Wise Agentic
Campaign Managementhuman at the helm.
Governed autonomy for the age of agents
The agents have arrived in advertising. The platforms shipped them, the official connectors opened accounts to AI, and automation climbed from delivery into management. The question on every board is no longer whether to automate, but which parts, at what level of autonomy, with what controls. Wise Agentic CM answers it.
Agentic partners.
Every major platform has shipped a named agent or an ads product, and official connectors now let AI act on live accounts. However you name it, agentic advertising, AI campaign agents, autonomous campaigns, it is the same shift: software that can touch the account. We govern who and what gets to.
What it is, in one minute.
The rest of this page goes deep. If you read only one part, read this: Wise Agentic CM is our proprietary framework that lets you adopt AI agents in your campaigns safely, automating what should be automated, keeping a senior human in control, and proving the value with a record you can audit.
A safe way to use AI agents in campaigns
A framework that decides, per workflow, how much an agent may do on its own, from advising only to executing inside guardrails, with a human approving anything that moves money or targeting.
Speed without losing the account
Agents can launch and optimise far faster, but ungoverned they can burn budget, break platform rules or leak data. Most executives have already had an AI incident. Governance is what turns speed into an advantage instead of a liability.
Levels, gates, and a proof log
A recommended autonomy level for every task, a human approval gate on the sensitive ones, and the Ship's Log, a full record that proves the automation worked and keeps you audit-ready and insurable.
In one line: we are first in line to adopt automation that improves results, and we govern it so it improves them safely, with a human at the helm.
Automation just climbed from delivery into management.
For years the platforms automated the delivery layer: the auction, the targeting, the optimisation. In 2026 the agents moved up a floor, into the management layer, the part that touches the account. The question is on every board already.
of North American ad buyers are focused on agentic AI for buying and campaign execution.
IAB Outlook Study, Jan 2026platforms with named agents: Ads Agent, Ask Advisor, Copilot, Brand Agents.
Google, Amazon, Microsoft, MetaMeta's stated target for fully automated ads, from budget to creative.
WSJ, Jun 2025reported cut in campaign launch time with AI agents, and fewer human buyers.
Jellyfish / Adweek, 2025We are not against automation. We are against autonomy without governance.
"Can I just connect the AI to the account and drop the agency?" It is a fair question, so here is the straight answer. Our work was never to manage campaigns; it is to deliver business results, in the short and the long run. That is exactly why we are first in line on any automation that improves them.
"Fire your agency"
Startups and content farms sell full autonomy with zero audited case studies. The flagship, a 21M-dollar company, collapsed in 2026 after pivoting back to human-run services.
First to adopt, by design
If automation improves results, we adopt it first, as we did with smart bidding, Performance Max and feeds. Whoever leads the transformation does not fear it.
Adopted, but governed
We adopt by workflow and by case, at the right level of autonomy, with senior oversight and a framework that mitigates security, platform and legal risk.
The transformation is going to happen, and it should. Our position is simply that it happens with governance, and with us on the front line. As always.
Governed autonomy is not an opinion. It is the consensus.
The people who build the models, the platforms, the standards bodies, the insurers and the European regulator all converge on the same design: agents with controls, and a human in command. We did not invent that requirement, we operationalised it for campaigns.
The reading is blunt: more than 70% of executives have already had an AI incident, and 40% have paused ads over it (IAB and Aymara). The industry's own agentic standard embeds approval gates before any commit of spend, by design, and there are still no guarantees against prompt injection in today's models. Ungoverned autonomy is not brave, it is uninsured.
We studied the leaders, then added the one thing they all lack.
We read the agentic standards, the security research and the existing maturity models, and integrated the best of each. Then we added what none of them offers: adoption by campaign workflow.
IAB Tech Lab · AAMP
The advertising industry's agentic protocols: an agent registry, trust and transparency, and gates before any commit of spend. We adopt the protocols and the agent transparency.
Willison, arXiv, OWASP
The lethal trifecta, the anti prompt-injection design patterns, and the catalogue of agentic threats. We adopt Plan-Then-Execute and least privilege.
Anthropic, SAIF, NIST
Human control, dynamic permissions and observability, the operational controls of safe agents. We adopt them, calibrated to the risk of each action.
Adoption by workflow
Enterprise AI maturity models give generic autonomy levels, none specific to campaigns. Adoption workflow by workflow is exactly where we come in.
The gap: every piece exists as a standard, a paper or a control. No one assembles them into a framework that decides autonomy campaign workflow by campaign workflow, with a proprietary layer, and the team to run it, to govern and prove it. That is where Wise Agentic CM lives.
Five layers, from the fit assessment to governed autonomy.
A complete system, not a switch. Four execution layers built on published standards, and one proprietary layer of governed autonomy: a framework, and the senior team that runs it, that no one else packages.
Two names people mix up. Wise Agentic CM is the whole framework, the five layers together. Wise Agentic Helm is the fifth of those layers: the framework that sets autonomy levels, holds the gates and keeps the log, run by a senior team that can actually govern agents.
