// RESEARCH / AGENCY OPERATING MODELS
2027 Marketing Agency Research: 50 Brands, One Harder Test
Every agency has an AI platform now. The question is whether the platform changes anything the client can inspect.
This research reviews public material from 50 agency brands and operating businesses across networks, consultancies, scaled independents and specialists. It asks a narrower question than “who has the best AI.” Which agencies publicly show a system that changes a client decision, the evidence governing it, the client’s control of it or the economics around it?
The answer is not a ranking. It is a baseline for a repeatable 2027 test.
What this research is and is not
This is a purposive public-source study reviewed September 17, 2026. It does not represent the full agency market, expose negotiated scopes of work or establish private capability. Strong public positioning can inflate apparent maturity. Private systems, client terms and operating practices can be missed.
The research records what was demonstrated in the material reviewed. “Not demonstrated” does not mean absent. It means the public sources did not provide usable evidence for that field during this pass.
Every result stays in its evidence lane. A product page can show that an agency claims a capability. A named case can show that a deployment and reported result exist. A filing can show company performance. None of those evidence types automatically proves that a named platform caused a business result.
| Evidence code | Meaning | What it supports | What it cannot establish |
|---|---|---|---|
| C | Agency claim or product description | The capability was publicly described | Adoption, causal impact or contract reality |
| K | Named client case or disclosed deployment | A reported deployment and result exist | Causality without a disclosed design |
| F | Filing, audited statement, client statement or independent evidence | A financial, audited or independently supported fact | That one platform caused the outcome |
| ND | Not demonstrated in the public material reviewed | The research did not find usable public proof | That the capability or right is absent |
The working question
The weak question is whether an agency uses AI. Forrester and 4As report that nine in ten U.S. agencies use generative AI. The public release does not provide its full sample or methodology, so the statistic needs caution. It still points to a useful conclusion: adoption alone cannot sort the market. Forrester and 4As describe AI as a cost of doing business for many agencies.
The stronger question is whether an agency’s technology changes six things:
- the client decision it is responsible for improving;
- the evidence allowed to govern that decision;
- the authority granted to the agency or its software;
- the client’s right to inspect, export and continue using the work;
- the measurement and counterfactual used to judge the result;
- the commercial agreement.
That framework separates three very different businesses that are often sold under the same AI platform pitch. Production systems reduce the time required to make work. Coordination systems connect people, channels and approvals. Decision systems define an objective, guardrails, evidence, decision rights, later measurement and retained learning. Each can be valuable. They are not interchangeable.
How the 50 brands were reviewed
The scan records twelve fields for every brand: positioning; named system; data and client-system connection; functional scope; measurement; evidence hierarchy; decision rights; client ownership and portability; delivery and commercial model; technical-builder evidence; named cases; and independent or financial evidence.
The sample keeps parent ownership visible. A network subsidiary can have a distinct public offer, but shared infrastructure is not treated as a separate market confirmation.
Networks, holding companies and consultancies
- Publicis Groupe
- WPP
- Omnicom
- Havas
- dentsu
- Stagwell
- Accenture Song
- Deloitte Digital
- IBM iX
- Capgemini / frog
Scaled independents, listed groups, digital and performance agencies
- Monks
- DEPT
- Power Digital
- Tinuiti
- PMG
- Wpromote
- Horizon Media
- Goodway Group
- Croud
- Brainlabs
- NP Digital
- Kepler
- Assembly
- Basis Technologies
- Known
- GALE
- Bounteous
- Jellyfish
- R/GA
- AKQA
Midsized and specialist agencies
- Directive
- Seer Interactive
- NoGood
- Tuff
- KlientBoost
- Disruptive Advertising
- WebFX
- Ignite Visibility
- Single Grain
- DAC
- Rise Interactive
- Adlucent
- M+C Saatchi Performance
- VaynerMedia
- Huge
- Code and Theory
- Critical Mass
- Razorfish
- iProspect
- Merkle
The full standardized scan records the public product claim, evidence state and caveat for each company. It supports six bounded findings. Platform language appears in every segment. Public proof is much richer for product functions and reported campaign results than for client control, decision rights, portability or commercial terms. Causal language has spread faster than public causal-method detail. Scale is not a reliable proxy for adaptation. Client-operated systems appear in a meaningful minority of cases. Finally, agencies with excellent reputations can expose limited public evidence of a proprietary decision system. That is a public-evidence finding, never a verdict on their private work.
