The Great Agency Repricing

The retainer model, the hourly rate, and the entire economic logic of "paying humans for time" are all having a very bad quarter. Whether you're an agency or a freelancer who just updated their day rate on a proposal — this one's for you.

At an internal presentation in late June, Deloitte showed its consultants a projection suggesting that traditional labour-based, hourly-rate consulting could shrink to a thin sliver of the total market by 2035, replaced by AI agents and productised solutions. One person in the room described the mood in terms that roughly translated to: the model, as they'd known it, was finished.

Now, this is Deloitte telling its own people that the way Deloitte makes money might not work anymore. That's like a restaurant putting "the food here is mid" on the specials board. When you're that honest internally, something genuinely structural is happening.

And it's not just consulting. The same repricing is hitting every service business in the ecommerce ecosystem — agencies, freelancers, SaaS vendors, and the brands trying to figure out which invoices still make sense. The common thread is painfully simple: when AI compresses the time required to complete a task, any business model that charges for time has a problem it cannot solve by working harder.

The Spreadsheet That Started a Crisis

McKinsey's internal AI tool, Lilli, now processes over 500,000 prompts per month, with consultants reporting up to 30 per cent time savings on knowledge work. About a quarter of McKinsey's fees globally are already linked to outcomes rather than hours. Bain says AI-enabled work represents roughly 30 per cent of its consulting business, heading toward 50 per cent. EY's leadership has openly floated something called "service-as-software" pricing — which sounds like it was named by someone who's never had to explain an invoice to a small business owner, but the concept is sound: pay for results, not labour.

The incentive misalignment here is almost elegant in its brutality. Technology enables efficiency, but revenue depends on inefficiency. If a team that used to need forty hours can now finish in ten, the client will eventually ask why they're still paying for forty. Not might ask. Will ask. And the honest answer — "because our entire compensation structure was built around billing for forty" — tends to land poorly in procurement meetings. (Procurement meetings being, as we all know, where joy goes to die.)

The Agency Numbers That Should Worry Everyone

The agency world is experiencing its own version of this, and the data is... bracing.

The Basis 2026 Advertising Agency Report surveyed 213 agency professionals and found that 87 per cent believe the traditional agency model is either broken or will be within three to five years. Nine in ten believe AI threatens their primary revenue streams. And for the first time in the survey's history, fewer than half of agency professionals feel optimistic about the future of digital advertising.

Fewer than half. Of the people inside the agencies. If pilots started expressing this level of confidence about aviation, we'd all be taking the train.

Meanwhile, 32 per cent of brands expect to handle nearly all creative in-house within twelve months. Sixty per cent of US senior marketing leaders said they're already spending less on agencies as a direct result of AI. Forrester forecasts a 15 per cent reduction in agency jobs in 2026.

But here's the number that really tells the story. In 2025, worldwide ad spending grew by 8.6 per cent. Over the same period, holding company revenues fell by 1.2 per cent. The market is growing. The share going to agencies is shrinking. That's not a downturn. That's the market repricing what "service" means.

And this isn't because agencies suddenly got worse at their jobs. It's because AI closed the capability gap that historically made in-housing prohibitively expensive. Three years ago, building an internal content operation meant hiring a small army. Today, one skilled marketer with the right AI stack can produce what that army produced. The economics shifted, and everyone — agencies and brands alike — is now scrambling to figure out what the new maths actually looks like.

What Actually Broke (And What Didn't)

Not everything got commoditised equally, which is worth being precise about.

The most exposed services are the ones that are easy to describe, easy to prompt, and easy to compare: content production, social copy, ad variations, monthly reports, standard landing pages, basic SEO content. These are tasks where a moderately capable in-house marketer with AI tools can now produce comparable output at a fraction of the cost. That's not an insult to agencies — it's just what happens when the floor of "good enough" rises dramatically.

The conventional wisdom says value migrates from execution to strategy. Comforting narrative. Also incomplete.

Here's the part nobody particularly enjoys hearing: most of what gets sold as "strategy" across the industry — auditing a brand's position, mapping competitive gaps, recommending a channel mix, building a quarterly plan — is structured analytical work. The reasoning is replicable. The frameworks are public. The data is accessible. And structured synthesis is precisely what large language models were built to do. (Which is either terrifying or freeing, depending on whether you've been charging $1,200 a day for structured synthesis.)

This doesn't mean strategy has no value. It means the word "strategy" has become a very comfortable hiding place for work that ranges from genuinely irreplaceable judgement to competent-but-replicable analysis — and AI is sorting those two categories faster than anyone expected.

The Thing That's Actually Scarce Now

So if execution is commoditised and structured strategy is increasingly automatable, what's actually left?

Context.

Not context in the "we understand your business" sense that appears on every agency credentials deck between "data-driven" and "full-service." (Two phrases that have never meant anything and now mean even less.) Context in the operational sense: the ability to determine what problem is worth solving, what data is worth trusting, what output is worth rejecting, and what risk is worth owning.

