Fractional CMO vs agency vs in-house: what are you actually choosing?
Founders and marketing leads choosing who runs marketing next year can use this guide to pick a starting structure and a short list of things to secure before signing anyone. Labor reached 24.5% of marketing budgets in Gartner's 2026 CMO Spend Survey, so the choice steers about a quarter of the money.
Most fractional CMO vs agency vs in-house comparisons price the options by output per dollar. AI made output cheap everywhere, so that measure no longer separates them. The better test is which capabilities must build up inside your company and which you can rent without losing anything.
Our rule: keep data and approvals inside the company, and rent execution. Buy judgment from outside, for example through AI marketing strategy consulting, until you can hire and keep it.
What did AI change about the in-house vs agency math?
Production got cheap and judgment stayed expensive. Three studies published in June 2026 measured that shift from different seats.
| Study | Who was asked | Finding | What it means for your choice |
|---|---|---|---|
| Forrester with the 4As, 24 June 2026 | US marketing agencies | The study found nine in 10 use generative AI and half use agentic AI for execution. In the same study, 61% still classify AI as a "cost of business" and 31% plan to monetise agentic AI within 24 months. | Agency AI savings mostly stay with the agency unless your contract moves them |
| ANA in-housing report, 22 June 2026 | ANA survey respondents | The report found respondents five times more likely to say in-housing is growing than shrinking. In the report, 53% agree the in-house agency's primary role is upstream strategy and brand-building work. | In-house teams now see strategy as their job, with production a smaller share |
| Gartner CMO Spend Survey, as reported by Marketing Dive, 9 June 2026 | 401 CMOs in North America, the UK and Europe, most at companies with $1 billion or more in revenue | Labor rose from 21.9% of the marketing budget in 2025 to 24.5% in 2026 | Companies are paying more for people as AI spreads |
Jay Pattisall, VP and principal analyst at Forrester, put the agency risk in one line: "AI has fundamentally transformed marketing agencies, but the industry is at risk of mistaking efficiency for effectiveness."
Treat all three as direction of travel. The samples are American or European, and Gartner's leans toward very large companies. A 40-person firm in Kuala Lumpur shares the direction. It does not share the ratios.
That raises a fair objection: if agencies keep the AI savings, renting execution looks worse. It still beats an underused salaried seat, the dearest way to buy a few hours of work a week. Renting pays only when the price follows the work.
Ask for fees tied to agreed scope and business outcomes instead of hours, plus a written note of where the agency uses AI on your account. Illustrative example: if AI cuts the production on your account from 60 hours a month to 35 and the fee stays flat, you are paying for 25 hours nobody works.
What does each marketing operating model do well, and where does it break?
Each marketing operating model sells one thing well and breaks under one predictable condition. Most mid-sized companies run two of them at once.
| Model | What you are buying | Breaks when | Usually fits |
|---|---|---|---|
| In-house team (your employees) | Context that stays, and fast internal decisions | Spend is too small or seasonal to keep specialists busy, or no senior leader directs them | Steady spend, a leader in seat, data worth protecting |
| Agency (an outside firm on retainer or project terms) | Execution capacity and channel specialists on demand | Nobody on your side can brief it, challenge it or read the numbers | Clear strategy and a client lead who sets direction |
| Fractional CMO (a senior leader working part-time across a few companies) | Senior judgment at part-time intensity | The job needs daily presence or management of a large team | Early or mid-stage firms, founder-led marketing, gaps before a hire |
| Consultancy (an outside team on a bounded problem) | A redesign or decision, with capability handed over | No internal owner runs the output, or the work is really ongoing | Restructures, measurement rebuilds, AI adoption plans |
In-house is cheaper per hour only when people are busy. Take an illustrative salaried paid social advertising specialist with 9 hours of real work in a 40-hour week. Divide 9 by 40 and the seat is busy for under a quarter of the week, so each useful hour costs about 4.4 times the hourly salary rate.
Count the hours of real specialist work each channel creates in a normal week, leaving out the bid and pacing work the platforms automate. Say search creates 34 hours and paid social creates 9.
Our rule of thumb: bring a channel in-house once it produces at least 30 hours a week for two quarters in a row, and rent it below that. Here that means one search advertising hire and a rented paid-social specialist.
One hire is also one point of failure. Keep the agency on a small backup scope, or document the account well enough that someone can cover leave. As AI removes more hours, fewer channels clear the bar each year.
Agencies earn their fee on breadth. A good one has seen your problem on many other accounts, and a team of three cannot copy that pattern library. Plenty of agencies work inside the client's own GA4 and CRM, the setup behind marketing analytics services. The risk appears when only the agency reads those numbers, because then it sets its own targets.
