The gap between composable infrastructure and connected customer experience isn't a technology problem; it's a structural one.
A marketing team with thirty tools and no centralized data strategy is not a technology problem waiting to be solved. It is an organizational problem wearing technology's clothes. The distinction matters because the fix is completely different, and most teams are still reaching for the wrong lever. Adding AI on top of that mess does not accelerate growth; it accelerates the debt.
Bree Graham spent seven years at Afterpay and CashApp, leading growth and marketing platform transformation, scaling the business from siloed manual execution to hundreds of millions of hyper-personalized customer communications per month. She built CRM and lifecycle teams from the ground up, which means she did not inherit a clean architecture. She designed one out of chaos. That experience gives her a specific and hard-won view: the tools are rarely the problem.
The Stack Mirrors the Org, Not the Customer
Most companies have not built a martech stack. They have built a monument to how their internal teams are structured. Paid media owns its tools. CRM owns its tools. The data team owns its own version of the customer. Nobody owns the customer experience end-to-end, and the customer is the one who notices.
Graham is direct about where the breakage actually starts. It is not in the technology selection. It is in the absence of a shared customer goal sitting above the channel-level OKRs. When teams are running five different priorities separately, each with its own tooling and attribution logic, you get a consumer who receives redundant or contradictory communications and reporting that cannot tell you what actually worked.
"If you're not looking at the customer experience first and you're looking at siloed channels, you're not servicing anybody. And that is where the breakage comes."
The structural fix she describes is a reframe, not reorg. Before any tool decision gets made, the question has to be: what are we trying to drive for this consumer, what does that end-to-end look like, and what does the tech need to do to support that? Working backward from the customer outcome, rather than forward from the channel playbook, is what separates an architecture from an accumulation.
One Customer Record Is Not Enough on Its Own
The pitch for centralized execution, one customer record, one orchestration layer, collapsed silos, is appealing. Graham agrees with the direction but is careful about where teams oversimplify it. Collapsing the execution layer without rebuilding the data pipelines that feed back into the source of truth just moves the problem upstream. You end up with clean-looking campaigns and a warehouse that is still fragmented.
The composable architecture Graham points to is specific in its operation. The orchestration layer reads directly from the data warehouse rather than receiving a downstream push. That distinction changes what is possible. Marketers are no longer limited to the 150 contact properties they manually piped into their CRM tool. They get access to a fuller, real-time picture of the consumer, and the decisioning can surface signals the marketer would never have thought to look for.
"When we leverage our decisioning, it really unlocked our mindset... the outputs were so different to what we were thinking. And it's because it had the context that we hadn't fed into our CRM tools to do the job."
The practical implication is that much of what gets called AI decisioning today is simply automation running on incomplete data. The ceiling on what AI can do is set entirely by the foundation underneath it, and if that foundation is a patchwork of channel-specific data models, the decisioning will reflect that.
Ownership Without a Named Owner
One of the more counterintuitive points Graham makes is about who should own the orchestration layer. The honest answer, in her view, is that it should not be a single named owner in the traditional sense. When orchestration is owned by a specific team, it tends to get built to serve that team's priorities. When it is treated as a shared accountability with the right people in the same conversation, it becomes infrastructure rather than territory.
The role she describes, sitting inside a growth org rather than a product org, can co-build the growth strategy as it is being developed, rather than being handed a brief after the strategic decisions are already made. That proximity matters. Martech decisions made in retrospect tend to produce workarounds. Martech decisions made alongside strategy tend to produce leverage.
"I don't think it's really a role that someone should own, but I think it's really a role where if you have the right people in the team and thinking about it, it should always be an ongoing conversation."
The best version of this, as Graham describes it, is a future state where decisioning runs in the background, optimizing itself based on real-time customer behavior, and marketers spend their time on strategic bets rather than maintaining 40-touchpoint journey maps that are already out of date by the time they go live.
Getting the House in Order Before the AI Conversation
Graham is unambiguous about the sequencing: data governance is not a prerequisite for AI in theory, but it is in practice. The work of centralizing data, auditing what is being sent where, and understanding what first-party data is actually necessary outside the ecosystem is what makes everything downstream meaningful. Most organizations skip it because it is unglamorous, and then they wonder why their AI outputs feel shallow.
In a regulated environment like financial services, this discipline is not optional. But Graham's argument is that it should not be optional anywhere. Any consumer who hands over their data to a brand is extending trust. The obligation to govern that data well, to keep it centralized where possible, and to send it downstream only through highly gated channels belongs to every marketer, regardless of industry.
"The best thing we can do for our customer is keep that data in one place. So why don't we leverage the tools that do that?"
In her framing, procurement is underutilized as a governance partner. Because procurement spans multiple business units, they have natural visibility into when marketing, sales, and product are each independently buying overlapping tools. Treating procurement as a thought partner rather than a transaction facilitator means they can flag redundancies before they become technical debt and help architects like Graham make the case internally for consolidation.
The Vendor Relationship as Architecture Input
Graham describes her vendor relationships in terms that most practitioners would find unusual. She treats them as extensions of the team, brings them into conversations about business goals before specific briefs exist, and is explicit with them when a new product they are pitching is not a current priority. That mutual honesty is what she identifies as the difference between a vendor and a partner.
The best vendor interactions she describes are ones where she brings a consumer problem that may not obviously map to any single tool, and the vendor either connects her to a capability she had not considered or surfaces a partnership with another tool that solves for it. That kind of conversation only happens when there is enough accumulated trust for both sides to think out loud.
"I've had partners where I've gone to them with problems that aren't necessarily their problems, and their tool may not solve. But you sit down in a room and like, you brainstorm, and then they're like, oh, we're actually building this."
The emerging market of AI-native tooling makes this kind of relationship even more important. Newer vendors are hungry for real-world customer insights to shape their product direction. Graham has worked with vendors who bring her into product development early, specifically because she can pressure-test their assumptions against lived experience. That is a competitive advantage that does not appear in a feature comparison matrix.
Three Takeaways
When your martech stack mirrors your org chart rather than your customer journey, you are not dealing with a tooling problem. You are dealing with an ownership problem, and purchasing new tools will compound it rather than fix it.
AI decisioning is only as good as the underlying data architecture. If your data is still fragmented across channel-specific tools rather than centralized in a warehouse that the orchestration layer can read directly, you are not running real AI decisioning.
Procurement, IT, and vendors are not administrative overhead in a martech strategy. They are architecture inputs. The internal credibility you build with those partners directly determines the quality of the external partnerships and the infrastructure you can actually deploy.
Bree Graham is a growth and marketing platform transformation leader, formerly at Afterpay and Cashapp.
Special thanks to Claude for helping to summarize this conversation.
