AI for CMOs: The Decisions a Marketing Leader Has to Make

Written on 29 October 2026 by Gaurav Phogat. External facts checked against their original sources on the same date.

AI for CMOs is mostly not about tools. It is about ten leadership decisions: where AI sits in the operating model, which roles the team needs, what AI agents may run without a person, where the budget comes from, which vendors to trust, how brand safety and consent are governed, how value is reported to the board, how people are trained, and what the function stops doing. Make those decisions explicitly and early, and the tools become a procurement question.

New to the topic? Start here: Hospital growth strategy sets out the reading order.

Every CMO I speak to has the same problem in a different shape. The CEO wants an AI story, the CFO wants savings, the team wants permission, and the agencies want to sell a pilot. Meanwhile the platforms are automating the work underneath all of them. The way through, and the real meaning of AI for CMOs, is to stop treating AI as a project and start treating it as a set of decisions that only the marketing leader can make.

I have made these decisions as a senior marketing and digital leader in large hospital groups, where every claim is reviewed and every patient record is sensitive. Healthcare made me cautious in some places and faster in others, and that is the lens I use below. For the practical side, what AI changes in each part of marketing, read the companion guide AI in marketing: strategy, examples and tools first.

Why AI for CMOs is a leadership decision now

Three facts frame the conversation. Budgets are not growing: Gartner’s 2025 CMO Spend Survey found marketing budgets flat at 7.7 percent of company revenue, and 22 percent of marketing leaders said generative AI had reduced their reliance on external agencies for creative and strategy work (Gartner, Demand Gen Report). Results are uneven: MIT’s Project NANDA reported in 2025 that only about 5 percent of integrated generative AI pilots were extracting significant value (Virtualization Review). And agents are overpromised: Gartner predicts that over 40 percent of agentic AI projects will be cancelled by the end of 2027 because of cost, unclear value or weak risk controls (Gartner).

Put together, that means AI will not rescue a flat budget by itself, most pilots will disappoint, and the leader who decides clearly will get more from less. Here are the decisions.

Decision 1: Where AI sits in the operating model

There are three models. A central AI team builds and governs everything. An embedded model gives each marketing team its own tools and freedom. A hybrid keeps a small central group that owns standards, the prompt and template library, the tool list and training, while the work happens inside each team.

I choose the hybrid almost every time. Central-only becomes a bottleneck and an innovation theatre. Embedded-only produces a dozen tools, no audit trail and no shared learning. The hybrid puts the work where the expertise is and the rules where they can be enforced. In a hospital group, that also lets one set of claims rules apply across every unit.

Decision 2: The team and the new roles

You do not need a large AI team. You need a few roles that did not exist in most marketing org charts three years ago, and you can usually fill them from inside.

• AI marketing operations lead: owns the tool list, the template library, the logs and the training calendar.
• Workflow designer: turns a repeated task into a prompt, a template and a review step that anyone on the team can run.
• Claims and content reviewer: checks facts, claims and tone before anything goes out. In healthcare, this sits alongside doctor sign-off.
• Marketing data lead: keeps conversion data, consent fields and definitions clean enough for AI and ad platforms to use.
• AI search analyst: tracks what AI assistants say about the brand and fixes the sources.

How I structure the wider digital team, and which skills to hire versus train, is in building a digital team inside a hospital group.

Decision 3: AI agents in marketing, and where a human must approve

An AI agent is software that plans and carries out a multi-step task across tools, rather than answering one prompt. Vendors now label almost everything an agent; Gartner calls this “agent washing” and estimates that only about 130 of the thousands of vendors claiming agentic AI are genuine (Outlook Business). The CMO’s job is to decide where agents may act alone and where they stop and wait for a person. My rule is simple: agents may prepare, recommend and monitor; people approve anything that is published, sent to a customer or spends money.

In healthcare the “never” column is longer: no agent gives clinical advice, triages symptoms or promises outcomes. Other industries have their own version, and the CMO should write it down.

This connects directly to How to Choose a Healthcare Digital Marketing Agency in India: A Buyer’s Checklist, Top marketing leaders in India: how the strongest CMOs lead and The commercial leader’s path to the top job.

Decision 4: Budget, and where the money comes from

With flat budgets, AI money has to come from somewhere. The honest sources are production, versioning, translation and reporting, the work where AI saves the most time. I fund the first year by moving a share of production and agency spend into tools, data work and review capacity, and I show the CFO the trade explicitly. A budget line called “AI innovation” with no source and no target will be the first cut in the next review.

Decision 5: Vendors, and whether to build or buy

The default order I use is: use the AI already inside the platforms you pay for, buy a specialist where a pilot has proven the workflow, and build only where the workflow itself is a competitive advantage. Building is tempting because it looks strategic, but it creates maintenance, security and model-update work that marketing rarely budgets for. My fuller framework, written for healthtech but general in logic, is in buy vs build for healthtech vendors.

