AI and Marketing Agencies: What to Keep In-House and What to Outsource

By Camellia · Updated Oct 11, 2026

The agency vs in-house decision is the question of which marketing work your own employees should do and which work an outside firm should do. AI changes the answer because it cuts the cost and skill needed for routine production, such as drafts, variations, reports and basic analysis, while leaving the expensive parts (strategy, taste, specialist craft, hard measurement) as costly as before. The practical result: more repeatable work belongs in-house, and agencies earn their fees on judgment and scarce skills instead of hours of output.

If you run marketing for a company with an agency on retainer, this isn't an argument for firing anyone. It's a reason to re-draw the line deliberately instead of letting it drift. This article walks through what's already happening in the market, a function-by-function split, how agency pricing is shifting, and a step-by-step way to renegotiate. For the team side of the equation, see AI marketing team structure, which this piece doesn't repeat.

The in-housing trend was already underway

AI didn't start this. Bringing work in-house was growing well before generative tools arrived. The Association of National Advertisers (ANA) has tracked it every five years. Its research reports that 82% of ANA members have an in-house agency in 2023, up from 78% in 2018, 58% in 2013 and 42% in 2008, and the ANA predicts penetration will eventually peak at 85 to 90%.

The detail is more useful than the headline. In the same 2023 study (162 respondents, fielded in February and March 2023), 92% of respondents still used external agencies, and 65% had moved some established business from an external agency to their in-house agency. So the real pattern isn't "agency or in-house." It's a mix, with the boundary moving over time.

The same coverage reports why companies do it. Cost efficiency was the primary benefit, followed by better brand knowledge, institutional knowledge and dedicated staff, and the desire to own, control and protect first-party data was a driver. The services most often moved in-house were digital creative (social media, search and email), several traditional creative formats, and some media work. Respondents who had considered bringing media in-house but hadn't said it was "too complex."

That study didn't discuss AI. So treat AI as an accelerant on a trend that already existed, not as its cause. The reasons companies in-house (cost, brand knowledge, data control) are exactly the reasons AI makes in-housing easier. Faster drafting and cheaper variation mean a small internal team can now cover work that used to need a bench of outside producers. For the data angle, see first-party data for AI marketing.

What AI changes about the economics

Three shifts matter for the decision.

Production gets cheaper. Digiday's reporting from May 2025 cites MediaLink figures showing average creative production savings above 27% from AI. Whatever the exact number at your company, the direction is clear: the same asset takes fewer hours. If you pay by the hour, you should expect either lower bills or a different billing basis.

The skill floor drops. A generalist marketer with good prompts and a tight brief can now produce acceptable first drafts, ad variations, subject lines and summaries. That doesn't make them a copy chief. It does mean you no longer need an outside firm just to get volume.

The scarce skills stay scarce. Brand strategy, original research, media negotiation at scale, complex platform engineering, PR relationships and creative that actually stands out are still hard. AI helps with the process around them, not the core talent.

The upshot: the work AI makes cheap is the work most worth examining for in-housing. The work it doesn't touch is where agencies still justify themselves.

What to bring in-house, what to keep outside, what to share

This table is a starting point, not a verdict. Your team's size, skills and risk tolerance change it. The "AI role" column shows where the tools do the heavy lifting.

A few notes on reading it. "Lean" means where I'd start the conversation, not where I'd end it. And "Hybrid" doesn't mean split down the middle. It usually means the agency owns the specialist craft while you own the brief, the approval and the data.

Examples by marketing function

Content. A team that pays an agency for 20 blog posts a month can often move the first draft in-house and keep the agency for the pieces needing deep expertise or original research. The in-house editor owns voice and accuracy. Google's guidance warns that producing many pages without adding value for users may violate its spam policy on scaled content abuse, so the decision isn't "AI writes everything cheaper." It's about putting people with real knowledge on fewer, better pieces.

Paid media. Day-to-day account upkeep (reporting, simple tests, creative rotation) is increasingly in-house work. Strategic planning, new-channel launches and anything involving large negotiated buys often stay with specialists.

Creative. Digiday's coverage includes an example of a localized campaign with 100 creative variations priced at $1,000 each rather than billed by hours. That is a clean illustration of volume work moving to per-output pricing. The flagship concept and art direction are a different category of work.

Analytics. Pulling data and drafting the first-pass narrative is cheap now. Deciding what the numbers mean for next quarter's budget stays with someone who knows your business. See AI marketing analytics.

Email and CRM. This tends to move in-house first, because it sits closest to your customer records and your product, the very data companies say they want to control.

How agency pricing models are shifting

Most agencies bill in a handful of ways. Here's how each one holds up when AI shortens the work.

