Govern the Decisions, Not the Tasks: What AI Should Decide in Security Marketing

Ask a marketing team how it uses AI and you get a list of tasks: drafting emails, summarizing calls, first drafts of blog posts, cleaning lists. Ask which decisions AI now influences and the room goes quiet. The second question is the one that matters, and at a security company it matters more than almost anywhere else.

A sequencing tool that picks which accounts get the next round of outreach has made a targeting decision. A scoring model that ranks intent signals has decided who sales calls this week. A writing assistant that cites a detection rate has made a claim on the company’s behalf. Nobody signed off on any of them, because nobody thought of them as decisions.

Gartner’s guidance to CMOs this year is to define where AI supports judgment and where people remain accountable. That is the right instruction. What most teams lack is a way to do it in a few working sessions instead of a quarter of committee meetings. This is the method I use. I turned it into a free tool, the Marketing Decision Map, plus a 16-page playbook and a blank workbook, so your team can run it on its own decisions.

Why do AI policies in marketing miss the risk?

Most marketing AI policies are task lists. They say which tools are approved, what people may paste into them and which outputs need a second look. That is useful hygiene, and it answers the wrong question. A task list tells you what the tools do. It does not tell the CEO who is accountable when the output is wrong, and it does not say which calls AI is now quietly making.

The gap shows up in the numbers. In Gartner’s 2026 CMO Spend Survey of 401 marketing leaders, 70% said their internal processes are not mature enough to implement and scale AI, and only 30% reported mature AI readiness. Fewer than a third of CMOs say their AI investments are producing the returns they expected. And only 15% of CEOs view their CMO as having strong AI savviness.

None of that is a tooling problem. Governing tasks produces a long policy nobody opens. Governing decisions produces a short list with names on it, and a short list with names on it is something a CEO can trust.

Observe: find the decisions AI already touches

Start by listing what marketing decides, not what marketing does. In a security company the list usually falls into six groups:

• Buyer and market: which buyer the message leads with, practitioner or CISO; which segments and verticals get priority this half; which accounts get one-to-one treatment.
• Positioning and category: which analyst category to position in or against; when to name and claim a new category; how the product is described against adjacent tools.
• Content and threat research: when and what threat research to publish; which competitive claims and customer names are approved; what AI-drafted content can publish without expert review.
• Budget: the annual split across brand, pipeline programs and open funds; which conferences to commit to; weekly paid moves between campaigns.
• Pipeline and channels: what outbound AI may send, and to whom; which intent signals trigger outreach; lead scoring and the handoff to sales.
• Measurement and governance: the attribution model; what marketing reports to the board; which AI tools may touch customer and prospect data; when AI gets more authority, or loses it.

Then find where AI already influences each one, including tools people adopted on their own. The honest version of this exercise always turns up a few. A sequencing tool with an AI audience builder switched on by default. A scoring model a vendor updated last quarter. A writing assistant someone connected to the shared drive. Those are not failures. They are the inventory.

Analyze: score value, risk and frequency

Score every decision from 1 to 5 on three things, with the people who make the decision in the room:

• Value: how much revenue, pipeline or reputation rides on getting it right.
• Risk: how bad a wrong call is, and how hard it is to undo. A misread threat report, a customer logo used without permission or a claim the product cannot back up all score high.
• Frequency: how often the decision gets made. A call made daily compounds small errors fast.

Priority is value times risk, plus frequency, for a maximum of 30. Value and risk multiply because a decision that is both valuable and risky is far more dangerous than one that is only one of those. Frequency adds rather than multiplies, so a trivial daily decision cannot outrank a quarterly one the board cares about.

A score of 22 or more means govern first. 14 to 21 means govern next. Anything lower gets a periodic spot-check. With the suggested scores in the tool, deciding when and what threat research to publish lands at 27, deciding what outbound AI may send lands at 25, and weekly paid budget moves between campaigns land at 11. That ordering matches how most security marketing leaders already feel about those three. The scoring just makes it explicit, and gives you a way to argue about the score instead of the vibe.

