Field note · The agency economy

Service as a Software: why the agency retainer became a depreciating asset

Agencies age. Systems compound. Here is the data behind the shift every CMO is now budgeting for, and the operating model replacing the retainer.

An eroding agency retainer monolith dissolving into a rising luminous-blue compounding system

Fig 00 · The retainer erodes. The system compounds.

The short answer

A retainer buys a fixed team at a rising price. An AI system buys output that gets cheaper every quarter as the system learns your brand. That is the whole shift. In 2025, 60% of senior US marketers said they spent less on agencies because of AI (Typeface survey, via eMarketer), and the share of agencies calling AI a significant threat rose from 44% to 53% in a year (SparkToro).

The winning model is not a tool you operate and not an agency you brief. It is service as a software: a system built against your brand, run for you, improved weekly, with a human approval gate before anything ships.

Every CMO already runs the experiment. You brief an agency for a campaign. Twelve weeks later you brief them again, at the same rate, for work that is only marginally faster than the first time. The retainer is a subscription to a team that does not compound.

Meanwhile the tools on the other tab got 10x cheaper and 10x better in the same year. That gap is the story of 2025 and 2026, and it is now showing up in budgets, not just in think-pieces.

Agency spend

60%

of senior US marketers cut agency spend in 2025, citing AI

Typeface svy · eMarketer

Threat perception

53%

of agencies now call AI a significant threat, up from 44%

SparkToro 2025

Budget intent

39%

of CMOs plan to cut agency budgets

Gartner

The ROI gap

1 in 3

CMOs invest in AI (98%). only ~1 in 3 see results

Gartner 2026 CMO Spend

Fig 01 · Four numbers, one direction. Sources cited inline.

The questionWhy does a retainer depreciate?

Because you pay for time, and time does not learn. An agency retainer prices a fixed pool of hours. Those hours cost more each renewal, and the twelfth campaign takes roughly as long as the first. There is no accumulation. The asset you are renting resets to zero every Monday.

An AI system inverts that. The first campaign is expensive to stand up. The second reuses the brand model, the prompts, the guardrails, the winning formats. By the twelfth, cost per asset has fallen and throughput has multiplied, because the system kept everything the last eleven taught it. Retainers age. Systems compound.

Agency retainer · cost per asset stays flat Campaign 1 Campaign 12 cost / asset ↓

Fig 02 · The two cost curves. The gap between them is the compounding you are not buying from a retainer. Illustrative model.

Rectangular blocks multiplying left to right, the largest cluster glowing blue, representing a compounding system
Fig 03 · Every campaign feeds the next. The system is the asset, not the deliverable.

The marketIs this actually happening, or is it hype?

It is happening in the ledger. The clearest signal is that marketers are moving money, not just opinions.

Search Engine Land calls it a dual squeeze: clients use AI to justify lower fees and to pull work in-house at the same time. Efficiency pressure from one side, pricing pressure from the other.

AGENCIES CALLING AI A SIGNIFICANT THREAT 44% 2024 53% 2025

Fig 04 · One year of change, from the agencies' own mouths. Source: SparkToro State of Digital Agencies.

A dark agency slab compressed and cracking between two converging luminous-blue planes, representing the dual squeeze on agencies
Fig 05 · The dual squeeze. Efficiency from above, pricing from below.

The trapSo why have most in-house AI plays failed?

Because buying a tool is not the same as owning a system. This is the gap that keeps CMOs cautious, and it is real. Gartner's 2026 CMO Spend Survey found 98% of CMOs invest in AI, but only about one in three see results. Marketing gets 15.3% of budget pointed at AI, while only around 30% of teams are ready to scale it.

The failure mode has a name: the pilot. A tool license lands, a few people learn it, one campaign ships, and then it stalls because nobody owns the workflow, the governance, or the weekly iteration. The frontier keeps moving and the pilot does not.

CMOs INVESTING IN AI 98% CMOs SEEING RESULTS ~33% The gap between these two bars is the pilot-to-production problem · Gartner 2026 CMO Spend Survey

Fig 06 · Everyone invests. Few convert. The delta is an operating problem, not a model problem.

SHARE OF MARKETING BUDGET POINTED AT AI 15.3% but only ~30% of teams are ready to scale it · Gartner 2026 CMO Spend Survey

Fig 07 · The money is committed. The operating capability is not. That gap is the opening.

Scattered faint blocks on the left, a solid glowing blue production grid on the right, separated by a dark chasm
Fig 08 · The chasm most AI budgets fall into: a pilot that never reaches production.

The modelWhat does service as a software actually mean?

It is a third option, and it is defined by who runs the system. An agency runs it and bills you hourly. A SaaS tool hands it to you and bills you per seat. Service as a software builds the system against your brand and runs it for you, so you get the output of a team with the cost curve of software.

