The 90-day AI integration plan for marketing leaders

If you’re a CMO or marketing leader and your team is using ChatGPT inconsistently, you don’t have an AI strategy, you have AI usage. Those are different things, and the gap between them is where most marketing functions are quietly losing ground right now.

Here’s a 90-day plan that closes that gap. Concrete, sequential, achievable without hiring or restructuring. The shape works in agencies, in in-house teams, in lean B2B functions and in commerce teams.

Weeks 1–2: audit what’s actually happening

You can’t plan the integration until you know the starting position. The audit is two questions.

What is everyone on the team actually using AI for, right now? Survey it. The honest answer is usually a mix of unsanctioned ChatGPT use, one or two paid subs nobody’s tracking, and a few power users running shadow workflows the rest of the team doesn’t know about.

What’s the brand voice doc, and is it operational? Pull it up. If it’s a Notion page describing tone with adjectives it’s not operational, because the voice doc has to be specific enough to be prompted from, with examples, bans and patterns. If it’s not, that’s your first fix.

By end of week two you should have a one-page picture of current AI usage by team member, current tool spend, voice-doc state, and three biggest pain points the team would change if they could.

Weeks 3–4: write the voice prompt and standardise the basic stack

The single highest-leverage thing you can do in the first month is get the brand voice prompt right, because it propagates through every other workflow you’ll set up later.

Three positive examples (paragraphs of strong on-brand work). Three negative examples (off-brand or close-but-not-quite drafts, annotated). Specific bans. Specific structural patterns. Get a senior writer or editor to own it. Iterate it until pasting it into a draft request produces something close to ship-ready. The brand voice extraction playbook in The Lens walks through building it.

While that’s happening, settle the tool stack. Pick one chat model (Claude, ChatGPT, Gemini, defendable choices, pick one), one research tool, one media-production tool. Three subscriptions, all paid for centrally. Kill the unsanctioned shadow stack. (For a practical way to pick, see how to evaluate AI tools without falling for the demo, where the four-week pilot rule saves a lot of budget.)

The goal at the end of month one is that every team member has access to the same tools, the same voice prompt and the same baseline expectation of what AI is for.

Weeks 5–6: rewrite one workflow end-to-end

Pick the workflow that’s the biggest drag on the team. Usually it’s content production, briefs to drafts to publication. Sometimes it’s reporting. Sometimes it’s campaign creative production. (We’ve broken down five workflow changes that lift content production specifically, so start there if content is your biggest drag.)

Whichever it is, rebuild it with AI in the loop properly. Not “the team uses ChatGPT during the workflow”, rebuilt. The brief is AI-generated and human-edited. The draft is AI-produced against the brief, the voice prompt and the research. The edit is human. The repurposing into other formats is AI-produced from the long-form. The whole workflow ships. (The brief-to-ship pipeline playbook is one worked version of this rebuild.)

Document it, run it on three real pieces, and measure the time and quality vs the old workflow.

This is the proof point. If the team can see one workflow that’s genuinely faster and at least as good, the whole adoption gets easier.

Weeks 7–8: roll the workflow pattern to a second area

Take the same pattern (defined inputs, AI-assisted production, human-owned editorial decisions, defined outputs) and apply it to a second function. Reporting, campaign briefs, ad creative production, social repurposing. Whichever is the next biggest pain.

Don’t try to do all of them. Two well-running rebuilt workflows beats five half-built ones, because the point of this stage is to prove the pattern travels, not to maximise coverage.

By the end of week eight you should have two workflows running cleanly, each documented enough that a new hire could pick them up in a week.

Weeks 9–10: build the measurement loop

A lot of AI rollouts stall here. The team produces more, but nobody tracks whether the extra output is actually working, and six months in there’s no honest answer to “is this paying off”.

Set up the loop. One dashboard, one source of truth, weekly cadence. Each piece of work tagged by workflow (which AI process produced it), by topic, by format. The model summarises what’s working weekly, and the senior person reads it and makes the call on what to adjust.