The three AI roles

Every decision gets one of three roles:

• Human decides. AI may gather inputs and summarize. A named person makes the call and owns it. Use this for anything a board, a customer or a regulator would hold you to.
• AI recommends for approval. AI drafts the options and a recommendation with its reasoning. Nothing ships until the owner approves it. Use this for decisions with real consequences that repeat often enough to benefit from a first draft.
• AI acts within guardrails. AI makes the call inside written limits, such as budget caps, approved claims, audiences and send volumes, and logs every decision. The owner reviews a sample on a set schedule and can pull the authority back.

Start every decision one role more conservative than feels necessary. Moving a decision up is easy once the log shows clean results. Pulling authority back after a public mistake is not.

Why security marketing sits at the high-risk end

Every B2B team can use this method, but security teams carry more decisions in the govern-first tier than most. The buyers are skeptical by profession and remember a claim that did not hold up. Customer names are often under NDA, and a logo on the wrong slide is a contract problem. Threat research can misattribute an actor or tip one off. Analyst category choices shape the whole year. These are decisions where one error lands in front of a CISO, a customer’s legal team or the board, and no amount of time saved on drafting makes up for it.

That is also why the answer is not to ban AI from these areas. The research, the drafting and the first pass at a comparison are exactly where AI helps a small team move faster. The rule is about who decides, not who types.

One card per governed decision

Each govern-first decision gets a one-page card that anyone, including an AI tool, can read and follow. It names the owner, who is consulted, the inputs, what AI may do, what stays human, the approval gate, the line nobody crosses, and how you will know it worked.

For what outbound AI may send, and to whom, a card might read: the demand generation lead owns it, with sales leadership and legal consulted. AI may draft and personalize sequences from the approved claims library and the agreed audience rules. New audiences, any claim not in the library and outreach to executives stay human. The owner approves each new sequence and audience before it sends. Nobody, person or tool, references a named company in connection with a breach or threat. The measures are reply rate, error rate and how often the owner overrides the draft.

Write the card where the work happens. An approval gate that lives in the sequencing tool gets used. One that lives in a policy document does not.

Execute: the 90-day rollout

The rollout follows Observe / Analyze / Execute, the same sequence I use for the first 90 days in any marketing leadership role.

• Days 1 to 30, Observe. List the decisions, find where AI already touches them, score them with the people who make them, and name one owner for each of the top five.
• Days 31 to 60, Analyze. Write the five cards. Build the shared context people and AI decide from in one place, starting with approved claims, the ICP and funnel definitions. Have the security team review every AI tool that touches customer or prospect data. Set the approval gates inside the tools.
• Days 61 to 90, Execute. Run the top five under their cards and log every decision AI touched. Review error and override rates every two weeks with each owner. Move one proven decision up a role, pull one back if the log says so, and report to the CEO which decisions improved and what you will govern next.

Three mistakes that undo the work

Writing a policy instead of cards. A ten-page AI policy changes no decision. Five signed cards change five.

Expanding authority on a vendor’s say-so. A new feature release is not evidence. Your own decision log is.

Judging AI on time saved. Hours back on drafting are real, but they say nothing about whether the decisions got better. Measure error rates, override rates and the business outcome each card names.

Get the tool, the playbook and the workbook

The Marketing Decision Map starts with 24 decisions security marketing teams make, already scored, and lets you rescore, add or switch off any of them. It plots each one on value and risk, drafts the top five cards, checks your shared context and builds the 90-day rollout, then exports a branded Excel file and PDF of your results. It is free and nothing is stored.

The Marketing Decision Map Playbook explains the method in 16 pages, with two worked cards, a readiness check and six mistakes to avoid. You can download the PDF directly, and the blank workbook gives your team live scoring formulas and a shortlist that reorders itself as you score. No form, no email gate.

AI is already making marketing decisions. The only choice left is whether your team decided it should.