Evidence record for the 50-brand scan
This is the public record behind the scan. Each entry identifies one public source already captured in the working ledger or standardized scan, the evidence code applied to the material, and the boundary on what that material can support. A direct source URL appears only where the canonical research already carries one. This record does not add a maturity score, infer private capability or turn a company case into proof of causality.
Publicis Groupe
- Source reviewed
- CoreAI and 2025 Universal Registration Document
- Evidence code
- C/F
- What the source shows
- Publicis describes CoreAI across identity, media, content and technology assets.
- What it does not establish
- Client portability, platform pricing or that CoreAI caused company performance.
- Review date
WPP
- Source reviewed
- WPP Open
- Evidence code
- C
- What the source shows
- WPP publicly describes WPP Open, managed and self-service use, and data collaboration.
- What it does not establish
- Negotiated decision rights, portability or platform-specific client outcomes.
- Review date
Omnicom
- Source reviewed
- Omni
- Evidence code
- C
- What the source shows
- Omni is presented as an integrated data, creativity, media, CRM, commerce and performance system.
- What it does not establish
- Client control, contract terms or the effect of the combined post-IPG platform.
- Review date
Havas
- Source reviewed
- Converged announcement
- Evidence code
- C
- What the source shows
- Havas describes Converged as connecting intelligence, content, production and measurement.
- What it does not establish
- Named causal client results or client-control terms.
- Review date
dentsu
- Source reviewed
- Merkury and The Trade Desk
- Evidence code
- C/K
- What the source shows
- dentsu names a Merkury deployment for addressability and closed-loop measurement.
- What it does not establish
- Commercial-model change or the causal design behind reported sales lift.
- Review date
Stagwell
- Source reviewed
- AI products and platforms
- Evidence code
- C
- What the source shows
- Stagwell publicly markets AI products and a separate Marketing Cloud business.
- What it does not establish
- That product revenue or company growth came from one system.
- Review date
Accenture Song
- Source reviewed
- Marketing Investment Navigator
- Evidence code
- C/K
- What the source shows
- Accenture describes a measurement product that combines MMM, attribution, lift measures, Amazon signals and scenario work.
- What it does not establish
- Causality behind reported deployment results or public commercial terms.
- Review date
Deloitte Digital
- Source reviewed
- Agentic Orchestration Engine
- Evidence code
- C/K
- What the source shows
- Deloitte describes a marketing-lifecycle system intended to work inside client technology.
- What it does not establish
- Public portability terms or business outcomes from that engine.
- Review date
IBM iX
- Source reviewed
- METRO case
- Evidence code
- K
- What the source shows
- IBM iX and METRO describe a multichannel marketing platform and standardized measurement.
- What it does not establish
- Commercial terms or a general causal result for IBM iX systems.
- Review date
Capgemini / frog
- Source reviewed
- Connected Marketing
- Evidence code
- C/K
- What the source shows
- Capgemini describes cross-functional marketing, sales, service and commerce work.
- What it does not establish
- Client ownership terms or a uniform causal method.
- Review date
Monks
- Source reviewed
- Marketing orchestration
- Evidence code
- C/K
- What the source shows
- Monks describes orchestration from insight through delivery and performance.
- What it does not establish
- Ownership of workflow logic, data transformations or decision history.
- Review date
DEPT
- Source reviewed
- Agentic AI and MCP case
- Evidence code
- C/K
- What the source shows
- DEPT publicly documents a client-facing agentic data-consumption product.
