Most marketing teams adopted AI in the same way companies adopted email in 1998. Slowly, skeptically, and mostly for one thing. A subject line generator here. A brief summarizer there.
That phase is over.
The shift happening right now is structural. AI isn’t being dropped into individual tasks anymore. It’s being threaded through the entire marketing cycle, with different agents running different stages, passing outputs to each other, and reducing the number of places where a human has to stop and manually move something forward.
For a large enterprise, the scale of what’s possible is significant. For teams that haven’t started thinking in systems, the gap is starting to show.
What “Agent” actually means in a marketing context
There’s a lot of loose language around this word. An AI agent, for practical purposes, is a model that does a job, produces an output, and (in more advanced setups) triggers the next step in a workflow without waiting for a human to prompt it.
A chatbot that answers questions when asked is not an agent. A system that monitors campaign performance overnight, flags underperforming ad sets, generates a revised brief, and routes it to a media buyer’s inbox by 7 a.m. that’s closer to what we mean.
The distinction matters because teams that treat agents as fancy autocomplete won’t get much out of them. Teams that redesign workflows around them will.
Walking the full cycle: Unilever as the frame
Unilever is a useful example because of scale. The company runs hundreds of brands across more than 190 countries, with marketing operations that span local activations, global brand campaigns, retail partnerships, and direct digital channels. Most enterprise marketing orgs face a compressed version of the same problems: too much data, too many markets, too little time to act on any of it quickly.
Here’s how the agent stack maps across their full marketing cycle.
Stage 1: Market intelligence and demand sensing
Before a campaign brief gets written, someone has to answer: what’s happening, what do people want, and where are the gaps?
At Unilever’s scale, that question used to require weeks of research, multiple agencies, and a synthesis document that was already slightly stale by the time it landed on a desk.
An insights provider agent changes that timeline significantly. It runs on predefined KPIs and connects to external data sources: search trend APIs, social listening tools, retail sales feeds, and syndicated market research. It doesn’t just pull data. It produces a structured view of what’s shifted since the last reporting period and flags anomalies that warrant attention.
For Unilever’s home care division in Southeast Asia, this means a brand manager in Singapore can start Monday with an automated brief that covers share-of-search shifts across three categories, competitor pricing movement from the previous week, and a ranked list of topics gaining traction with their core audience. That brief used to take three days to compile manually.
A demand forecaster agent runs in parallel, pulling from both internal sell-through data and external signals like weather patterns, cultural calendars, and economic indicators. For a product like Domestos or Omo, seasonal demand spikes are predictable, but only if you’re combining the right variables. Agents don’t get tired of checking 40 variables at once.
Stage 2: Strategy and brief development
This is the stage most marketing teams haven’t automated yet, and for understandable reasons. Strategy feels like it requires judgment. It does. But a lot of what passes for “strategy work” at the beginning of a project is actually pattern-matching against what’s worked before, plus filling in a template.
An idea generator agent trained on past campaign briefs, brand guidelines and performance data can do a reasonable first pass at this. It won’t replace a strategist. It will give that strategist something to react to instead of a blank page, and research shows that’s faster.
At Unilever, this looks like an agent that ingests the insights brief from Stage 1 and produces three positioning options with supporting rationale for each, flagging which one has the strongest alignment with recent consumer sentiment and which one is most differentiated from what competitors have run in the past 18 months.
The strategist still decides. But they’re deciding faster, and with a documented audit trail of why each option was considered.
Stage 3: Content and creative production
This is where most companies started with AI, and where the most visible progress has happened. But individual content generation tools are not the same as a content agent operating within a workflow.
Unilever’s “People Data Centre,” a proprietary consumer intelligence platform built over several years, feeds creative agents with real-time information about what messages are landing with which audiences. The output isn’t just copy. It’s copy with a predicted resonance score attached, calibrated by market and channel.
What’s changed in 2025 and 2026 is less about generation quality and more about handoffs. A creative agent now produces social copy, a long-form article draft, and a 30-second script from the same brief, formatted for different platforms, and routes them to different approval queues automatically. A human reviews and approves. The agent doesn’t need to be asked to do the next step. It already knows what comes after.
That sounds minor. At a company running 400+ campaigns simultaneously across markets, it’s not minor at all.
