Five ways
forward

Sometime in the next year, a client’s procurement lead will ask a fair question. If AI makes the rollout assets, why does the rollout cost the same?

Small shops have more room than the headlines suggest. Half of the 376 agency owners, consultants and freelancers SparkToro surveyed in late 2025 had grown revenue over the previous 12 months.⁴ Focus matters more than size. Promethean Research puts net margins at 19% for studios under 10 people, 12% at 10 to 24, 9% at 25 to 49 and 8% at 50 or more.⁵ Agencies that narrowed their services averaged about 30%, a figure it calls directional.⁵ For a 30-person shop in that squeezed middle, the move is to narrow.

What an AI-native shop looks like

An AI-native agency is built differently from the payroll up. Four traits define it.

1. A senior pod that owns judgment. A few experienced people hold the relationship, choose which idea goes to the client, push back when the client is wrong and put their names on the result.
2. Systems for the repetitive bulk. Versioning, research, first drafts and reporting run through workflows a staff member builds and tunes.
3. A memory of each client. Brand rules, feedback and results live in a system the agency maintains, on terms the client has agreed to.
4. A price per output. Per asset, rollout or month, with outcome-based fees as the long game.

The clearest template comes from law. At Crosby, lawyers and engineers sit at alternating desks.⁶ AI takes the first pass on each contract, a lawyer approves it, and the client pays per document while the firm tracks human review time.⁶ Most of that carries over, except the review: an agency’s senior makes a creative call rather than checking a clause, and clients pay for that call.

Sequoia Capital Deal Velocity, Not Billable Hours Play video: Deal Velocity, Not Billable Hours: How Crosby Uses AI to Redefine Legal Contracting 50 min

Picture a 15-person shop: three partners who own clients and sign off, six senior strategists and makers in two pods, two workflow builders, two producers, and two juniors who learn by reviewing system output. On a rollout, the system drafts variants from the client’s brand file, a senior keeps the best dozen and fixes three, and a partner signs. The client pays per rollout. Senior hours go to ideas and client conversations, and the versioning grind goes to the systems.

An existing 30-person agency can get most of the way by reorganizing: seniors into pods, two or three people onto the systems, execution staff retrained to review and direct.

Pick a
direction

The ground is open: just 6% of agencies sell AI as a line of business.⁷ Five directions follow, each with a working example. A shop needs one.

The guide

Help clients put AI to work: training, adoption programs, custom agents, governance. More than half of B2B marketing leaders call data strategy and AI readiness important when choosing an agency, and few are satisfied.⁸ Time Under Tension, an Australian agency and OpenAI Services Partner, runs AI Academy courses and in July won Miele, Early Settler and Amotiv for adoption work.^9,10

Fits if

your team already uses AI daily and clients keep asking how.

First move

package your internal AI training as a paid program for your two best clients.

The toolmaker

Turn your internal tools into products. Rise and Shine and Partners more than doubled revenue in 2024, and Adweek credits its own AI platform, Black Rabbit.¹¹ basement.studio, a 35-person design and engineering studio, spun out two products, BaseHub and XMP.¹²

Fits if

you have built an internal tool clients ask about.

First move

list your internal tools and ask three clients what they would pay for one.

The brand-systems builder

Brand guidelines written for people now need a version written for machines. Focus Lab turns guidelines into voice files, design tokens and evaluation scripts, then re-tests them as models change, so the work recurs.¹³ Standards ships a machine-readable brand file that AI tools can apply.¹⁴

Fits if

you are a design studio with brand clients.

First move

turn one client’s guidelines into a voice file and token set, test them in the client’s AI tools, and price the re-testing monthly.

The productized studio

Sell the repetitive bulk at a fixed price and keep ideas as the premium. Plus Also Studios, Howatson+Company's sister studio, launched a service in March that delivers unlimited rollout assets for a campaign at one flat fee of AU$5,000, ordered through a portal and returned in 72 hours.^15,16 Document the human decisions in each asset: after Thaler v. Perlmutter, purely AI-made work cannot be copyrighted in the US.¹⁷

Fits if

you have high-volume rollout or adaptation work.

First move

take the most repeated deliverable on last year’s invoices, fix its scope and give it one price.

The senior specialist

Go narrow and senior, and sell in sprints. X&O, 10 people or fewer, made Adweek’s 2026 Fastest Growing list selling strategy sprints instead of accounts.¹⁸ JK, launched in February by two TBWA alumni with Hitachi as founding client, runs a lean core with freelancers and wants off time-and-materials.¹⁹ For scale, networks such as Worldwide Partners and IKPN let independents pitch together.²⁰

Fits if

there is one practice clients keep rehiring you for.

First move

name that practice and list the services you would stop selling to own it.

What the next five years pay for

Recurring revenue steadies a business whether or not it is ever sold, and AI capability draws buyers: R/GA’s first acquisition in nearly 50 years was Addition, a 12-person AI studio.²¹

Price is where the gains are kept or lost. S4 Capital cut headcount about 11.5% in 2025 while its like-for-like revenue fell 8.7%,²² the pattern to avoid. Blair Enns, founder of Win Without Pitching, argues AI may finally end labor-based pricing, since a fee that tracks effort shrinks as work speeds up.²³ Only 17% of large advertisers in a WFA study still pay by labor as standard,²⁴ but a fee built from a rate card is still an hourly price, and gets discounted the same way at renewal.

For now there is a window. A vendor survey of 174 agencies found client requests for AI discounts flat over six months.²⁵ Of 400 client decision-makers in a 2026 survey, mostly at smaller organizations, 79% agreed that if AI cuts the time required, they expect fees to fall, though only 29% expect their agencies to cut fees because of AI in the next year.²⁶ The pressure is waiting for renewal.

Start with per-output or subscription pricing. Outcome-based fees will take longer: WPP’s chief executive says they are a few years away, with Jaguar Land Rover its only client on them.²⁷ A workable sequence: reprice at the next renewal, build client memory and workflows over two or three years, and reach year five with the system as the asset.

If you could not sell a single hour next year, what would your clients still pay you for?

What to design around

Execution-heavy work is the first target. This month the AI-agent company Enso announced plans to buy the performance agency PPG Digital and run it on its agents.²⁸ General Catalyst, which buys service firms to rebuild them around AI, has avoided digital ad and PR agencies because of high client churn.²⁹ Don’t count on that reprieve lasting.

The harder problem is the missing junior rung. Workers aged 20 to 24 held 10.5% of US advertising and PR jobs in 2019 and 6.5% in 2024.³⁰ Stanford research on AI-exposed jobs traces early-career declines mainly to fewer hires,³¹ and AI now does work that junior hires used to do.²⁵ A shop that stops hiring them will have no one to promote in the 2030s and will buy its seniors from competitors who kept training. One place to start: each junior owns a workflow and reviews system output before a senior does, with that time priced into the fee.

Most of the work ahead is editing. Pick one direction. Cut the services that don’t serve it, even ones that pay well this year. Then decide on purpose what the people in the room are for, and build the rest around that.

Hours were always the easiest thing to count, never the thing worth buying.