- What it does not establish
- Agency-wide measurement, portability or commercial boundaries.
- Review date
Power Digital
- Source reviewed
- nova
- Evidence code
- C/K
- What the source shows
- Power Digital describes a Snowflake-connected system for first-party and advertising data.
- What it does not establish
- Underlying causal design, ownership or pricing terms.
- Review date
Tinuiti
- Source reviewed
- Bliss Point
- Evidence code
- C
- What the source shows
- Tinuiti presents Bliss Point with audience, creative, media and measurement integrations.
- What it does not establish
- Client control, evidence precedence or commercial-model detail.
- Review date
PMG
- Source reviewed
- Alli and technology
- Evidence code
- C/K
- What the source shows
- PMG describes shared client access, governance and measurement around Alli.
- What it does not establish
- Public pricing or a platform-level causal result.
- Review date
Wpromote
- Source reviewed
- Polaris IQ
- Evidence code
- C
- What the source shows
- Wpromote describes data, media, creative, incrementality and MMM through Polaris IQ.
- What it does not establish
- Client ownership, contract terms or independently verified outcomes.
- Review date
Horizon Media
- Source reviewed
- NEON retail-media measurement
- Evidence code
- C/K
- What the source shows
- Horizon describes standardized retail-media measurement across networks.
- What it does not establish
- Portability, pricing or independent outcome validation.
- Review date
Goodway Group
- Source reviewed
- GOES
- Evidence code
- C
- What the source shows
- Goodway describes outcome engineering, scenario planning, MMM and exports.
- What it does not establish
- Responsibility for modeled outcomes or named causal case evidence.
- Review date
Croud
- Source reviewed
- Croud and ECI partnership
- Evidence code
- C/K
- What the source shows
- Croud describes planning, workflows and execution through Croud Control.
- What it does not establish
- That technology alone caused company growth or public client-control terms.
- Review date
Brainlabs
- Source reviewed
- Marketing measurement methods
- Evidence code
- C/K
- What the source shows
- Brainlabs assigns MMM, incrementality and attribution different decision jobs.
- What it does not establish
- Client control or ongoing commercial terms.
- Review date
NP Digital
- Source reviewed
- Agency profile
- Evidence code
- C
- What the source shows
- NP Digital publicly presents search, content, media and technology capabilities.
- What it does not establish
- A client-owned decision system or verified platform-specific outcomes.
- Review date
Kepler
- Source reviewed
- Kip
- Evidence code
- C/K
- What the source shows
- Kepler publicly names Kip and its analytics, commerce and workflow scope.
- What it does not establish
- Licensing, portability or detailed commercial terms.
- Review date
Assembly
- Source reviewed
- STAGE
- Evidence code
- C/K
- What the source shows
- Assembly describes a planning and media system drawing on more than 400 sources.
- What it does not establish
- Client-control terms or independently verified performance.
- Review date
Basis Technologies
- Source reviewed
- Basis platform view
- Evidence code
- C/K
- What the source shows
- Basis describes omnichannel planning, buying and measurement and states transferability during agency changes.
- What it does not establish
- That the product’s reported outcomes are independently causal.
- Review date
Known
- Source reviewed
- Skeptic
- Evidence code
- C/K
- What the source shows
- Known publicly offers Skeptic and names geo and holdout testing.
- What it does not establish
- That public ROI claims have independent causal reconstruction.
- Review date
GALE
- Source reviewed
- Alchemy products
- Evidence code
- C
- What the source shows
- GALE describes a system connecting transactions, identity, creative and media.
- What it does not establish
- Portability or attributable client results.
- Review date
Bounteous
- Source reviewed
- Client enablement
- Evidence code
- C/K
- What the source shows
- Bounteous describes client enablement and independently operable data collaboration.
- What it does not establish
- A universal operating model or causal attribution for every case.