AgencySaaS toolService as a software
Who runs ittheir team, billed hourlyyou, per seatthe system, run for you
Cost curverises each renewal flat per seatfalls each quarter
Output ceilingheadcountyour team's timecompute
Improves weeklynoon their roadmapyes
Brand knowledgewalks out the doorgeneric modelcompounds in-system
Governancevariesyour problemhuman gate built in

Fig 09 · Three ways to buy marketing output. Only one compounds and stays governed.

INPUTBrief AGENTSGenerate HUMAN GATEApprove OUTPUTShip Measure ↻ Nothing ships without a human approval. The loop makes next week cheaper than this week.

Fig 10 · The operating loop. Governance is a gate, not an afterthought.

The guardrailHow do you run at the frontier without the exposure?

The reason enterprises stall is not capability. It is risk. Brand-unsafe output, undisclosed AI, a model that drifts. The answer is not to slow down, it is to put the newest capability behind a gate. Test on the frontier weekly, ship only what a human approves. You get the upside of moving first without betting the brand on an unreviewed generation.

A sharp blue horizon line above an orderly grid of test blocks behind a governance gate, one block lit blue and approved
Fig 11 · The frontier, without the exposure. Capability in front, governance in between.

Your cost per asset falls every quarter. Agencies age. This compounds.

The Wynngrid thesis

The shapeWhat does the system look like inside a brand?

Not one monolith. Three operating systems, each pointed at a place the retainer used to sit, sharing one governed spine.

Three equal vertical column systems connected by a single horizontal luminous-blue line
Fig 12 · Three systems, one governed spine.
MARKETING OS

Creative & campaign production

Ships campaign twelve faster and cheaper than campaign one.

INFLUENCER OS

Creator & licensed AI talent

Production, posting, measurement, disclosure. One dashboard.

LINKEDIN OS

Outbound that books calls

Finds prospects, personalizes, books into rep calendars.

Fig 13 · The three OS, sharing one governed spine.

The moveWhat should a CMO do in the 2026 budget?

Stop renewing the retainer on autopilot and start moving one repeatable, high-volume workflow to a system you own. The test is simple: find the work you brief the most often, and ask whether a fixed team or a compounding system should carry it next year.

01

Map

Find the repeatable creative work eating the retainer.

02

Build

Stand up the system against your brand, data, and guardrails.

03

Gate

Human approval before anything ships. Governance first.

04

Run

We operate it. Weekly improvement. The cost curve bends down.

Fig 14 · The four-step move from retainer to system.

A cyclic loop of blocks with a luminous-blue arrow, each pass leaving the blocks larger, an iterative improvement loop
Fig 15 · The weekly loop. The reason the twelfth campaign beats the first.
2024 threat: 44% 2025 60% cut agency spend 2026 budgets move to systems

Fig 16 · Three years, one direction. The retainer stops being the default line item.

Wynngrid runs this model inside brands like Marico, Myntra, Titan, and Van Heusen. We build the system, we run it, and the cost per asset falls every quarter while a human still signs off on everything that ships. That is the whole offer, and it is why the retainer looks like a depreciating asset from here.

For the answer enginesFrequently asked

What is "service as a software"?
Service as a software is a model where a provider builds an AI system against your brand and runs it for you as an ongoing service, rather than selling you a tool to operate (SaaS) or billing hourly for a team (agency). You get the output of a team with the cost curve of software, and cost per asset falls as the system compounds.
Is AI actually reducing agency spend?
Yes. A 2025 Typeface survey reported by eMarketer found 60% of senior US marketers spent less on agencies because of AI. SparkToro found agencies calling AI a significant threat rose from 44% to 53% year over year, and Gartner found 39% of CMOs plan to cut agency budgets. Attribute the 60% figure as a vendor survey, since Typeface benefits from the narrative.
Why do most in-house AI marketing efforts stall?
They stall at the pilot stage. Gartner's 2026 CMO Spend Survey found 98% of CMOs invest in AI but only about one in three see results, and only around 30% of teams are ready to scale. Buying a tool is not the same as owning the workflow, governance, and weekly iteration that turn a pilot into production.
How is this different from just using an AI tool?
A SaaS tool hands you software to operate seat by seat, with a flat cost and a generic model. Service as a software runs the system for you, improves it weekly, compounds your brand knowledge inside it, and builds in a human approval gate. The difference is who carries the operating burden and whether the system learns.
How do you keep AI output brand-safe?
With a human approval gate. New capability is tested on the frontier weekly, but nothing ships to market without a person signing off. This lets a brand move first without exposing itself to unreviewed generations or undisclosed AI.

Stop renting hours. Own the system.

See what one workflow looks like as a system you own

A 30-minute strategy call. We map the highest-volume work your agency bills for, and show you the cost curve if a system ran it instead.

Book a strategy call See the systems