Hello Reader

Kaan Catalkaya manages 15 clients by himself.

Each client has different goals, customers, campaigns, platforms and ways of working. Keeping track of it all would be difficult for a team.

Kaan does it as a solo operator.

He can manage that workload because every client gets its own command center.

The command center contains everything his AI system needs to work on that account: meeting transcripts, audits, reports, agreements, instructions, standards and a record of what changes over time.

When Kaan begins a new task, he doesn’t have to reconstruct the client’s history from memory, because the context is already there.

This is snapshot of Kaan’s client onboarding agent:

Most agencies (and marketing teams) have pieces of this information, but they’re scattered across call recordings, account plans, old chat threads, project-management tools and someone’s notes.

If this sounds like your team, then you’re likely to end up creating a campaign using the wrong messaging, basing future ad spend on out of date performance metrics or writing a draft that uses the wrong tone of voice.

The best way to fix this is by building an AI command center.

The AI Junk Draw

At a recent all-day workshop one of the participants was showing me his Codex set up.

“I have one Project for everything," he said proudly.

I had to suppress my look of horror. “How do you find something you were working on from a week ago? Isn’t it like hunting through a junk draw?”

“I can usually find it after a few minutes,” he replied.

I nudged a little harder. “You don’t find everything gets confused?”

He didn’t seem worried, but four weeks later I returned for another workshop and he pulled me aside to confide that his master Project was getting a bit out of hand.

“I can’t seem to find anything any more,” he admitted. “And I’m burning so many tokens. I think I did this all wrong.”

I told him it wasn’t too late to set it up correctly.

The Clash of Competing Context

The first few times you use AI for a particular client or brand, a collection of individual conversations is manageable.

You open one conversation for research, another for copy and a third for campaign planning.

You paste in a brand guide when you need it. Meeting notes live in Granola. The approved strategy is saved in Google Drive.

This works fine until several people are using AI across several accounts and projects.

Then you start repeating information, working from different versions of a brief and searching through old chats for decisions that should have been recorded somewhere permanent.

And when several clients or products share the same workspace (or Project), their instructions get confused.

I call this the clash of competing context.

If you don’t organize your files, instructions and connections appropriately, ChatGPT (and any other AI tool) aren’t going to know what applies where.

They’ll end up guessing which rule to follow and you’ll be left with the clean up.

Kaan kicking off a task with his agent:

Setting up your AI Command Center

Setting up your AI command center means creating two levels of context and putting them in the right place:

1. An agency or team Core for the processes your team runs
2. A separate Command Center for every account or campaign

These two parts let you reuse your company and team processes while keeping each client, campaign or product distinct.

What Belongs in the Core

Your Core contains the knowledge that applies across accounts, projects or campaigns.

This could include:

• Your campaign planning process
• Research methods
• Brief templates
• Reporting formats
• Review checklists
• Approval stages
• Writing standards
• Words and phrases your team avoids
• Lessons your team can apply across accounts

Think of this as your operating manual.

Easy route: load this context is in your custom instructions under settings.

Advanced route: If you’re familiar with Codex or Claude Code, then you can add this to your agents.md or Claude.md file.

Here’s what my custom instructions look like in Claude:

The 5 Layers of an AI Command Center

Now you have your Core built it’s time to start on the Command Center. To work effectively you need enough context to answer five questions:

1. Who is this client/what is the product?

Start with the information that remains relatively stable:

• Company overview
• Products and services
• Target audiences
• Positioning
• Competitors
• Brand voice
• Approved claims
• Words and phrases to avoid

This gives AI a clear picture of the business.

If you’re working at a company with a single brand you can adapt this approach for different products or solutions. Instead of a company overview and brand voice, you’re saving the product or solution specs and user benefits.

2. What are we working toward?

Add the information that changes throughout the relationship:

• Current business priorities
• Campaign goals
• Active offers
• Target segments
• Deadlines
• Budgets
• Success measures
• Known constraints

Make sure you record the date on any of these context documents so AI knows its recency. A strategy document from six months ago can sound authoritative while sending the work in the wrong direction.

3. What evidence can AI use?

Give the Command Center a clear collection of source material:

• Customer interviews
• Sales-call transcripts
• Research
• Performance reports
• Previous campaigns
• Approved creative
• Product documentation
• Meeting notes

Source material helps AI base its work on what the client’s customers have said, real campaign data and prior learnings.

4. What’s in production?

Create one place for current work:

• Briefs
• Drafts
• Feedback
• Final deliverables
• Open questions
• Next actions

Keep the status clear: label drafts, approved versions and archived work so the AI doesn’t treat them as equally current.

5. What’s already been decided?

Keep a running decision log containing:

• Approved recommendations
• Rejected directions
• Client or leadership feedback
• Budget decisions
• Scope changes
• Brand choices
• Final sign offs

Meeting transcripts capture what people said and a decision log records what the team agreed to do. Without it, AI ends up resurfacing old ideas, ignoring preferences or recommending a change that’s already been discussed.

Kaan also records where the AI’s authority ends. His system can examine performance and recommend changes, but he approves decisions involving budget.

Every Command Center should contain equally clear boundaries.

If you’re using the browser version of ChatGPT or Claude you’ll want to add all of this to a project. If you prefer working with folders on your desktop then set this up using Codex or Claude Code.