- Review date
Jellyfish
- Source reviewed
- Creator intelligence
- Evidence code
- C/K
- What the source shows
- Jellyfish describes creator, search, media and AI-discovery intelligence.
- What it does not establish
- Client ownership, pricing or independent outcome proof.
- Review date
R/GA
- Source reviewed
- AI Products team
- Evidence code
- C/K
- What the source shows
- R/GA publicly documents a product-building AI team and shipped tools.
- What it does not establish
- A uniform enterprise decision layer or client-control terms.
- Review date
AKQA
- Source reviewed
- AKQA
- Evidence code
- C/ND
- What the source shows
- AKQA publicly shows design, innovation, brand and product work.
- What it does not establish
- A distinct public decision system, portability or pricing model.
- Review date
Directive
- Source reviewed
- Stratos
- Evidence code
- C
- What the source shows
- Directive describes CRM, customer voice, media and finance framing around revenue allocation.
- What it does not establish
- Ownership, contract structure or independent validation.
- Review date
Seer Interactive
- Source reviewed
- AI skills
- Evidence code
- C/K
- What the source shows
- Seer documents reusable AI skills, source traceability and human review.
- What it does not establish
- Client portability or independently verified savings.
- Review date
NoGood
- Source reviewed
- NoGood
- Evidence code
- C
- What the source shows
- NoGood publicly describes growth squads, proprietary systems and agents across lifecycle work.
- What it does not establish
- Contract, export or independent performance detail.
- Review date
Tuff
- Source reviewed
- First 90 days with Tuff
- Evidence code
- C/K
- What the source shows
- Tuff describes an embedded growth-team and experimentation model.
- What it does not establish
- A proprietary platform, portability or independently validated cases.
- Review date
KlientBoost
- Source reviewed
- KlientBoost
- Evidence code
- C/K
- What the source shows
- KlientBoost publicly presents paid media, creative, CRO and lifecycle work with cases.
- What it does not establish
- A comparable operating platform, client-control terms or system-level proof.
- Review date
Disruptive Advertising
- Source reviewed
- Strategy-led service
- Evidence code
- C/K
- What the source shows
- Disruptive Advertising publicly describes managed media, lifecycle, creative and analytics work.
- What it does not establish
- Causal authority rules, portability, pricing or sufficiently disclosed methods.
- Review date
WebFX
- Source reviewed
- RevenueCloudFX
- Evidence code
- C/K
- What the source shows
- WebFX exposes detailed product tiers, integrations, seats and limits around a revenue platform.
- What it does not establish
- Causal reconstruction behind its headline revenue totals.
- Review date
Ignite Visibility
- Source reviewed
- RevIntel
- Evidence code
- C
- What the source shows
- Ignite describes CRM outcomes, ads, local and franchise marketing through RevIntel.
- What it does not establish
- Client data rights, exports, causal design or independent result evidence.
- Review date
Single Grain
- Source reviewed
- Single Grain
- Evidence code
- C
- What the source shows
- Single Grain says it deploys proprietary systems inside client accounts.
- What it does not establish
- Named architecture, ownership, causal measurement or commercial terms.
- Review date
DAC
- Source reviewed
- IRIS
- Evidence code
- C/K
- What the source shows
- DAC describes cross-channel work spanning local, media, CRM, analytics and business data.
- What it does not establish
- Portability or a general causal conclusion from named cases.
- Review date
Rise Interactive
- Source reviewed
- Rise Interactive
- Evidence code
- C/ND
- What the source shows
- Rise publicly describes cross-channel media and analytics services.
- What it does not establish
- Current technical detail, client control, platform pricing or stronger system evidence.
- Review date
Adlucent
- Source reviewed
- Adlucent
- Evidence code
- C/ND
- What the source shows
- Adlucent publicly describes retail, paid search and commerce-signal work.
- What it does not establish
- Proprietary-system ownership, portability, pricing or cross-enterprise decision proof.
