The Work Comes Home: How AI Quietly Moved B2B Marketing's Agency Budget Inside

There is a conversation happening in B2B marketing right now that almost never makes it into a press release. It happens in a renewal meeting, usually in the last ten minutes, when a CMO looks at an agency retainer and says some version of: we produced most of this ourselves last quarter.

Not all of it. The positioning work still came from outside, and so did the brand film. But the campaign variants, the localized landing pages, the fifteen versions of a product one-pager for fifteen segments, the social cutdowns, the first draft of nearly everything. A year ago that work justified a line item. Now a marketing ops person with a decent prompt library turns it around in an afternoon, and the retainer starts looking less like partnership and more like a subscription nobody re-examined.

For CMOs, B2B Marketing Leaders, Marketing Operations Teams, and anyone who signs an agency statement of work, this is what the data says about where that money is actually going, what in-house teams are genuinely absorbing, and the specific capabilities that get quietly destroyed when a company brings the wrong work home.

The money moved before anyone announced it

Start with the budget, because the budget tells the truth earlier than the strategy deck does.

Gartner's 2026 CMO Spend Survey, fielded with 401 senior marketers across the US and Europe, found marketing budgets sitting at 7.8% of company revenue, barely moved from 7.7% the year before and roughly 18% below where they were four years ago. Flat top line. But underneath that flat number, the mix shifted hard. Labor climbed to 24.5% of the marketing budget, up from 21.9% in 2025. Paid media grew to 31.4%. Gartner is direct about where the money came from: cuts to agencies funded it.

Meanwhile CMOs are putting 15.3% of budget into AI, and 70% say becoming an AI leader is a defining goal for the year. Only 30% describe their AI capabilities as mature enough to scale.

Put those three numbers next to each other and the year makes sense. Marketing leaders are spending more on their own people and more on AI tooling, funding both by shrinking what they send outside, and doing it while two-thirds of them admit their internal AI capability is not fully built yet. The bet is being placed before the capability is finished. That gap is where most of the pain in this transition lives.

What B2B is pulling back, specifically

The B2B numbers are sharper than the general marketing numbers, and more useful.

Forrester's 2026 B2B Brand And Communications Survey found that 93% of B2B marketers still use agencies in some capacity. Agencies have not been fired. What changed is the trajectory of spend. The share of B2B marketers expecting agency budget increases dropped 13 percentage points year over year, with roughly half now expecting flat spend. In digital marketing specifically, the share expecting to increase agency investment fell from 51% to 31% in a single year.

The prior year's survey showed the same current running underneath. Expectations for increased agency investment in digital marketing fell 14 points, content creation fell 10 points, and the share of marketers expecting outright decreases in both rose 8 points. Brand development and management, long considered the work you would never hand to an internal team, saw increase expectations fall 11 points.

That last one matters. Digital production going in-house is a decade-old story. Brand strategy going in-house is new, and it says something about how confident marketing leaders now feel about work they used to outsource on instinct.

Agencies were losing this argument before AI showed up

It would be convenient for agencies if this were purely a technology story. It is not.

Forrester's survey data on B2B agency relationships shows satisfaction gaps that have nothing to do with generative models. 80% of B2B marketing leaders say clear communication is critical to the relationship, and only 55% are satisfied with what they get, a 25 point gap. Seventy percent say they prioritize value over price, and 53% feel they receive it. On specialized skills, the kind of expertise that is supposed to be the entire reason you hire outside help, importance outran satisfaction by 15 points.

AI did not create that dissatisfaction. It removed the reason to tolerate it. When the only way to get forty campaign variants was to call your agency, a 25 point communication gap was the cost of doing business. When your own team can produce those variants internally, the gap becomes a reason to cancel.