Agentic Fit Assessment
Decide, before executing, where AI as support, agents without money, or governed autonomy make sense.
Security & Access
Least privilege, credentials in a vault, and only platform-sanctioned paths. Read-only by default.
Senior Oversight
A senior human in the loop across campaigns, analytics, business, agentic AI and security.
Compliance & Trust
AI Act, GDPR and platform terms, operationalised into checklists and labels.
Governed Autonomy
The Autonomy Ladder, the Agentification Matrix, pre-registered gates and the Ship's Log.
The foundation, proven.
Every foundation layer rests on published standards and research, not on opinion.
01 · Sounding · Agentic Fit Assessment
Before any execution, an assessment fixes what makes sense per client, moment and part of the campaigns: maturity, risk and reversibility, security, financial exposure, geography and law, typology and platform. Each workflow leaves with a mode, from AI on demand to governed autonomy, or not automated.Evidence · Gartner: over 40% of agentic projects fail on unclear value and weak risk controls
02 · Keel · Security & Access
Read-only by default; writes only through official APIs and MCPs with their own token; credentials in a vault, MFA and per-client isolation. No browser automation of the UI, no shared credentials.Evidence · Google requires manual end-user sign-in on the API; UI automation is not a sanctioned path
03 · Crew · Senior Human in the Loop
An approver with real authority to stop (GDPR art. 22, and AI Act art. 14 human oversight), a five-competency profile, documented AI literacy (AI Act art. 4), and a hard rule: whoever proposes is not whoever approves.Evidence · OWASP names "overwhelming the human in the loop"; a junior rubber stamp is an attack vector
04 · Beacon · Compliance & Trust
Prohibited-practice screening (art. 5), transparency labels (art. 50): agents identify themselves and synthetic creative is labelled. Plus DPIA, vendor due diligence and contractual clauses.Evidence · AI Act art. 50 in force 2 Aug 2026; campaign management is transparency, not high-risk
The Keel layer is the same discipline as our Agentic Security practice, applied to live ad accounts: least privilege, hard guardrails, human gates and a kill switch. We never hand campaign management to full autonomy. There is always a senior human in the loop and a security layer under the agent, or we do not run it.
The 5th layer: the governed-autonomy engine.
Now it gets proprietary. This is the fifth layer, not a separate product you license: Wise Agentic Helm is our framework for governed autonomy, and the senior team that runs it. It assigns an autonomy level to every workflow, holds the gates that stop sensitive actions, and keeps the log that proves the value and protects the account. It starts with one ruler that everyone can read: the Autonomy Ladder.
One level per workflow
Each campaign workflow gets an autonomy level, set before adoption and reviewed with evidence. Not the whole account at once, and never at the hype's pace.
Pre-registered, like a study
A workflow's level only moves with evidence recorded in the Ship's Log, like a clinical study, not on a hunch. The rule is declared up front and frozen.
Human in the loop, always
From A0 to A4 there is a senior approver and a full record. The green marker on the ladder is the human, present at every level we execute.
We do not execute A5
A5 can emerge in edge cases: aggressive risk appetite, low-risk sectors, small budgets, data outside the EU and US. It is a market phenomenon, and it is 0% with Wise.
We govern the layer the platforms do not.
"So is Performance Max not A5?" No. Platform automation is the delivery layer. The ladder governs the management layer: who touches the account, including switching that platform automation on.
Management layer · what the ladder governs
Who touches the account: creating campaigns, moving budgets and bids, editing feeds and audiences, and activating or configuring the platforms' own automation. Every one of these is scored, gated and logged.
Delivery layer · the platform's engine
The automation inside the platform: the Performance Max auction, Advantage+ targeting, Smart+. It is Google, Meta and TikTok's product, and it runs within the budgets, caps, feeds and creatives the management layer sets.
The reading: activating Performance Max is a management action, scored, gated and logged like any other. The platforms govern their layer; we are the ones with a published model to govern the other. As the SAE levels govern the driver, not the engine.
The right level, workflow by workflow.
Not everything gets agentified, and what does, does not all go to the same level. The matrix fixes the recommended level and the gate for each type of work.
An example: reporting runs at A4 from day one; budget changes stay at A3 until 90 days of clean log, and the move up to A4 is decided with data, not belief. The matrix is the map; the Ship's Log is what lets a workflow move on it.
Approve the plan, not the click.
Four controls in series, designed on the published security patterns, because approval fatigue is itself an attack vector. The agent proposes a full plan; a human approves it at plan level; execution runs only on sanctioned paths; every action and approval is logged, immutably.
The golden rule, from the research: "as soon as an agent ingests untrusted input, it must be prevented from triggering actions with consequences." arXiv 2506.08837, with researchers from Google and ETH Zurich, 2025.
The 2026 distribution, declared before we measure.
The fit score is calibrated to our reading of market risk and applicable cases. This is the average distribution we expect the model to produce across a typical portfolio, declared before any measurement, the way a study pre-registers its hypothesis.