- Review date
M+C Saatchi Performance
- Source reviewed
- OneView
- Evidence code
- C/K
- What the source shows
- M+C Saatchi Performance describes always-on MMM, planning and creative-system work.
- What it does not establish
- Ownership, pricing or independent causal proof.
- Review date
VaynerMedia
- Source reviewed
- VaynerMedia reports
- Evidence code
- C/ND
- What the source shows
- VaynerMedia publicly shows social, creative, media and commerce expertise.
- What it does not establish
- A client-operable proprietary decision system or contract evidence.
- Review date
Huge
- Source reviewed
- CoinTracker case
- Evidence code
- K
- What the source shows
- Huge documents client-specific growth and product work with reported outcomes.
- What it does not establish
- That the reported outcomes result from a single system or a general agency model.
- Review date
Code and Theory
- Source reviewed
- AI systems
- Evidence code
- C/K
- What the source shows
- Code and Theory publicly documents product, engineering and content-system capacity.
- What it does not establish
- Client ownership terms or a comparable platform-wide decision layer.
- Review date
Critical Mass
- Source reviewed
- Critical Mass
- Evidence code
- C/ND
- What the source shows
- Critical Mass publicly demonstrates digital experience, product and commerce delivery.
- What it does not establish
- A distinct public decision system, portability or pricing model.
- Review date
Razorfish
- Source reviewed
- Beta Lab
- Evidence code
- C/K
- What the source shows
- Razorfish publicly names Beta Lab and connected marketing capabilities.
- What it does not establish
- Client-control terms, pricing or system-level independent evidence.
- Review date
iProspect
- Source reviewed
- Performance marketing
- Evidence code
- C/K
- What the source shows
- iProspect publicly describes first-party data, media, commerce and performance marketing.
- What it does not establish
- Platform independence, ownership or commercial detail.
- Review date
Merkle
- Source reviewed
- JYSK case
- Evidence code
- K
- What the source shows
- Merkle documents client infrastructure and a named reported business outcome.
- What it does not establish
- Causal attribution for one platform or every Merkle engagement.
- Review date
Twelve operating models that show the real differences
Twelve profiles create enough contrast to see the underlying business models without turning the scan into a leaderboard.
| Model | What the public evidence shows | What remains unresolved |
|---|---|---|
| Publicis | CoreAI links identity, media, content and technology assets in an integrated organization | Public client ownership and platform-specific commercial terms |
| WPP | WPP Open spans strategy, creative, production, media, commerce and measurement; WPP states client-environment deployment and outcome-linked remuneration principles | Negotiated portability, decision authority and platform-specific client impact |
| Omnicom | Omni brings Acxiom identity, Flywheel commerce, media, CRM and performance together | Control, portability and the effect of the post-IPG combined platform |
| Accenture Song | Marketing Investment Navigator combines MMM, attribution, sales lift, brand lift, Amazon signals and scenario testing | Causal design behind the reported deployment results and public commercial terms |
| Monks | Monks.Flow connects production, performance and orchestration; it states clients own creative assets built on their platforms | Ownership of workflow logic, transformations and decision history |
| DEPT | A named Model Context Protocol client build demonstrates product and integration capacity | Public measurement, portability and commercial boundaries at scale |
| Tinuiti | Bliss Point describes 70-plus integrations, econometrics and human decision authority | Client control, evidence precedence and commercial-model detail |
| Power Digital | nova joins first-party and ad data with Snowflake, incrementality and contribution economics | Underlying causal design and ownership terms |
| PMG | Alli presents shared access, enterprise governance and independent measurement language | Public pricing and system-level causal result detail |
| Brainlabs | InsightMix, MMM, incrementality and attribution are assigned different decision jobs | Public client-control and ongoing operating terms |
| WebFX | RevenueCloudFX makes tiers, seats, limits and add-ons more visible than most agency platforms | Public causal reconstruction behind headline revenue totals |
| Directive | Stratos connects CRM, customer voice, media and finance framing around capital allocation | Ownership, contract structure and independent validation |
These models describe five distinct businesses. Networks are integrating huge owned assets. Consultancies can alter client infrastructure. Scaled independents coordinate delivery through proprietary systems. Productized agencies expose software-like packaging. Builders leave behind client-operated products. Those categories often use the same nouns. The operating and commercial boundaries differ.
