Two previously unrelated profit opportunities are colliding in a big way for ad agencies and consultancies-you’re going to want to pay attention.

Recently every professional services firm has been preoccupied with how they plan to actually monetize their usage of LLMs, which currently are a major expense, but which they hope become an absolute profit bonanza, eventually.

A longer simmering but still very controversial topic among procurement professionals, ad agency executives, and really anyone with any stake in modern ad buying is principal media, the practice wherein agencies buy ad space themselves at one price and resell it to brands at another price, for a profit, the size of which agencies typically do not disclose.

It’s may be helpful to think of it as allowing an ad agency to play a futures market game with the value of advertising space, buying inventory for themselves at one price and reselling it to clients later a different price, though the analogy is imperfect.

What do these two things have to do with each other?

A Digiday article highlighted the recent development of ad agencies blending these two often opaque cost-and-or-profit centers, and if you’re an investment or legal professional working with companies in the advertising space, you’re going to want to read our deep dive on this.

We ran a survey on an audience of over 100,000 people following an ad industry meme page to round out our view on AI usage rates at agencies right now, as well.

So let’s get into it, after we run down some recent items worth your attention.

Agencies are offering to keep subsidizing AI costs for brands in exchange for those brands allowing them to do more principal media buying.

Principal media buying offers agencies an opportunity to hit uncommonly high profit margins, and nothing but the final price at which it is sold to the brand is disclosed.

The amount agencies are spending on AI is highly unclear currently, as they are only recently beginning to factor it into contracts with their clients.

It's one unclear bucket of money chasing another, and the notion of trading them arrives just as the prospect of much cheaper LLMs in the near future looks increasingly possible.

Let’s look at principal media buying from both sides of the argument.

The principal media buying advocates start out with a pretty solid point:

Almost every business in the world that resells goods works by selling you something for what you’re willing to pay for it, without disclosing to you what the party selling it to you paid for it.

When you go to an auto mechanic, they might put separate line items for an oil change and a new air filter on the bill, but if you demanded to know the cost of goods paid for the oil and the filter unit, it would be seen as strange an unreasonable.

This is because the mechanic, competing with other mechanics, faces downward pressure on parts markups to attract car owners, though in practice this competitive discipline is conditioned on customers' ability to compare prices and verify work quality.

Car owners are incentivized to shop around in order to not be taken in a mechanic putting a huge markup on the same parts any other mechanic can buy and install.

The cost of motor oil and car parts is arguably irrelevant to the buyer so long as they’re happy to pay the price, because it’s worth it to them and they know the total amount their mechanic charges them is similar to general market rates in their area.

Thus, the buyer and seller dance the delicate dance that makes most transactions simply work without the true cost of goods being known to the buyer.

So why, then, do those against principal media buying believe this kind of cost transparency should be the industry standard for transactions between brands and agencies when it comes to advertising space purchased on their behalf?

Arguments against principal media are a bit more complicated and require context specific to the advertising industry.

Fit into the above metaphor, with the mechanic, the con side says something like this:

Ad agencies buying your media are not mechanics, they are personal assistants managing your day to day errands for you, whose job in this case is to find you the mechanic that gives you the best value in terms of the quality of service relative to what they charge. The minute they themselves can make money, secretly, by recommending a mechanic that is not the best value to you, there’s a problem, and now you also need to be suspicious of the butchers, bakers, and candle stick makers they have selected for you.

You expect the mechanic to buy parts at one cost and mark them up to another to make their money. This is not what your agreement with your personal services shopper who hires a mechanic entails.

Advertiser concerns around non-transparent pricing from agencies really amped up with the release of a 2016 report by the Association of National Advertisers and K2 Intelligence.

The report is complex and surfaces many issues, but the one that raised it to high alert status in the ad industry is a practice the ANA report found and characterized as a breach of contract and fiduciary duties:

“Numerous non-transparent business practices, including cash rebates to media agencies, were found to be pervasive in the U.S. media ad buying ecosystem.”

This means some media agencies had abandoned what many clients and industry groups consider to be their core service, which is to give their clients good counsel on which is the right media to help their advertising campaigns succeed.

These secret kickbacks indicate some of them were willing to obfuscate financial details to hide that they were not advising clients to buy media to meet their goals, but to buy whatever media increased their profits the most.

Principal media’s key distinctions from non-principal media buying are that it offers the buying agency an opportunity to make additional profit (or incur losses), and it is typically non-transparent, which is to say it stands out as the ultimate tool for an agency looking keep making money hand over first, without dealing with the potential breach of contract issues that may arise from taking cash kickbacks.

So while many media buying practices came under scrutiny as concerns raised in the ANA report spread through the industry, it’s easy to see why lots of the negative attention focused on media agencies’ opaque profiteering back in 2016 eventually re-settled later on principal buying.

Besides being a perfect vehicle for brands’ imaginations to run wild with fever dreams of fiduciary malfeasance, some other elements of the way media agencies priced and sold their services rankled advertisers when principal media was pulled into the mix with them.

One issue involves another revenue stream for large media agencies: a % commission charged on the media they bought. For instance: at a 10% commission, for every $100,000 of ads they verifiably bought for a brand, they were paid $10,000.