Stage 4: Paid media and distribution
This is where sales assistant agents and media optimization tools do some of their most measurable work. Unilever’s marketing operations teams run programmatic spend across dozens of platforms in multiple currencies, each with its own auction dynamics, audience behaviors, and creative fatigue patterns.
A media agent doesn’t just automate bidding. It monitors creative performance at a granular level, identifies when an ad unit is burning out with a specific audience segment, and either pulls the creative or triggers a refresh request to the content agent upstream. The loop closes without a human having to spot the problem first.
For a brand like Dove running a real-time activation tied to a cultural moment, this responsiveness is the difference between catching a window and missing it. Campaign managers who used to spend 60% of their week on manual reporting are now spending that time on decisions that actually require a human.
Stage 5: Sales enablement and retail
For a company like Unilever that sells through retail partners rather than direct-to-consumer, the “sales” stage of the marketing cycle is really about equipping retail teams and category managers with the right information at the right time.
A sales assistant agent here prioritizes accounts that show early signals of churn or expansion, surfaces the most relevant case studies and promotional materials for a specific retailer conversation, and generates a customized pitch brief before a field rep walks into a meeting.
Unilever piloted exactly this kind of tool with its beauty and personal care division in Europe. The field team, which had been spending roughly three hours preparing for each retail buyer meeting, got that time down to under 45 minutes. The quality of the prep reportedly went up, not down.
Stage 6: Customer service and post-sale retention
Marketing doesn’t end at acquisition, though a lot of marketing teams act like it does.
A support assistant agent handles tier-one customer queries, routes complex issues to human agents with full context already attached, and identifies patterns in complaints that signal a product issue or a messaging problem. For Unilever’s direct-to-consumer brands like Hourglass or Paula’s Choice, this agent is also a retention tool. It can identify customers who’ve gone quiet, flag them for a re-engagement sequence, and surface the most relevant product recommendation based on purchase history.
This is where the marketing cycle genuinely closes. The data from customer interactions feeds back into Stage 1. What people are complaining about, what they’re asking for, what language they’re using about a product all of it becomes input for the next insight brief.
Teams that haven’t connected these stages yet are running their marketing cycle with a gap in the middle. They’re acquiring customers and then starting over from scratch each time, rather than getting smarter with every transaction.
The horizontal layer: Employee copilots
Running underneath all of this is something less visible but arguably more important for most companies in the short term: the enterprise-wide copilot layer.
Microsoft 365 Copilot, Google Workspace AI, and similar tools are already inside the daily workflow of most enterprise marketing teams. A brand manager at Unilever drafting a regional campaign plan is using Copilot to pull from previous briefs, summarize stakeholder feedback, and generate a first-draft timeline. That’s not a pilot program anymore. It’s standard operating procedure.
The organizations that are getting the most out of this layer are the ones that have built internal knowledge bases specifically to feed these tools. Generic copilots give generic output. A copilot trained on five years of Unilever’s brand guidelines, campaign post-mortems, and consumer research gives answers that are actually useful.
What’s actually slowing teams down
Two things, mostly.
The first is data fragmentation. Agents are only as good as the data they can access. Most large marketing organizations have customer data in one system, campaign performance data in another, sales data in a third, and almost no clean connection between them. Before any of this works properly, someone has to solve the plumbing. That’s an unglamorous project that takes time and usually requires buy-in from IT and finance. Teams that delay it will delay everything else.
The second is role confusion. When an agent is doing the research, the briefing, the copy drafts, and the performance monitoring, what exactly is the marketing manager doing? This is a real question, and it doesn’t have a comfortable answer yet. The teams handling it well are the ones having that conversation explicitly, rather than hoping nobody brings it up.
Where this goes next
The agent stack described here is largely available today. Most of the tools exist. The integrations are being built. What’s missing in most organizations is a decision to treat AI as infrastructure rather than a feature.
By 2027, the gap between companies that have built agent-connected marketing workflows and those still using AI as a collection of disconnected point tools will be significant and measurable. Not in vague “efficiency” terms. In campaign speed, cost per acquisition, and time from insight to market.
Unilever has a head start, partly because of resources and partly because they started early. But the architecture isn’t proprietary. The decisions are available to any marketing team willing to make them.
Want to audit your current marketing stack against this model? The next article in this series covers how to map your existing tools to the agent framework and identify the gaps that are costing you time.
Want your team ready before this shift hits? Reach out, and we’ll map the AI agents stack that fits your growth, budget, and timeline.