The honest state of play is mixed. Digiday reported in May 2025 that AI is speeding up advertising work but agency compensation is still mostly tied to billable hours. In the same piece, Forrester analyst Jay Pattissall said AI had not yet transformed the remuneration model, and executives named culture, technology and management as obstacles to outcome-based pay. The article also described attribution as a hurdle, since tying payment to outcomes is hard when results depend on things the agency doesn't control.

Still, some agency leaders were already signaling change. The article reported that Monks' co-founder and chief AI officer Wesley ter Haar criticized reliance on billable hours and headcount and wanted to move toward output-driven models. It also reported that Mighty & True's CEO expected fees to fall as AI lowers costs, and that Luckie's chief AI officer had built a plan projecting at least 25% overall efficiency gains and more than 70% in copywriting while exploring value-based pricing.

So as a buyer, don't assume outcome-based pay is available or even desirable. For most clients today the realistic move is a hybrid: a retainer for the strategic core, per-output pricing for volume work, and a smaller performance component tied to metrics the agency actually controls. Be careful with the last one. If you pay for leads, but your sales team can't convert them, you've just created an argument about whose fault it is.

A step-by-step way to re-draw the line

Do this after you've set your goals and scored use cases, as covered in how to build an AI marketing strategy, so the split follows priorities rather than habit.

1. Inventory your agency spend. List every deliverable you pay for, with the fee, the hours behind it where visible, and who inside your company approves it.
2. Sort the work. Mark each item as strategic, specialist, or routine production. Routine production is your in-housing candidate list.
3. Check your internal capacity honestly. Do you have people who can brief, edit and approve? AI drafts still need an editor. If you don't have one, you're not ready to bring that work in.
4. Pilot one function. Pick something low-risk like email variants or social calendars. Run it in-house for a quarter and compare cost, speed and quality against the agency's output.
5. Write down quality standards. Voice guidelines, fact-checking rules and approval steps. See AI brand voice guidelines.
6. Open the pricing conversation. Ask your agency how AI changes their delivery and what they plan to pass on. Good partners will have an answer. Ask whether per-output pricing or a smaller retainer is available for routine items.
7. Set the new scope. Move routine work out of the retainer, add the strategic work you were underbuying, and define what the agency will not do.
8. Agree measures. Cost per deliverable, time to publish, error rate and business results. Revisit quarterly.
9. Decide who owns the AI tooling and data. Contracts should say where your data goes, which tools the agency uses on your account, and who owns the outputs.
10. Review annually. The boundary will keep moving as tools improve.

The build vs buy logic behind this is the same one used for software. If you want a framework for that kind of choice, see AI build vs buy and the build, buy or integrate decision.

Common pitfalls

• Treating it as a cost-cutting exercise only. The ANA's own reasons for in-housing include brand knowledge and data control, not only price. Savings that destroy quality aren't savings.
• Bringing work in-house without an editor. AI produces plausible text that can be wrong. Google warns that generative models don't retrieve facts and that AI-generated content needs manual fact-checking. Someone has to own that.
• Ignoring the talent problem. In-house teams need people who can brief, judge and improve output. If your best marketer is stretched thin, adding AI work makes it worse.
• Demanding discounts without redefining scope. If you cut the fee but keep the same deliverables, you may get worse attention. Change the work and the fee together.
• Going straight to outcome-based pay. Without clean attribution, it's a fight waiting to happen. Start with metrics the agency controls.
• Letting agencies use AI on your account without rules. Ask what tools they use, what data goes into them and who reviews the output. If you're running lean, AI marketing for small business covers a lighter-weight version of these checks.
• Moving media buying in-house too early. ANA respondents who hadn't done it cited complexity. Don't underestimate it.
• Forgetting to measure. If you can't compare in-house and agency output on cost and results, you're guessing. The measurement side is covered in AI marketing ROI.

Key Facts: Agency vs In-House in the AI Era

• 82% of ANA members had an in-house agency in 2023, up from 78% in 2018, 58% in 2013 and 42% in 2008; the ANA predicts penetration will peak at 85 to 90% (ANA).
• In the same study, 92% of respondents still used external agencies and 65% had moved some established business from external agencies to their in-house agency (162 respondents, February to March 2023) (MediaPost).
• Cost efficiency was the primary reported benefit of in-house agencies, followed by brand knowledge, institutional knowledge and dedicated staff (MediaPost).
• MediaLink reported average creative production savings above 27% from AI, while agency pay remained mostly tied to billable hours as of May 2025 (Digiday).
• Forrester's Jay Pattissall said AI had not yet transformed the remuneration model; barriers to outcome-based pay include culture, technology, management and attribution (Digiday).
• Google says producing many pages with AI tools without adding value for users may violate its scaled content abuse policy (Google Search Central).