A1 · Advisor · 72.5%
AI with no connection to the accounts, fed with exports and advising on demand. The dominant market mode in 2026.
A2 · Assisted · 20%
Connected in read-only, analysing and suggesting continuously; the human executes.
A3 · Copilot · 5%
The agent prepares drafts and bulk changes; a human approves and launches.
A4 · Governed · 2%
The agent executes inside guardrails; only with top-tier access and a high fit score.
A5 · Autonomous · 0%
A market phenomenon in edge cases. Zero with Wise Pirates.
The commitment: the real distribution comes from the aggregated Ship's Log, quarterly. A material deviation forces a recalibration declared in the changelog, never a silent adjustment. Pre-registration, applied to our own calibration.
The log that proves, and protects.
Autonomy rises and falls with evidence, and the Ship's Log is where the evidence lives: the single record of actions, approvals and results, and the basis of the ROI of automation. It is also what makes the method hard to copy, because the history cannot be.
Audit trail
Every agent action: what, when, on what input, approved by whom. The basis of contractual accountability and of insurability.
Autonomy reviews
Levels per workflow rise or fall with evidence from the log; changes are captured in a versioned changelog, not made silently.
Agentic ROI
Hours freed and redeployed to senior work, errors avoided, launch speed and incrementality, measured against a pre-registered baseline.
The moat
Documented governance is what insurers already require to cover AI, and the accumulated log makes the methodology impossible to replicate.
Approving an agent is not junior work.
The Helm needs a rare profile, with real authority to stop, and it is exactly the profile Wise Pirates trains and has. The scarcity is not of agents; it is of senior people who can govern them.
The profile · five competencies, one person
Campaigns and platforms at expert level, not a weekend certificate. Analytics and attribution, to read real impact, not the platform's own metric. Business: margin, LTV and client priorities. Agentic AI: model limits, failure modes, prompts. Security: prompt injection, credentials, least privilege. With real authority to stop, under GDPR art. 22.
Why Wise Pirates
500+ brands managed on Google and Meta, pattern recognition no agent has. 100+ certifications, including Claude, Gemini and OpenAI. A full-stack Google partner, Ads Premier plus Cloud and AI, rare in Europe. Proprietary technology and methods, Feed'em, WAndrome, WiseGEO. A documented AI-first security practice, and a top-10 European performance agency.
The point: the scarcity is senior people able to govern agents, and that is our first line. It is why we are the partner for this transformation, not a casualty of it.
What no one else does, and we do.
We audited nine reference agencies and the agentic-tools market in July 2026. As far as we could verify, none of them publicly documented a per-workflow autonomy model.
| Capability | "Fire your agency" startups | Platform agents | AI-first agencies | Wise Agentic CM |
|---|---|---|---|---|
| Automation of execution | Yes | Yes | Yes | Yes |
| Autonomy levels per workflow | – | – | – | Yes |
| Senior HITL defined (5 competencies) | – | – | Partial | Yes |
| AI Act compliance embedded | – | – | Rare | Yes |
| Pre-registration + independent audit log | – | Partial | – | Yes |
| Contractual responsibility for errors | – | – | Yes | Yes |
Set reviewed (July 2026): Adsmurai, T2O, Publicis Sapient, Making Science, Brainlabs, Incubeta, DEPT, Jellyfish and Artefact. As of that review, none publicly documented a per-workflow governed-autonomy model. That is the whitespace, and Wise Agentic CM is built around it.
"The platforms govern their engine. We govern who touches the account, and we prove it."
Three ways to work, one owner of the governance.
Tech-agnostic and without lock-in. We fit the model to your organisation, and in every model the gate on sensitive actions and the Ship's Log stay with us.
Assessment & levels
We assess agentic maturity, fix levels, gates and the security design. Your team implements, with our QA.
Holistic partnership
We operate Helm, the gates, reviews and Log, and implement part; your internal teams keep execution where it makes sense.
End-to-end operation
Campaigns, agents, gates and Log run by the Wise team, integrated with your data and business teams.
In every model, the gate on sensitive actions and the Ship's Log are always assured by Wise Pirates, one owner of governance, whoever executes.
Agentic CM does not run alone.
The framework connects to the rest of what we do, so governed automation turns into measured business results, with one partner.
Tell us your accounts and your risk appetite. We will map what to automate, and at what level.
Wise Agentic CM, answered.
What is Wise Agentic CM?
Is this about firing the agency and connecting AI to the account?
What is Wise Agentic Helm?
What is the Autonomy Ladder?
How is this different from Performance Max or Advantage+?
Who approves the agents' actions?
How do you prove the automation is worth it, and keep it safe?
Do we have to let Wise Pirates run everything?
Where do the facts and numbers on this page come from?
The next chapter of campaign management.human at the helm.
Wise Agentic CM, the framework of governed autonomy. Wise Agentic Helm, the framework and the team: the Autonomy Ladder, the Agentification Matrix and the gates. The Ship's Log, the record that proves the value and protects the account.
Start with Wise Agentic CM →See all methodologies →
