Why Marketers Must Completely Rethink Data for the Agentic Era

The era we just lived through was defined by a problem of human capacity. There was more data than any person could act on, so we summarized it, sampled it, picked the metrics that mattered, and built dashboards that surfaced the headlines. We talked about drowning in data because we had a hard ceiling on attention and a fixed constraint on time.

Agents don’t have those limits. They don’t get tired. They don’t lose focus at row 10,000. They don’t need the headline because they can hold the whole picture in working memory at once. Where a human analyst has to choose what to look at, an agent can evaluate everything everywhere all at once.

I’ve spent twenty years building data foundations for marketers, and the shift I’m watching now isn’t like the others. Dashboards, BI tools, attribution platforms. Each one was a better way to surface data for the people who had to read it. This one changes who the data is for.

An AI agent doesn’t run better on a pre-defined subset of your data. It runs best on all of your data, as long as it’s correct, clean, connected, and in context. All the campaigns, all the publishers, all the SKUs, all the metadata, in a single unified schema. The more context you give it, the better it performs. The headlines aren’t enough anymore. The full picture is the input.

That flips the entire question. For a decade, we asked how to compress data down to what a human could act on. Now we have to ask whether our data is in a shape an agent can actually use.

Most of it isn’t… yet.

Data for humans and data for agents are two different things

Four bars matter, and they’re worth taking in order. These come from watching what actually breaks when marketers try to put agents on top of their existing data stack.

Correct. Numbers reconcile, sources of truth are agreed on, the reports tie out. Most marketing organizations are here. In the agentic era, correctness comes with new stakes. A wrong number a human catches in a Monday meeting is a small problem. The same wrong number that an agent acts on at scale is a problem of a different category entirely. Bad data scales as quickly as good data when an agent is moving the levers.

Clean. Duplicates, orphans, leftover test rows, residue from a migration two years ago. A human analyst sees a duplicated row and ignores it. An agent sees two rows and reasons against both, and ten minutes later, it’s confidently telling you sales are double what they actually are. The cost of “good enough” data goes up sharply once something is acting on it without a human checking the work.

Connected. This is where it breaks for most companies. Your experienced teams carry the relationships in their heads. They know that “revenue” in this table is the same as “net sales” in that one, that this campaign maps to that SKU family, that the attribution window changed in March and the year-over-year comparison has to account for it. The agent doesn’t know any of that. It will guess. Sometimes the guess is right, but often it’s wrong, albeit confidently so. 

I’ve sat in plenty of these conversations where a smart team confidently shows me a unified view, and the agent test exposes the seams within minutes. An agent that reasons across media performance, sales, inventory, content health, and competitive signals simultaneously runs circles around one that sees only a single channel, but only if the connections between those datasets are made explicit rather than left to inference.

[in] Context. Every organization runs on tribal knowledge. The unwritten rules, the abbreviations, the reasons your category taxonomy looks the way it does, the carve-outs nobody documented because everyone just knew. This is what makes the difference between an assistant and an agent. A new employee, even one with deep industry experience, doesn’t really get up to speed for weeks or months, because that’s how long it takes to absorb the context. Agents work the same way. The tribal knowledge of the organization is what gives them the judgment to act on your behalf instead of just summarizing what they see.

If your data isn’t unified, it isn’t usable for AI. The marketers who’ll win the next five years aren’t the ones with the smartest individual analysts. They’re the ones who manage their data, and their data flows, well enough to feed their agents the whole picture.

This is an industry rethink.

Why MCP is the first place this shows up

Since we started building Skai’s data foundation in 2006, no feature has ever been requested more than MCP. By the time requests arrived in volume, the build was already well underway. We launched it in April.

MCP, the Model Context Protocol, is essentially an API for agents. User interfaces let humans talk to software. APIs let software talk to software. MCP lets agents talk to software. Same idea, new audience.

MCP became the most-requested feature in our history because of what the protocol signals. People click. Agents reason. Both need data, but in completely different ways. The agentic era will rewrite the foundational pillars of advertising from the ground up.

That rewrite starts with the connection between the agent and the data, which is exactly what MCP is.

What our MCP actually does

Right now, Skai’s MCP does one thing, and it does it really well: it gets your campaign data out of Skai and into the hands of any agent you want to put it in front of.

That sounds modest until you remember what that data actually is. Some of our clients are running over $100 million in Skai across paid search, social, and retail media. Beyond the clicks and conversions from the campaigns themselves, most marketers haven’t fully internalized that they’re sitting on a nine-figure proprietary data asset. Every keyword, every creative, every audience, every bid decision, every outcome – all of it normalized and tied together across hundreds of publishers. AI can do extraordinary things with that, if it can get to it.