The pressure on the agency side is severe and worth understanding, because it shapes what you will be able to buy in eighteen months. Forrester expects 15% of agency jobs to be eliminated in 2026, following an 8% reduction across major holding companies in 2025. The economics are brutal in a specific way: 75% of agencies are absorbing the cost of AI work themselves, and only 6% have found a way to monetize it. A labor-based pricing model does not survive contact with a technology that collapses labor hours. Forrester's analysts describe in-housing as having "commoditized creative ideation and execution," which is a polite way of saying the thing agencies sold by the hour is now something clients make themselves.

Talent is moving accordingly. Survey data fielded by NewtonX for Adweek in March 2026 found 54% of agency leaders somewhat or very likely to start their own firm within two years, while 32% of brands expect to handle nearly all creative in-house within twelve months and another 23% expect to bring at least half of it inside. The holding company shrinks, the senior people leave and hang a shingle, and the work they used to do gets split between a boutique and your internal team.

The part of the story that does not match the headlines

This is the point where I think most marketing leaders are getting ahead of their own evidence.

The ANA's 2026 State of In-Housing research asked in-house agency leaders what AI has actually changed about their relationship with external spend. 49% said AI is letting the internal team do more work while external spend stays the same. Only 19% said AI is letting them produce work that was previously outsourced.

Sit with the difference between those two numbers. The dominant effect of AI inside marketing organizations so far is not substitution. It is absorption of new volume. The internal team is producing more, the demand for content keeps expanding to fill whatever capacity exists, and the agency invoice looks about the same as last year.

The productivity claims are similarly restrained when you ask the people doing the work. On creative development, only 18% of in-house agency leaders report a significant positive impact with clear cost savings and productivity gains. Another 49% report moderate impact, and 21% say it is simply too early to tell. This is from a group where 64% are already using AI in creative services. These are practitioners with hands on the tools, and two-thirds of them will not claim more than a moderate gain.

None of this means in-housing is a mistake. In-house agencies have been growing for fifteen years on their own merits, well before anyone had a language model. The ANA found 82% of member marketers operating an in-house agency, up from 58% a decade earlier, with penetration expected to top out somewhere around 85% to 90%. In-house teams handle an average of 61% of the work at companies that run both models, and 92% of those companies still use external agencies. The steady-state answer was always going to be a mix.

What the data warns against is the specific move I keep seeing: cutting a retainer this quarter based on a productivity gain you have projected rather than measured. The marketers closest to the work are reporting moderate improvement. Budget decisions are being made as if the improvement were total.

What actually breaks

Three failure modes show up repeatedly, and each has a cost that lands somewhere other than the budget line where the savings were booked.

Sameness. When every company in a category runs the same models over the same competitive research and the same category conventions, output converges. Agencies were never a perfect defense against this, but an outside team that also works in adjacent categories brings pattern recognition your internal team cannot manufacture from inside the building. Pull all ideation inside, run it through the same tools your three closest competitors use, and you get a category where everyone's messaging reads like it was written by the same tired committee. Which, in a sense, it was.

Governance. Forrester's 2026 B2B predictions carry an uncomfortable number: B2B companies will lose more than $10 billion in enterprise value through falling stock prices, legal settlements, and fines tied to ungoverned generative AI use, driven by untested functionality colliding with lagging user skills. Agencies, whatever else you think of them, came with review layers. Legal review, brand review, fact-checking, a second set of eyes with professional liability attached. When production moves to a two-person internal team under deadline pressure, those layers are the first thing dropped, and nobody notices until something ships that should not have. Forrester also found 19% of buyers using AI applications feel less confident in their purchasing decisions because of inaccurate information. Your unreviewed content is part of what feeds that.

The capability you cannot rebuild. Some agency work looks like production and is actually judgment: positioning that survives a competitive attack, category narrative, the outside read on whether your own story makes sense to someone who does not already work there. That capability is expensive, infrequently used, and easy to cut because you only miss it during a repositioning or a launch. Cut it and you will not feel the loss for two years, at which point you will discover that rebuilding it takes longer than a fiscal quarter.

A sorting rule that holds up

The question worth arguing about is which work belongs inside, and the B2B teams handling this well tend to sort it the same way.