A fractional CMO is a leadership decision. You rent judgment for a few days a month. It fails when the role quietly grows into managing six people and three vendors, which is a full-time job on a part-time contract.
A consultancy is only as good as its exit. The test is whether the company runs better a year after it leaves. The familiar failure is a strategy deck or a tech stack your team cannot run, followed by an agency hired to operate it.
We sell consultancy, so weigh this list with that in mind. Skip one when:
- you already know what to do and only lack hands;
- you cannot name the person who will run the output after handover;
- the work is continuous, like monthly content or weekly reporting;
- the budget covers the recommendation and not the change.
The economics sit in the agency model vs the AI consultancy.
Which marketing capabilities should you own, buy or rent?
Keep the data and approval layers inside and rent channel work and production. Strategy is bought until you can hire for it. The table sorts the five layers of a marketing function, with the 30-hour bar from the example above as the line for channel work.
| Layer | Verdict | What AI did to it | What to do |
|---|---|---|---|
| 1. Data, measurement and platform accounts (ad accounts, analytics, CRM, consent records, conversion definitions) | Own | Automated bidding in Google, Meta and Microsoft optimises toward the conversion signal you define | Keep every account in your company's legal name, with an employee as top admin |
| 2. Brand, claims and compliance approval | Own | More AI drafts to read | Outsource drafting if you like. An employee signs off, above all in insurance, banking, education and health |
| 3. Strategy and trade-offs | Buy, or own once you can hire and keep a strong leader | Little change: senior hours are still scarce | Fractional CMO for a standing need, consultancy for a one-off step change |
| 4. Specialist channel operation | Rent until a channel clears the 30-hour bar | Bid and budget work largely automated | Pay for oversight, not button-pushing |
| 5. Production volume | Rent or automate | Compressed hardest, and still getting cheaper | Use whichever option is cheapest this year and check again next year |
Once layers 1 and 2 are owned, the in-house vs agency choice shrinks to layers 4 and 5. That is utilisation arithmetic. Fractional CMO vs consultancy becomes a question about layer 3: a standing part-time leader, or a bounded answer for a few months. Our CMO AI implementation roadmap covers the order in which to build layer 1.
Open the admin screens for Google Ads, Meta Business Manager, your analytics property and your CRM. For each, write down who holds the top admin role and which legal entity owns the account. If any answer is "the agency" or "not sure", fix that before you change anything else.
Which marketing operating model fits your company right now?
Three questions set the main structure, starting with whether a leader is in seat. Three more add rules on top. The tree shows the logic and the toggles below apply it to your answers.
Tick every statement that is true for your company. The result updates as you go.
With nothing ticked, judgment is the gap and volume does not justify a team. Rent senior direction part-time, keep every account in your company's name, and rent execution project by project.
Illustrative heuristic. Readiness score = one point each for the first three statements (leader, steady spend, data asset); 3 of 3 points toward an in-house core. The band follows the tree. It cannot see your budget or your people.
Watch for one expensive habit: picking a model to avoid a hire. A fractional CMO kept for years because recruiting felt risky, or a broad retainer kept because nobody wants to own the brief, costs more than the hire it replaces. A company that needs a fractional CMO today may need a full-time one within 18 months. A change in your agency roster is a good moment to rerun the tree.
How do you switch marketing operating models without losing a quarter?
Most model changes fail in the handover. Someone rebuilds the campaigns instead of transferring them, and the platform's automated bidding relearns from scratch.
- Move ownership before you announce anything. List every ad account, analytics property, tag manager container, CRM, creative library and data feed, with its legal owner and top admin. Move each one to your company.
- Write down the conversion definitions. What counts as a lead or a sale, and which events feed which platform. This is the most valuable document in any transition, and it rarely exists.
- Pay for an overlap. Four to eight weeks of the incoming party shadowing the outgoing one is a reasonable default.
- Transfer live campaigns. Keep account history intact where the platforms allow, and restructure later with a measurement plan.
- Set a 90-day scorecard on business outcomes. Pipeline or revenue, and the cost of each. Activity counts do not belong on it.
- Name one internal owner per layer, even where the work is done outside.
Your market changes the hiring half of the fractional CMO vs agency vs in-house decision; the framework stays the same. Before planning an in-house team in Singapore or Malaysia, put the senior performance or analytics role through one full hiring round. If it does not close, plan a fractional or hybrid structure and try again at the next budget cycle.
In Australia and Canada, privacy law gives you a governance reason to own the data layer as well as a strategic one; confirm the specifics with your privacy adviser, as this is not legal advice. For the role-by-role version, see marketing org design when AI does half the work.
Run the Monday test this week. At your next budget cycle, count the real hours in each channel and answer the six toggles again with the new numbers.