MCP is how it gets to it. If APIs are pipes, an MCP is a toolbox. Skai’s MCP exposes the data, the metrics, and the dimensions that any Skai user has in a form that an agent can reason about. The agent decides which tool to use for the job, asks Skai for it in plain English, and gets a clean answer back.

No custom code. No developer build. Configuration in minutes.

Permissions aren’t bolted on; they’re inherited. When you connect your agent to Skai through MCP, that agent sees exactly what you can see. The governance model that has run our platform for two decades is the same one running our MCP.

This isn’t theoretical. Celeste, our purpose-built agent for commerce media, has been operating on this foundation for the past year. MCP extends that model from our agent to any agent. As Jason Taam, VP Data and Analytics at The Barcode Group, put it when we launched: “With MCP, our agents can directly access performance and AMC data in Skai without added complexity, enabling faster decisions and greater efficiency.”

A few of the most exciting ways our clients could put Skai’s MCP to work:

• Cross-business reasoning. Connect Skai data to an agent that also sees sales, supply chain, or finance data. Now one conversation can cover whether a campaign is performing, whether the inventory exists to scale it, and whether the margin justifies it.
• Unified ad data views. Pair Skai with the team’s other ad data sources within a single in-house agent. No rebuilding connections. One agent, one picture.
• Custom marketing agents. Skip the part where you rebuild integrations to 200-plus publishers. Start with the foundation, build your agent on top.

Picture a brand director asking her agent which campaigns are at risk of missing the quarter. The agent doesn’t hand back a list. It traces the slowdown to a competitor promo that spiked CPCs, flags where reallocation would recover the most spend, and asks if she wants it to draft the changes. That’s the shift.

That’s what MCP does today. The bigger question is what comes next.

The five surfaces every marketing organization has to rethink in the agentic era

MCP is one piece. Every marketing organization has data moving across five surfaces, and each one changes shape in the agentic era.

• In. How first-party, third-party, and publisher data gets into your unified foundation.
• Out. How that data gets to wherever decisions are being made.
• Inside. How data gets organized once it’s in, with metrics, dimensions, and views the team can actually use.
• Through. How signals feed the AI inside publisher platforms so their algorithms optimize for what matters to your business.
• Across. How your stack integrates with the broader analytics ecosystem.

Here’s what changes for each one.

“Out” in the agentic era. This is MCP, and it’s live today. Every other agentic data flow will follow the same pattern: open access to the customer’s reasoning system of choice, governed by the permissions already in place.

“In” in the agentic era. Data ingestion has always been a developer problem. Custom integrations, mapping spreadsheets, schema reconciliation, the kind of work that takes months and requires engineers, most marketing teams don’t have. Agents change that. An agent can read a new data source’s structure, propose how it should map into your foundation, and configure the ingestion in a conversation. The work that took a quarter starts taking an afternoon.

“Inside” in the agentic era. Custom metrics, dimensions, dashboards, and views have always been pre-built answers to predicted questions. In the agentic era, the way data gets organized becomes a conversation. The agent composes the view on demand. You ask, it assembles. Celeste already does the early version of this. The trajectory is clear: fewer pre-built dashboards, more questions answered against the same unified foundation, with the structure shaped to the question instead of the question forced to fit the structure.

“Through” in the agentic era. Skai already feeds first-party data into AI-native campaign products like Google’s AI Max and Amazon’s AI bidding tools. The next step is for agents to automatically orchestrate which signals flow where and when, with the right privacy controls and cadence. Our 13,000-plus preconnected data feeds become the substrate. Agents become the conductors. The better the data going in, the better the publisher AI performs coming out.

“Across” in the agentic era. Today, integrations are pipes between platforms. Tomorrow, they’re agents talking to agents. MCP to MCP. The role of a unified data layer is to be the thing other agents can rely on for a consistent, decision-ready view across publishers and retailers, no matter how many specialized agents are working alongside it. [GAP FOR GAL: one number or proof point that grounds Skai’s position in the broader agent ecosystem, parallel to the 13,000 feeds figure in “Through.” If nothing fits, leave as is.]

Conclusion: The urgent case for agent-ready data

It comes back to one thing [that is five things]: Correct, clean, connected, in context. If your data isn’t all four, AI will use it anyway, confidently, badly, at scale.

The marketers who win the next five years are the ones who fix this before their agents expose it. The question is whether your data is ready to be the input to an agent, not just the output of a report.

Skai is the AI-driven commerce media platform for performance advertising. For nearly two decades, the world’s top brands and agencies have trusted our award-winning technology to bring retail media, paid search, and paid social together into a single, strategic commerce media program. With embedded AI, connected data, and automation throughout, Skai helps marketers move faster, make smarter decisions, and drive more meaningful growth.

Curious what agent-ready data looks like in practice? Schedule a quick demo.

---

Frequently Asked Questions