For the last two years, AI in commerce media mostly delivered faster answers: a recap of last week’s performance, a first draft of a QBR deck, and a fresh keyword list. Now AI agents can make recommendations, coordinate workflows, and act across live advertising accounts. For enterprise brands, that shift raises a new question about the technology required to support it.
In February, Amazon Ads introduced its Model Context Protocol (MCP) server. It lets agents built with Claude, ChatGPT, or custom applications create campaigns, pull reports, and manage billing without custom integrations. At the same time, the barriers to building software are falling. According to McKinsey’s 2026 State of AI report, 32% of organizations have decided against buying at least one software product because they could build it internally with agentic coding tools.
The build-versus-buy debate is louder than ever among enterprise commerce teams. But the most useful question is whether your organization has the infrastructure to make AI reliable, connected, and accountable at scale.
Where Building Your Own AI Tools Makes Sense
For advertisers with straightforward needs, connecting an AI assistant directly to a retailer delivers immediate value. A brand managing a handful of Amazon campaigns can use Claude and Amazon’s MCP server to pull reports, investigate trends, and prepare quarterly business reviews.
The challenge emerges when those workflows scale beyond a single retailer or team. An enterprise needs a tool that can help answer where to put its next media dollar, show if underperforming campaigns are a result of ad strategy or inventory constraints, and act without losing financial control.
Four Challenges of Scaling DIY Commerce AI
More Retailer Connections Don’t Eliminate Data Silos
Every retailer brings its own reporting structure, attribution method, product taxonomy, and campaign capabilities. Retailer MCPs make each environment easier for an agent to reach, but they don’t reconcile the differences between them. An agent connected to Amazon and Target can retrieve data from both and still fail to produce a consistent, fair comparison.
The silos extend inside the business, too. Media, commerce operations, and finance often work from different systems with different definitions of success. More connections give an agent more data, but not a shared understanding of the business.
Connectivity Doesn’t Equal Commerce Intelligence
When Amazon launched its MCP server, it acknowledged that connectivity alone doesn’t guarantee reliable outcomes, especially when advertising tasks span dozens of steps.
Good decisions require multi-year history, competitive Share of Voice, retail signals like inventory and content health, and proof of what drove incremental sales. Without those signals, the agent risks optimizing one metric at the expense of the larger business objective.
Prompts add another layer of risk. The same request phrased in two ways can return two different answers. When a data point is missing, an agent will often fill the gap with a confident guess, and without a shared data layer, every analyst’s prompt becomes its own version of the truth.
Write Access Requires Governance
The stakes change when AI moves from answering questions to taking action. At enterprise scale, a single budget or bid change can affect millions in media investment and established financial controls. Your team needs to know who authorized a change, what it touched, and how to undo it.
Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. If you’re the one building, you must find out where your approval rules live. If they sit inside a prompt, one configuration change removes them.
Ownership Costs More than the Build
AI development tools have shrunk prototyping time from months to days. Everything after launch still takes the same work: maintaining integrations, updating business logic, managing permissions, passing security review, and responding when retailers change their APIs.
Operational continuity matters too. When the person who build a workflow changes roles, someone must inherit it. Without a shared foundation, AI can recreate the silos enterprises spent years eliminating. The result is a new form of technical debt, built faster than ever before.
Before You Build, Ask These Five Questions
- Can the agent meaningfully connect advertising, competitive, operational, and financial signals across the business?
- Does it have the commerce-specific context to make reliable recommendations, or will your team need to build and maintain that logic?
- How are permissions, approvals, and audit trails enforced as agents start taking action?
- Who maintains integrations, reporting logic, and security as retailers and requirements change?
- Can you connect AI-driven decisions to execution and measurable outcomes?
If your team can answer these confidently, custom development may be a strong investment. If not, the greater opportunity is building your own AI experiences on infrastructure that already answers them.
Build on a Connected Commerce Foundation
Your strategy, customer expertise, business priorities, and workflows are worth owning, and that’s where custom agents create a meaningful advantage. Cross retailer reporting, data normalization, competitive intelligence, and access controls are foundational work a team shouldn’t have to rebuild.
Consider a global brand evaluating where to increase media investment ahead of a major retail event. Its agent sees strong advertising performance on Amazon and recommends more spend. But a connected view shows the same product running low on inventory, while another retailer offers stronger growth and room to gain Share of Voice. That’s the gap between a good Amazon decision and a good business decision.
How Pacvue Connects the Enterprise Commerce AI Ecosystem
Pacvue’s AI-Powered Commerce Media OS brings advertising performance, retail operations, competitive intelligence, and commerce data from Amazon, Walmart, and more than 100 other retailers into one foundation. That includes Share of Voice, historical performance, inventory, and pricing signals that individual retailer connections don’t provide on their own. Our engineers maintain those retailer connections, so when an API changes, it becomes our ticket instead of yours.
Teams can use that foundation two ways. Pacvue MCP opens it to the AI tools teams already use, including Claude, ChatGPT, and Copilot, through one secure endpoint into Pacvue data with access that follows each user’s existing Pacvue permissions. Teams keep control of their own prompts, agents, and decision workflows.
For teams that prefer an integrated experience, Pacvue Agent is built on Pacvue’s commerce-specific data and AI capabilities and works inside the platform to diagnose performance changes, surface opportunities, and move from recommendation to approval to execution in one place.
The Future of Commerce AI
AI models will keep improving. New assistants, retailer MCPs, and agent-building frameworks will make sophisticated commerce workflows easier to build every year, so every brand will have access to capable AI. The leaders will be the ones that connect their data, apply commerce-specific intelligence, coordinate decisions across teams, and execute with confidence.
The winning strategy is to build on a foundation that makes your strategy yours. Then to run them on a foundation that makes every AI tool you choose more valuable, so enterprises can spend less time managing the complexity behind AI and more time using it to drive growth.