Here’s what the instructions look like in one of my client projects in ChatGPT:

And here are the sources and connectors:

Use AI to Build it For You

This probably sounds like an awful lot of work, but don’t worry!

AI can do it for you.

Kaan uses the client kickoff to create the first version of his Command Center. He records the conversation, adds the transcript to the system and asks AI to extract the client’s business context, audience, goals, constraints and open questions.

He then reviews the extracted information and that becomes the foundation for the account.

You can follow the same process in Claude or ChatGPT:

1. Create a project for your client or product
2. Connect your call transcriber (or load the transcript manually)
3. Ask AI to extract the key information you need.
4. Correct any gaps or assumptions.
5. Save the approved context in the project instructions.
6. Add new source material and decisions as the engagement or campaign progresses.

In-house marketers can use a product kickoff, annual planning session or regional briefing in the same way.

Here’s Claude at work creating my client context:

Your Command Center Needs an Owner

A good setup can become outdated within weeks if nobody maintains it.

Assign a person to keep each command center current. This could be the account lead, product marketer or campaign executor—who ever is closest to the project is the best fit here.

After each client meeting, add the decisions, feedback and assigned actions and after each campaign, save the final assets, performance summary and lessons back into the project.

Once a month, review the command center for outdated instructions, old drafts and missing decisions.

Kaan’s system connects with client tools and advertising accounts. It can retrieve performance data and recommend campaign changes. More access creates more ways for the system to help.

--

Have you built your own AI command center? I’d love to know. Hit reply and tell me.

Tahnee

Your Client Context Prompt

Use this prompt to create your client context doc:

Use my call transcriber tool to find my client call with [CLIENT OR CONTACT NAME].

Search the meeting title, attendee names and transcript content. If you find more than one possible call, show me the matches and ask me to select the correct one before continuing.

Once the call is confirmed:

1. Retrieve the complete transcript and meeting notes.
2. Identify the participants and their roles when stated.
3. Extract the client information needed to create a Client Command Center.
4. Use only information contained in the call. Don’t fill gaps with assumptions.
5. Label unclear, conflicting or unconfirmed information for follow-up.

Create the command center using the following structure.

1. Client identity

Capture:

• Company overview
• Products and services
• Business model
• Target audiences
• Customer needs and buying triggers
• Positioning
• Competitors and alternatives
• Brand voice
• Approved claims
• Language or approaches to avoid

2. Current direction

Capture:

• Business priorities
• Goals for the engagement
• Active offers, products or campaigns
• Priority audience segments
• Deadlines and milestones
• Budget information, if discussed
• Measures of success
• Current challenges
• Scope boundaries

Add the call date and label this section “Current as of [DATE].”

3. Source material

Create a list of every source mentioned during the call, including:

• Documents
• Research
• Customer interviews
• Campaigns
• Performance reports
• Brand materials
• Product information
• Tools and platforms

For each item, record what it contains, who owns it and whether it still needs to be added to the Project.

4. Active work

Capture:

• Current projects and deliverables
• Status
• Owner
• Deadline
• Dependencies
• Open questions
• Next actions

Use a table with these columns:

| Work item | Status | Owner | Deadline | Dependencies | Next action |

Use only these status labels: Proposed, In progress, Awaiting approval, Approved, Complete or Unclear.

5. Decisions and approvals

Capture:

• Decisions made during the call
• Directions that were rejected
• Client preferences
• Changes to scope
• Who can approve each type of work
• Actions AI can support
• Actions requiring human review
• Actions requiring client approval
• Financial, legal, privacy or publishing boundaries

For each decision, include the supporting transcript excerpt and timestamp when Granola provides one.

Output requirements

• Write in direct, specific language.
• Preserve the client’s terminology.
• Separate confirmed information from implied or missing information.
• Don’t include credentials, passwords or access keys.
• Don’t treat casual discussion as an approved decision.
• If the call doesn’t contain enough information for a section, write “Information needed” and add the gap to the follow-up list.
• Cite transcript timestamps where available.
• Format the finished result as a clean Client Command Center document that I can add to this Claude Project.

After creating the document, give me a short quality report showing:

• Which sections have strong source coverage
• Which sections rely on limited information
• The five most important questions to answer before using this command center for client work

Wait for my review before treating any extracted information as approved Project context.

A useful campaign system carries the audience and message from one stage into the next. You need to see what goes into the research, how it changes the strategy and what the creative brief needs to preserve.

On Sep 16, we’ll demonstrate that sequence in a 90-minute Masterclass.

You'll watch the campaign agent at work and receive a setup guide, seven skill files and bonus templates. And if you’re busy that day, don’t worry you get lifetime access to the student portal and the recording.

Use AGENT at checkout for $100 off. Deadline midnight tonight.

​Register here​

Already registered? Your seat is booked, and we'll see you there.

The AI: ROI Conference

AI is getting expensive. This is how leading orgs solve ROI.

Cutting all AI spend isn’t an option, but neither is going bankrupt.

So how do top orgs solve ROI?

At the AI: ROI Conference, 10 enterprise leaders explain how they drive ROI without blowing the budget.

A number of my Section colleagues, including Scott Galloway, will be presenting at the event and sharing what’s working in enterprise organizations today.

It’s one day, virtual, and costs nothing.

​Register for the Sept. 17 session.​