If principal media is added to a deal, it struck advertisers as “double dipping.”

When there is less than perfect clarity around what was principal media and what wasn’t in an agency’s total portfolio, brands might be paying strictly a % commission on one thing and commission plus principal media profits on another thing, with principal profit rates being variable across different line items.

The end result is that a brand couldn’t know what portion of their investment in a placement was going to an agency and what portion was going to a media outlet, even if they were paying the same blanket 10% commission on both. Advertisers find it important to understand this split, which they often call “working media.”

A second issue is that it distorts an advertiser’s ability to understand the real ROI of a media line item beyond their specific deal with their current agency.

Let me give you a hypothetical example to illustrate this-

Let’s say you buy TV ads on a major sports network, and on a Spanish language telenovela superstation. The agency reporting comes back reporting that both buys have an ROI of $2.

However, unbeknownst to you, your agency is selling the sports network at a 100% markup, and the telenovela station at a 20% markup. This would mean that the ROI of these actual media buy line items alone isn’t an identical $2; the sports network ROI is 33% less than the telenovela network. The gap is made up entirely by your agency’s principal pricing power and decisions.

If you design next year’s overarching media strategy around the assumption that these channels yield an equal return, and extend this to all kinds of other assumptions that guide your creative development and ad production and audience research, you might be in big trouble!

The third and last issue I’ll cover is an important thread that runs through a lot of what I’ve already discussed: many brands have historically demanded full auditability of their paid media buys.

Principal media with no margin disclosure makes that kind of contractual obligation extremely difficult to enforce in practice, and most current agency contracts have not been updated to demand it.

We’ve made a little media buying simulation in which there are two agents serving a client with a $1 million media budget that wants a 3x ROI on their advertising.

This is, again, purely hypothetical, with data modeled quite loosely off of the media market, to demonstrate how the incentives work and behavior plays out.

One agency do not have a principal media practice, and ergo no markup on any media, so they will allocate to varied channels purely based on client ROI, as channel mix cannot impact their profitability.

The other agency can make a principal markup, and knows it needs to hit the 3x ROI goal, but generally seeks to maximize profit, whittling down as close to that flat 3 as it can get.

If our media universe looks like this…

…we get pretty similar results from both of our agencies:

Hey, a $3.63 ROI is only .1 lower, and you are WELL over your goal as a brand, so why begrudge the principal buyers that they made an extra $112k of profit?

Okay, well, what if we remember that those “chumboxes” (ahem, sorry, native ads!) that appear beneath the article on news publisher sites, usually promoting dubious miracle cures, exist?

Let’s see how our buyers act:

Oof! Okay, so now the principal buyer has moved the ROI down to $3.30, and in the name of making almost triple the profit of the non-principal buyer.

Hey, though, you’re still above your target-no fiduciary duty has been breached at the ROI level and you’re literally getting what you asked for.

But let’s go once more, and this time let’s not forget MFA (Made For Advertising) trash sites, newly turbocharged by generative AI:

Okay, let’s see how the buyers do with that kind of sewer to swim in:

You’ve missed out on a lot of potential surplus performance, and the people who delivered you these results have been rewarded much more handsomely than their non-principal trading compatriots for it.

This loose simulation was just meant to broadly outline how principal buying can play out, and this is not meant to illustrate anything like the actual likely distribution of goal attainment and relative agency ROI out there in the market.

These concerns I teased adding these dubious inventory sources are not a fantasy, though-they were the subject of the SECOND report that the ANA put out on transparency and fraud issues in 2023, positing as much as 23% of programmatic open web investment was waste.

If you’re curious how this simulation worked or want to use it yourself, here is the Github project, feel free to go crazy.

Principal dealmaking is on the rise, growing faster than advertisers’ understanding of it, in the view of the ANA.

Agencies have both an appetite for more principal media deals, and an awareness that they brands are wary of this way of working.

How, then, might one make one’s clients more amicable to an increased amount of their media budget being principal buying?

Much like every other industry, there has been much fanfare made in the advertising agency world over the rapid adoption of LLMs and related tools.

The initial belief some experts had, as it pertained to revenue and contracts, was that the billable hour model many professional service firms still use to charge for their services was going to crumble as workers became so efficient that they could do all the necessary work in a fraction of the time.

The hope was that more firms were going to be able to sell outcomes, or at least find even higher value work than the grunt work AI took over, that they could still bill hourly.

There have been ad agency job cuts on the media side where AI efficiency gains were cited as a reason, which arguably points to some kind of overall productivity gains from AI, just not in the way that agency staffers would have hoped.

On the other hand, it has been reported that agencies’ costs for these AI tools have been increasing significantly.

“Who should pay for these AI tools, agencies or brands” is actually a tough question; agency technology fees are alternately eaten by the agency or passed on to the client on a more or less case-by-case basis.

AI models fall into the technology category where brands often talk about “ownership” versus “renting” from an agency, much like customer data platforms and other data infrastructure.

If what AI models need to be a differentiator for you is lots of high value context, from rich proprietary data that only your company has, than this is a classic build-and-own case for brands with sufficient resources.