Buyer pressure is rising faster than contract clarity
The buyer side has a reason to ask harder questions. Gartner’s 2026 CMO Spend Survey covered 401 CMOs and marketing leaders, primarily from companies with more than $1 billion in annual revenue. Respondents allocated 15.3% of their marketing budgets to AI, only 30% described readiness as mature or fully developed, and paid media reached 31.4% of budget. Gartner says agency budget cuts helped fund that spend.
Agency Core’s 2026 study included 579 agency leaders and 400 clients. Ninety-one percent of clients said their agency helps them succeed. Forty-two percent said they planned to reduce the relationship within a year. Agency Core reports intention and perception, not proof that AI changed retention. The tension remains useful: client satisfaction does not automatically make a commercial model durable.
Procurement is moving ahead of measurement. The WFA reports that 70% of procurement leaders say AI has changed agency conversations, 34% have begun adapting remuneration and 57% plan to. Three in four lack KPIs for AI’s effect. The public summary does not state a respondent denominator, so its figures are directional. WFA should be read alongside the ANA’s public September 2026 note that no respondent to its 2025 agency-compensation study reported a significant AI effect on compensation agreements.
Platforms are absorbing the obvious work
| Platform | Public capability | Agency work under pressure | Remaining question |
|---|---|---|---|
| First-party data, Analytics, DV360, Meridian, GeoX, bidding and budget functions | Planning, activation, reporting and measurement | Who independently arbitrates cross-platform evidence and margin? | |
| Meta | Automated audience, creative, delivery and incrementality tools | Manual optimization and creative variation | Meta performance claims are company-reported and the platform grades its own inventory |
| Salesforce | Marketing agents, shared data, campaigns, goals, budgets and guardrails | CRM-to-campaign workflow | Some capabilities were announced or in pilot when reviewed |
| HubSpot | Marketing, sales, service, CRM context and agents | Midmarket operating workflows | It cannot supply every outside platform signal or business constraint |
| Adobe | Content, approvals, activation, performance and brand controls | Content production and workflow coordination | Agency value must exceed vendor configuration |
| Amazon | Commerce signals, AMC, DSP, Ads Agent and an advertising Model Context Protocol server | Retail-media insight and activation | Its MCP server was announced as open beta; Amazon is also inventory owner |
| Snowflake | Governed data, clean rooms and agent infrastructure | Agency-owned data middleware | Infrastructure still needs methods, choices and adoption |
The opportunity left for agencies is not clicking a vendor’s new agent button. It is work that crosses systems and incentives: reconciling platform attribution with finance, testing incrementality, connecting sales quality to marketing allocation, defining escalation rules and preserving a client-owned decision record. A Department of Justice remedy summary requiring Google data export and rival integrations in 2026 shows why portability and access have practical consequences.
The missing contract layer
Buyer guidance is already more concrete than many agency platform pages. ISBA’s Creative Services Framework added generative-AI clauses. Its proprietary-media guidance calls for granular data, governance and oversight. ISBA and the ANA’s principal-media study turn opacity into a contract issue rather than a branding debate.
A serious agency agreement should make twelve items inspectable:
- the named decision and business owner;
- read and write authority, approval thresholds and a kill switch;
- the objective and guardrails;
- evidence precedence when systems disagree;
- a baseline and counterfactual;
- data, prompt, output, audience and experiment-design rights;
- provider retention, training, substitution and cross-border rights;
- model version, retrieved evidence, tool-call and approval traceability;
- export, transition and deletion rules;
- principal-media, preferred-vendor and resale conflicts;
- fees for software, integration, operation, judgment and usage;
- failure, rollback, liability and exit.
