In my recent articles, we've explored two critical dimensions of how AI is reshaping marketing operations: First, we mapped out the landscape of accumulated intelligence, defining the different categories of learning (sovereign, shared, ecosystem, portable) that persist inside AI systems and how those distinctions matter for who controls what. Then we examined the infrastructure layer beneath the agents themselves: the persistent harness systems where intelligence actually accumulates, and the foundational architectural decisions vendors are making about ownership and federation without brands even realizing it.

Now the question gets more practical: What do brands need to ask before signing on the dotted line?

What this means for your brand

While vendors are reorganizing around persistent intelligence systems, most brand/vendor relationships haven't caught up to what that means. The nature of what you're buying is changing from pure execution to operating inside an intelligence environment.

Whether you’re working with agencies, tech platforms, or service providers, your vendors aren't just executing for you anymore. They're building persistent AI systems that accumulate learning based of your decisions, your budgets, your audience strategies, and your optimization patterns over time.

The learning you should own

After a few months, your vendor's AI learns which audiences respond to your messaging, which creative angles drive conversions and what your optimal spend looks like. After a year, it's learned deeper patterns specific to your business: which audience segments drive repeat purchases, how you make budget decisions, what messaging resonates with new customers vs. existing ones, how supply constraints impact your ad spend.

That's irreplaceable intelligence that you need to own and is the kind of accumulated learning that becomes your competitive advantage. That learning could be locked inside systems you don't have access to, embedded in proprietary infrastructure. Which means you can't take it with you if the relationship changes.

Before that happens, you need to ask the right questions to define who owns what in this new shift.

The Six Questions Every Brand Should Ask

Before deploying persistent AI systems with any vendor, ask these six questions first:

1. What intelligence stays with you?
When your vendor's AI system learns from your campaigns, your decisions, your budget patterns, and your audiences, does that learning stay accessible to you? Can you export it? Can you take it to another partner if needed, or does it stay embedded inside your systems? For example, if your vendor’s AI learns that a specific audience segment has a 3x higher lifetime value for your product, that’s intelligence you need to know. But if that insight only exists in the platform’s knowledge graph and can’t be exported, you may never even know about it, let alone be able to export or build a strategy around it.

2. Where does that intelligence live?
Is it living inside your vendor's infrastructure? Inside their partner's systems? Inside a third-party cloud platform? The location determines governance, access, portability, and control. I asked this exact question recently myself. The vendor went on to explain how they implement a knowledge graph at install, and that over time, accumulated learnings and context would improve and grow this intelligence within their graph. When I asked if I take the graph with me if I part ways in 5 years, the answer was “I’m not sure, I’ll follow-up and get back to you.” Honestly, I expected this answer, because it’s not always easy to parse out what intelligence is theirs, mine, or shared within that graph given that we are all still working it out. What you should be looking for in a response is transparency, not the perfect answer.

3. What can be exported today?
What can actually leave the system right now? Think about what you’re measuring today that your team might need readily available such as:

• Campaign learnings
• Optimization patterns
• Audience insights
• Approval workflows
• Measurement logic

Perfect portability of intelligence does not yet exist within the major commercial AI platforms, so manage your expectations accordingly. But if a vendor can’t extract basic sovereign intelligence like campaign performance, as something you could access today, that’s likely a red flag for down the road.

4. What's proprietary to vendor partners?
Your vendor might be using multiple third-party AI systems within their AI product. You need to know: Are there multiple harnesses in the solution, is the workflow logic, the memory systems, orchestration layer, or intelligence in the models themselves fragmented within this solution? Extraction of intelligence may be more challenging if the answer to that question is “Yes.”?

5. Can I swap components?
Understand what has to change in the system if you want to change a model, integrate a different data source, or use another tool. ”Composable” gets thrown around as a buzz word a lot these days, but if swapping a model requires rebuilding your entire workflow and governance guardrails, that’s a sign that components aren’t truly composable.

6. What will I need to rebuild if I leave?
An industry AI expert used a great analogy: When you start using an agentic harness it’s like having a very smart friend who you’ve known for years and knows you really well, from the restaurants you like or dislike, shows you enjoy or hated, to the parties you like or don’t like to go to.

When you move to a new AI harness, it’s like making a new friend who doesn’t know all these things about you. They’re still very smart, but it will take time and new experiences to get back to the same place. If you end the relationship, what business dependencies will you have to replace or recreate? What accumulated learning will you lose? Ask your vendor to be specific about what tools, workflow, context skills or you would need to replace from scratch.

The Brands Getting Ahead

The main point here is that brands need to understand what they're actually buying, who controls what, and what the long-term implications are. Vendor lock-in is not necessarily bad, but unintentional lock-in is.

The brands that get ahead will be the ones treating intelligence governance as a strategic priority from day one. They'll ask these questions before they commit to understand where their learning lives. They'll know what they can take with them and will seek to maximize sovereignty over their competitive advantages.

The question isn't whether to embrace AI-driven partnerships, but rath

er if you maintain control over the intelligence they accumulate on your behalf.

At dentsu, we consider composability and client ownership a main priority. Learn more about dentsu.Connect, our end-to-end workflow here.