An Amazon agency running your PPC is selling execution. An agency auditing your account and recommending a channel mix is selling strategy. But the partner — agency, consultant, freelancer, whoever — that tells you which products should not be advertised at all, which competitive signal everyone has misread, and where the data is leading toward a decision you'll regret in two quarters? That's context. And context is what AI cannot generate, because context requires knowing what to exclude, not just what to include.

This connects directly to The Great Flattening. AI homogenises outputs because language models converge toward the statistical centre of everything ever written. But the decision about what your brand should say, why it should say it, and what it should deliberately refuse to say sits upstream of the model. That's context curation. It gets more valuable as execution gets cheaper — not less.

Three Models That Might Survive This

Within this repricing, three distinct models are emerging for service providers who plan to still be doing this in 2028.

The Strategic Partner positions itself as an irreplaceable advisor, monetising creative IP and strategic impact through outcome-based partnerships. The catch: if your "strategic expertise" is really structured analysis with a nice deck around it, you're exposed to the same compression that hit execution. This model only works for partners with genuinely deep vertical expertise — the kind of category-specific pattern recognition that a general-purpose AI can't match. (Yet. Always yet.)

The Productised Orchestrator builds and sells the systems that do the work, rather than doing the work itself. Dept's Agent Studio is the clearest example — they completed an ecommerce redesign for Blackroll 3.8 times faster than estimated without agents. The value isn't in the deliverable. It's in the infrastructure that produces it. Less "hired help," more "the people who built the robot." A subtle but important distinction when your client starts asking why they're paying for forty hours.

The Accountability Partner ties their compensation directly to outcomes. McKinsey reports over 30 per cent of its fees now come from outcome-linked pricing. BCG says three-quarters of its largest AI engagements use variable-fee structures. The logic: if AI compresses the labour, the client is buying the result, not the hours. And a provider willing to stake their own compensation on that result is making a statement about confidence in their own judgement. (Also a statement about their stomach for risk, which is arguably the more impressive quality.)

The In-House Trap (Because There's Always a Trap)

Before anyone on the brand side does a victory lap about pulling everything in-house, Deloitte's own research found that AI pilots built through strategic partnerships are twice as likely to reach full deployment compared to those built internally. PwC puts a finer point on it: technology delivers only about 20 per cent of an initiative's value, while the other 80 per cent comes from redesigning how work gets done around the technology.

That 80 per cent is where external partners still have a legitimate claim. Not because brands can't do the work, but because the partner has seen the same implementation fail in fifteen other organisations and knows which three variables determine whether it succeeds.

Internal teams have an inherent limitation that has nothing to do with talent: they see one company's data. One set of experiments. One set of results. External partners working across multiple clients build pattern recognition that only emerges from watching the same thing go wrong — and occasionally right — across dozens of comparable situations simultaneously. That cross-pollination of failure is, ironically, one of the most valuable things a service provider offers.

The Bottom Line

The billable hour isn't dying because clients decided to stop paying for time. It's dying because AI made time a poor proxy for value, and the market is correcting.

This affects everyone in the ecosystem. Agencies need to articulate whether they're selling context, orchestration, or accountability — because "we do the work" is no longer a defensible value proposition when the work takes a quarter of the time it used to. Brands need to figure out which capabilities genuinely require external help and which ones they're outsourcing out of habit. And freelancers... well, freelancers need to update their proposals. (Again.)

The repricing is uncomfortable for anyone built around the old model. But for the partners who can prove they offer something the client can't replicate internally — and who are confident enough to put their fees where their mouth is — this is the market finally recognising a form of value that the billable hour was never equipped to measure.

P.S. If your agency's answer to "how are you adapting to AI?" is a new slide in the capabilities deck, that's not adaptation. That's a font change.

P.P.S. And if your brand's answer to the same question is "we're pulling everything in-house," please circle back to the section about the 80 per cent. I'll wait.

• Amazon's black market for insider access persists: middlemen on WeChat and Telegram sell leaked seller records and account reinstatements for a cut. One suspended seller documented a contact offering to recover his frozen $90K for 20%.

• Amazon starts buying ads on ChatGPT, routing users back to its own storefront while still blocking OpenAI's bots from scraping its catalog. It treats AI platforms as marketing channels, not data-sharing shopping partners.

• OpenAI pitches ChatGPT ads as the shift from attention to intelligence economy, targeting the 20% of its 900M weekly users with commercial intent. Cost-per-click already outpaces CPM spend, just 19 weeks after launch.

• TikTok launches Symphony Agent, an agentic AI that builds TikTok-first campaigns, sources creator content, and matches creators across markets. Starbucks pilots a custom Creator Network paying employees ad revenue for their videos.

• OpenAI signals it will add third-party ad measurement, breaking from self-graded platform norms. The catch: it won't share chat contents with advertisers, and no one has yet cracked how to measure influence inside a conversation.

⚠️ Norm AI - Eliminate compliance risks for your business at scale with AI

🌎 Globe Explorer - Visual-hierarchical search on any topic you want.

🗒️ Circleback - AI-powered meeting notes and workflows.

🤖 Fireflies AI: Automate your meeting notes. Transcribe, summarize, search, and analyze voice conversations.

😍It was great, all I needed

🙂Good, not great

🙄It sucked