Bring inside the work where context beats craft. High-volume, high-repetition output that depends on knowing your product, your segments, and your customers cold. Campaign variants, localization, sales enablement collateral, email programs, social, first drafts across the board. Your internal team knows things about your buyers that an agency learns slowly and imperfectly, and AI multiplies that context advantage. This is also where the ANA's respondents report their real gains.

Keep outside the work where craft beats context, or where you need someone who does not work for you. Positioning and category strategy, brand identity, original research design, analyst relations, and anything requiring specialized regulated expertise. Also anything where the value comes precisely from an outsider's willingness to tell you your messaging is confusing.

Do not confuse volume savings with capability. If the pitch is that AI lets your team produce four times the output, ask what happens to quality review at four times the volume. Capacity gains are only gains if the review layer scales with them. Budget the review, staff it, and treat it as part of the cost of in-housing rather than the thing you cut to pay for it.

Restructure the agency relationship before you cancel it. Forrester expects agencies to shift from selling services to selling solutions, with pricing moving toward outcome-based and product-led models. That shift is available to you now as a negotiating position. The retainer that paid for production hours can become a smaller, sharper engagement for strategy, research, and review, priced against outcomes rather than headcount. You will pay less and get more of the part that was actually hard to replace.

Measure before you cut, not after. Run one quarter where you track internal output volume, cycle time, review escalations, and performance against the agency-produced baseline. The 18% of in-house leaders reporting clear gains have this data. The ones reporting "too early to tell" do not, and they are the majority.

One more thing worth pricing honestly: the work does not disappear when it moves inside, it changes shape. The agency was absorbing project management, version control, asset libraries, trafficking, and the unglamorous coordination that makes a campaign ship on the day it was supposed to ship. Bring the production inside and that coordination lands on someone, usually a marketing ops lead who was already at capacity. Gartner's finding that labor climbed to 24.5% of budget is partly this. You are not only hiring makers. You are hiring the people who keep the making organized, and if you skip that hire, your new internal capacity gets eaten by scheduling.

The honest counterargument

The strongest case against everything above is that the same predictions were made in 2010 about digital production, in 2016 about programmatic, and in 2019 about content studios, and agencies are still here. In-housing has never been a one-way door. Companies pull work inside during cost pressure, discover the fixed cost of maintaining a bench for uneven demand, and push some of it back out two years later. The ANA's own data hints at the cycle: 65% of member marketers have moved work from an external agency to their in-house team, which was actually down slightly from the prior study.

There is also a real risk of overcorrecting on the strength of a single survey year. Forrester's numbers show expectations for agency spend cooling, not agency spend collapsing. Expectations are not invoices. A 13 point drop in the share of marketers planning increases still leaves a market where nearly all B2B companies buy outside help and about half plan to spend the same amount they spent last year.

What is different this time is the price of the capability. Previous in-housing waves required hiring the people who could do the work. This one requires hiring fewer of them and giving them tools, which changes the fixed-cost math that used to push work back out. That is why I think the current shift sticks in a way earlier ones did not, even though the endpoint is still a mix rather than a clean break.

What this looks like in twelve months

The endpoint is not the agency-free marketing organization. Ninety-three percent of B2B marketers use agencies, in-house penetration tops out around 90%, and the two models have coexisted at most large companies for years. Predictions of agency extinction have been wrong for two decades and are probably wrong again.

What is ending is the retainer that paid for production capacity. That model was built for a world where making things at volume required people, and it is being dismantled by a technology that makes volume cheap and judgment expensive. The agency relationships that survive in B2B will be smaller, more senior, more specialized, and considerably harder to justify cutting, because what they sell will be the part your internal team cannot produce no matter how good its tooling gets.

The marketing organizations that come out ahead will be the ones that made the trade deliberately. They moved production inside because they measured the gain, kept judgment outside because they understood what it costs to rebuild, and spent some of the savings on the review layer that keeps the whole arrangement out of Forrester's ten billion dollar column.

The ones that struggle will be the ones that cut the retainer first and figured out what it was buying afterward.