This loop is what makes the engine compound. Without it, you’re just producing more output of unclear quality.

Weeks 11–12: the team conversation

By month three the rollout has produced real changes. The team has shifted what they do day-to-day. Some people will have leaned in, some will have resisted, and a few might be checked out because they think their job is about to disappear.

This is when the leadership work matters most.

Run a one-hour session with the team. Not a strategy presentation, a conversation. What’s working, what’s not, where they’re nervous, what they want more of. Be specific that AI doesn’t reduce the senior marketing problem, it magnifies it, and the people who learn to operate the workflow well end up doing more of the strategic work, not less.

This conversation is the one most leaders skip. Without it, the people-side of the change goes underground and the workflow gains never quite stick.

What 90 days actually gets you

End of quarter, a team that wasn’t doing structured AI work can be producing substantially more, at brand quality, with measurement in place, against two well-run workflows. Not the agentic-future version of AI in marketing, the practical version that exists today and produces real lifts.

The next 90 days is about widening the surface (more workflows), the 90 after that is about depth (better pipelines, better models, better measurement), but the first 90 sets the floor.

The teams that don’t do this work will spend the next year being out-shipped by teams that did. The differentiator is disciplined integration, and that’s a leadership job rather than a tools job.

I lead marketing at [company], a [type of team: agency, in-house, B2B,
commerce] team of [number] people. I'm starting a 90-day plan to move us from
inconsistent AI use to structured AI work, and I need an honest picture of
where we are now.

Attached are:
- the team's answers to a survey on which AI tools they use, what for, and
  how often (including tools the company doesn't pay for)
- our current software spend, or the list of AI subscriptions on expenses
- our brand voice document

Do three things.

1. Give me a table of current AI usage with the columns: Team member or role |
   Tools used | What they use them for | Paid for centrally? (yes/no/unknown) |
   Monthly cost (from the spend file).
2. Assess the voice document. Is it operational, meaning specific enough to
   prompt from? Check for positive examples, negative examples, specific bans
   and structural patterns, and list which are missing. Quote any lines that
   are only adjectives describing tone.
3. List the three biggest pain points the team raises, with how many people
   mentioned each one.

Finish with a one-page summary I can share: current usage, current tool
spend, voice-doc state, top three pain points. Don't estimate any cost or
count you can't see in the files; mark it "not in data" instead.

I need an operational brand-voice prompt for [brand], one we can paste into
any draft request so the output comes back close to ship-ready. Our senior
[writer/editor] will own it and refine it; you're writing version one.

Attached are:
- three pieces of strong, on-brand work: [titles]
- three pieces that were off-brand or close-but-not-quite: [titles], with a
  line from me on what was wrong with each
- our current brand voice document, if we have one

Write the prompt with these sections:
1. Who we are and who we write for, in two or three sentences.
2. Positive examples: a paragraph from each strong piece, with a line on what
   makes it on-brand.
3. Negative examples: a paragraph from each weak piece, annotated with what
   to avoid.
4. Specific bans: words, phrases and habits never to use.
5. Structural patterns: how our pieces open, how long paragraphs run, how we
   use headings, how we close.

Base every rule on the attached examples, not on general good writing advice.
After the prompt, list anything you weren't sure about or couldn't tell from
the examples, so our editor knows where to look first.

Set this up as a routine that runs every Monday morning.

Each week I'll attach an export from our marketing dashboard. Every piece of
work in it is tagged by workflow (which AI process produced it), by topic and
by format, with its performance figures.

Each week, give me:
1. A table with the columns: Workflow | Pieces published | Main performance
   figure | Change on last week.
2. The same for topic and for format.
3. Three short bullets on what's working and three on what isn't, each tied
   to a number in the export.
4. Anything that moved sharply since last week.

Don't recommend what to change; a person makes that call. If a piece is
missing a tag, or a figure is missing or looks wrong, list it separately
rather than guessing.