The IPA’s 2026 Pricing Playbook covers retainers, commissions, subscriptions, project fees, performance fees, equity and hybrids. It warns that outcome pricing needs stable scope, credible measurement and agreed risk. IPA supports a modular view of agency pricing, not a blanket claim that AI makes every fee outcome-based.
Evidence that narrows the thesis
This research includes failure cases because platform adoption is not proof of client value.
WPP’s 2025 annual report described a 5.4% like-for-like decline in revenue less pass-through costs, £939 million in adjusting items, roughly $1.2 billion, organizational complexity and inconsistent execution. WPP does not show that WPP Open caused those results. It shows that platform adoption, organizational integration, financial performance and client value are separate variables.
The ANA found that 90% of qualified respondents worried whether principal-media recommendations were in their interest, while only 57% had governing guidelines. DAC’s public ChatGPT advertising test reported booked leads at roughly 100 times its typical cost and CPMs 8.7 to 9.5 times its benchmarks. DAC is a useful negative test, not a universal verdict on a new channel.
Promethean Research’s 2026 survey of 119 agency leaders found value-pricing use fell from 31% to 18%, with lower average growth among agencies using it. The study is cross-sectional and cannot identify the cause. Promethean also found service reducers outgrew expanders in its sample. A narrow specialist can beat a broad system.
The counterexamples are clear. Strong agencies may thrive without public proprietary systems. Major vendors may own the decision loop before agencies do. Client ownership can shrink future scope. Standard tools may be sufficient for many buyers. Contract change may lag past 2027. Those facts refine the thesis: proprietary technology earns its place only when it reconciles contested evidence, encodes a repeatable method, connects context the standard stack lacks or leaves a client-owned learning asset behind.
Repeat protocol for 2027
The scan becomes more useful if it can fail. A repeat review will test whether announced products became named deployments, client outcomes replaced productivity claims, control and offboarding terms became explicit, commercial structures changed and AI branding became ordinary.
Five public thresholds make the forecast falsifiable. At least 26 of the 50 brands would need to lead with a named decision or constraint before an AI label. At least 15 would need to disclose two or more client-control terms. At least 10 would need a distinct paid offer for independent measurement. At least 10 agency or buyer documents would need to show a modular pricing structure. At least 15 named cases would need to leave a client-operated workflow, experiment library, measurement system or decision history behind.
Frequently asked questions
What does “not demonstrated” mean?
It means the reviewed public material did not provide usable evidence for a field. It does not mean the agency lacks the capability or right in private work.
Is this a ranking of marketing agencies?
No. The research does not assign maturity scores or winners. It records evidence types and limits so agency claims, named cases and independent evidence are not treated as the same proof.
What makes an agency AI platform defensible?
A stronger platform helps improve a named decision, defines which evidence governs, connects context the standard stack lacks, provides real client control and leaves a usable learning asset behind.
What should a client own when an agency builds marketing technology?
The client should have explicit rights to its data, outputs, decision history, export format, transition support and deletion confirmation. Ownership needs to cover the useful operating record, not only raw data.
Will AI replace marketing agencies?
It will replace production work and platform tasks. Agencies retain a role where clients need independent cross-platform evidence, business context, change management and accountable decision-making.
Will marketing agency pricing move to outcome-based fees in 2027?
Some agreements will add limited performance components. The broader movement is likely to be modular pricing across maintained technology, integration, operation, judgment and carefully bounded shared risk.
Source record
The evidence record uses public product, investor, client-case, regulatory and association material already captured in the working ledger and 50-brand scan. Direct links appear only where the canonical research already carries the source. The record does not add links from memory, infer a capability from a missing page or treat a source’s marketing copy as independent proof. Related analysis: Every Agency Has an AI Platform Now. Show Me the Contract. For adjacent operating context, read The 2027 Marketing Stack, AI Governance for Commercial Teams, The AI Audit Log and AI Pricing Has to Respect Compute Reality.