That being said, in the early days of the agency AI revolution many firms have been entirely paying for and managing the AI tools they use to do work for brands, while footing the bill.

Most of the energy put into solving this problem so far has gone into considering new pricing models that fairly allow brands to share the burden of these costs.

That is, until some intrepid dealmakers came up with a different path to balancing the books despite rising AI costs:

What if it continuing to eat these AI costs could be “traded” in exchange for clients knowingly opting into more principal media buying?

Agency AI costs are almost a dark mirror of principal media itself-right now clients don’t always have contracts that guarantee them detailed information on much their agencies spend on AI, akin to the general reduced transparency terms around principal media buys. The difference is that while the client suspicions around principal media run towards it being shockingly cheap, the current general expectation of corporate AI bills is that they are excessively expensive.

This perception is a major opportunity for agencies-

“Hey, AI is really eating our wallet, but we’ll throw it in for free if you let us do a little principal media trading to lessen the sting!”

Much like principal media, its advocates can spin a pretty decent win-win story here.

The thing we find very interesting is the timing.

The latest AI headlines have not been exclusively about the stunning achievements of frontier modelsl, even as Fable and Mythos upend notions of how cybersecurity will work in the future, and, you know, other things like that.

Often stealing the (lower wattage) spotlight since early 2025 are the new waves of high efficiency open parameter models that point to a future that is “smaller” in many ways, critically in terms of the compute bill that will come with many more basic LLM functions.

There is, of course, all the internet laypeople talk about the Jevons Paradox and the possibility that LLMs are a domain where efficiency only leads to increased usage. Still, that aside-

We are also seeing organizations become increasingly concerned about the actual ROI of these tools in the near-term, and in the hungover metaphorical morning after tokenmaxxing, we’re beginning to see people experiment with cost controls.

Advertisers, in fact, are beginning to examine this, on the brand side and at agencies.

Much of the work currently being done at professional services firms will likely be subject to efficiency gains from lower cost models or otherwise-information aggregation, basic presentation creation, and other things that make up a lot of the busywork agencies are seeking to trim should be easily manageable with tools far more modest than today's frontier models, let alone those likely to emerge in the coming years.

This is the possible tension-subsidizing the cost of AI tools being a major concession to a brand feels great if your knowledge is based primarily on a snapshot of the last 18 months.

In the news cycle of the last 30 days or so, which is the kind of small sample we must admit is relevant in an area as fast moving as LLMs are right now, it looks a little different.

It looks almost like agencies are offering brands something for free right before the cost of that thing drops precipitously. In big brand and agency world, where you still occasionally see three year enterprise contracts, the total savings over the lifetime of a deal from free AI might not be nearly what you hoped.

Through one of our partners working with Daypart AI, we were able to run a survey of the followers of a social media account very popular with agency staffers, as well as some brand and publisher side advertising folks.

Remember: this is a social media survey of anonymous users who purport to work in an industry, whose identities and employers are in no way verified, and their views represent nothing but their own opinions.

We also received some direct messages clarifying that there had been very strong, uh, incentives from management to use AI as much as possible:

“Caveating that I am tokenmaxxing (because I am forced to)”-anon

“(I am) not mandated (to maximize AI usage) but highly encouraged and pushed (to do so)”-anon

We were also told there is strong variance in terms of AI usage pressure, holding company to holding company.

“(Redcated TLA holding company #1) was shoving it down our throats, but I feel much less pressured to use it at (Redacted TLA holding company #2) now.”-anon

Are agency staffers seeing the tracking and usage control measures agency leaders have mentioned they are starting to look at? Some have:

“They updated (a holding company’s proprietary LLM tool) recently to show token usage and asked us to be more mindful of how much we are using lol”-anon

Others have been told to be aware of their usage but have not been given any systems or tools to do this:

“We’re being asked to but I haven’t heard anything about measurement / tracking yet.”-anon

And finally, some of them would probably advise you to wonder whether you want usage of these agency AI tools as part of your contract with them at all:

“Yes (I have been tokenmaxxing) but the (holding company proprietary LLM tool) fucking sucks ass.”-anon

Much to ponder! (with an appropriately huge grain of salt, given that this is an anonymous social media survey)

If you or a company you’re invested in are being offered AI tools and services at no cost, you need to obtain the most verifiable possible estimate from the vendor in question as to how much this will amount to, especially if it is explicitly in exchange for concessions elsewhere in the contract.

When offered principal media, you need to ensure you can write and sign the kind of contract that will ensure you’re happy with what you get, no matter how much more profit the counterparty makes. Know how much this reduces the auditability of your deal and if you yourself have a contractual or professional duty of loyalty to others, ensure they understand this and are aligned with you.

In the event someone offers to swap you free AI for the unlimited ability to act as a principal media buyer when activating your advertising campaigns-well, make sure you keep ‘em busy with the really tough Claude prompts, and I’ll see you in the funny papers (at the bottom, under the article, in an Outbrain or Taboola ad).

September 2019 ANA Media Agency Compensation Practices Report

WFA Guidance on Generative AI in Agency Contracts