structuring, and campaign creation once accounted for about a fifth of the cost of a typical contract, and he estimates that work can be reduced by almost 80% now that the platform’s algorithms handle segmentation and targeting better than a human team slicing up campaigns “for review.” Manual bid and pacing adjustments tell a similar story, with AI already outperforming human pacing decisions since late 2024, and Kaushik makes a more specific point here that deserves attention beyond his own newsletter. He claims that an agency intervening to “rescue” a dip during an AI learning cycle isn’t helping, it’s actively sabotaging the algorithm’s ability to learn, and each of these manual rescues resets the clock.
The reports tell the same story from a different perspective. Weekly presentations, twice-weekly status meetings, and hand-typed comments on numbers that are already in a dashboard used to account for nearly a third of the cost of a contract, by Kaushik’s estimate, and he thinks 60% of that may disappear now that AI-driven data tools can explain what happened and why without a human translating a spreadsheet. None of this means that the work disappears. This means that the work is transferred to a machine and the real work of the agency is to decide what the machine should optimize towards, without building or reporting on what the machine is already doing on its own.
The contractual structure with which Kaushik wants to replace it
This is where the column stops being descriptive and starts being prescriptive, and it’s the part I think SEO managers will recognize the most quickly. Kaushik wants agency fees to be split three ways. A Lean mandate base, representing around 40 to 50% of a now smaller total, covers governance, management and data engineering. The project fee, an additional 30-40%, covers work that still requires true human judgment, creative design, complex strategic analysis and portfolio strategy. And an outcome incentive, the other 15-25%, is directly tied to incremental profit or verified revenue increase, never to a platform-reported metric like ROAS that the platform itself has a vested interest in inflating.
This last point is the same argument I made Quote Share of voice And GEO measurement for most of this year. Kaushik’s warning against paying agencies a “percentage of media spend” is structurally identical to the warning I would give to any SEO team that is still paying an agency for deliverables, number of pages, audits shipped, blog posts published, instead of paying for organic revenue actually generated. A contract built around activity always rewards more activity. It never rewards the judgment of doing less and getting more, and AI-powered search environmentjudgment is the one thing a machine cannot do entirely for you.
What SEO teams should actually do with this
Kaushik’s own examples apply to performance marketing and Google Ads, but the underlying calculations apply just as directly to relationships between SEO and content agencies, and here’s how I would apply them.
First, pull the current SOW from your SEO or content agency and sort each line item into one of Kaushik’s three buckets. Anything that looks like keyword research work, manual ranking tracking, or template-based technical audits belongs in the “the platform already does it” pile, because AI-driven crawlersGoogle Analytics 4 anomaly detection and automated technical monitoring have filled most of this gap. Be honest about how much of your current provision is quietly paying for this stack.
Second, offer to split your next renewal the same way as Kaushik, a smaller retention base for data governance and ownership, project fees for work that actually requires a strategist’s judgment, like entity construction, content architecture for AI Mode and AI Previewsor GEO strategy, and a results incentive tied to organic revenue or a verified increase in quote share of voice, not a vanity metric like keyword rankings or the number of deliverables any AI tool can now generate on command.
Third, and this is Kaushik’s warning applied directly to SEO, make sure you own your own data before renegotiating anything. The GA4 access, Search Console, log files, and AI citation tracking you use should be in your hands, not your agency’s, or you have no way to request any of the above.
My opinion
I don’t think this means SEO agencies are becoming worthless, and neither does Kaushik about agencies in general. I think this means that the agencies that survive over the next couple of years will be the ones that stop selling hours and start selling the judgment that a machine still can’t provide, and the ones that continue to charge for monthly rebuilds and manual reporting will lose the client before they lose the discussion.
Kaushik ends his article with a sentence that I keep coming back to. You can pay for the past or for the present. After 20 years of watching agency-client relationships evolve in this industry, I think this is the right framework, and I think most SEO teams are still writing checks for the past without realizing it.
More resources:
Featured image: Master1305/Shutterstock