Whatever you buy, ask the same questions: where does our data go, does it train someone else’s model, can we see what the system did, what does it cost per unit of work, and can we leave with our data and templates.

Decision 6: Governance and brand safety

Governance is what lets a marketing team move fast without the CMO approving every asset. I use three risk tiers: internal work, public content, and anything that makes a claim, uses a real person’s likeness or talks to customers in real time. Each tier has a reviewer and a record. Disclosure is part of this: in India, ASCI’s guidelines on labelling synthetically generated content in advertising, released on 29 September 2026, take effect three months later (ASCI). I cover what that means in practice in ASCI AI labelling rules for hospital marketing.

Decision 7: Data and consent

The quality of AI output in marketing depends on the quality of the data it reaches, and the legality of using that data depends on consent. India’s DPDP Rules were notified on 13 November 2025, with most notice and consent obligations applying from 13 May 2027 (S.S. Rana summary). The CMO should insist that consent purposes are stored as fields that automation reads, so the system refuses to send what a customer did not agree to. This is a joint decision with IT and legal, but marketing should own the requirement.

Decision 8: Measuring and reporting AI value to the board

Boards do not want an AI update. They want to know what changed in the business. I report AI in three layers: efficiency (time and cost per unit of work against a baseline), effectiveness (conversion quality and customer acquisition cost) and new capability (work that was not possible before), plus one risk line for incidents and corrections. The structure of the pack itself is in the board pack from digital and growth, and the way I agree metrics with finance is in measuring non-clinical AI projects.

Decision 9: Talent and upskilling

Generic AI training fades in weeks. What sticks is training on the team’s own tasks, with the team’s own templates, followed by a month of using them under review. I ask each team to nominate one person who learns faster than the rest and give them time to teach. I also change job descriptions, because people do not invest in skills their role does not recognise.

Decision 10: What to stop doing

This is the decision most CMOs skip, and it is the one that frees the capacity. Candidates include manual weekly reports that an agent can draft, hand-built versions of the same asset for every channel, agency retainers paid by the hour for work AI now shortens, and campaigns optimised for lead volume when the platforms will happily find cheap leads that never convert. If nothing stops, AI simply adds work.

A CMO’s first 100 days with AI

If you are doing this inside a group with several businesses or units, align the plan with the wider enterprise plan. The version I recommend for a hospital group is in an AI strategy a hospital group can execute.

Questions the CEO and board will ask, and how I answer them

• “What are we getting for the money?” I answer with the three layers against a baseline, and I say what the saved time was used for.
• “Are we behind our competitors?” I show what AI assistants and search engines say about us versus competitors, because that is visible and comparable. Tool counts are not.
• “What could go wrong?” I name the top risks, the control for each, and the incidents so far, including the ones we caught.
• “Can we cut agency or headcount?” I show which tasks have shrunk and which roles have grown, and what I propose to move. I do not promise cuts before the pilots prove the savings.
• “Who is accountable if the AI gets it wrong?” The person who approved it. AI is a tool in a reviewed process, never the accountable party.
• “Why not just let the platforms do it?” We do, for buying. But the platforms optimise what we feed them, so the inputs, the data and the brand rules stay ours.

What I learned the hard way

I have made most of these mistakes myself. Each one cost time, and a couple cost credibility with people whose support I needed.

• I announced an AI programme before I had an owner for it. The announcement created expectations, and for weeks nobody was accountable for meeting them. Now I name the owner and the first two problems before anything is said in a town hall.
• I let every team choose its own tools. It felt empowering. It produced overlapping licences, data going into personal accounts and no way to share what worked. A short approved list fixed more than any training did.
• I reported activity to the board. Assets generated and hours saved sounded impressive in marketing and meant nothing in the boardroom. The second time, I reported cost per unit of work and what changed in acquisition, and the questions changed.
• I automated a broken process. AI made a messy approval chain faster at being messy. Fixing the process first, then adding AI, took less total time.
• I expected the agencies to resist. Some did. The better ones rewrote their own scope once I was clear about what AI would take over and what I still wanted from them: ideas, craft and judgement.
• I underestimated how much the role of the marketing leader changes. Less time approving outputs, more time setting rules, defending data quality and explaining trade-offs to the CEO. I wrote about that shift in what I had to unlearn moving from digital head to growth leader.

Pilots that never scale are the most common way AI programmes die, and the reasons are usually organisational, not technical. I go through them in why AI pilots never reach production. For the CMO role itself, including what it now owns beyond campaigns, see what the CMO role owns now.

AI will not make the decisions in this article easier. It will make the cost of not making them higher, because the platforms, the competition and the team will all move anyway. The CMO’s advantage is to decide first, write it down and hold the line on what stays human.

Questions people ask