AI agents may finally deliver the original promise of one-to-one marketing.
Back in the 1990s and 2000s, I was selling and managing CRM campaigns for brands like General Mills and Best Buy. We talked a lot about the promise of one-to-one marketing. In reality, most of what we delivered was relatively basic personalization.
In 1993, Don Peppers and Martha Rogers popularized the idea of one-to-one marketing, envisioning businesses that would develop individual customer relationships, learn from every interaction, and adapt their behavior accordingly.
It was a compelling vision. It also turned out to be incredibly difficult to execute at scale.
Over the following decades, we built increasingly sophisticated technology to approximate that vision. CRM systems, customer data platforms, marketing automation, behavioral triggers, and personalized journeys.
I’ve spent much of my career building and implementing these systems. They’re powerful, but there’s a fundamental limitation in how they work.
We personalized the content, but we still predetermined the conversation.
Marketers designed the campaigns, defined the audiences, established the rules, and mapped the possible journeys. Even sophisticated personalization generally selected among possibilities someone anticipated in advance.
We got remarkably good at making one-to-many marketing look like one-to-one marketing.
What changes when the campaign can think?
Recently, I’ve been experimenting with AI agents designed to handle prospecting, lead nurturing, and parts of the sales process.
The obvious opportunity is automation. Agents can research companies, identify prospects, write personalized messages, and manage follow-ups. But as I’ve started connecting these capabilities, something more interesting has emerged.
Imagine an agent researching a prospective customer, understanding their business, and developing a hypothesis about where a product might be relevant. It initiates a conversation based on that hypothesis, then adapts as it learns.
Perhaps the original assumption was wrong. Maybe the prospect reveals an entirely different problem, or the timing isn’t right. Each interaction changes what the agent knows and what it should do next.
Consider a company selling marketing software to businesses with large dealer networks. One prospect might struggle with decentralized marketing. Another might have challenges managing cooperative advertising funds. Rather than delivering variations of the same predetermined campaign, an agent can explore those distinct problems and adjust its approach based on the responses.
The next interaction isn’t simply step three of a nurture sequence. It’s a decision informed by what was actually learned.
Every prospect effectively has their own campaign, continuously evolving as the relationship develops.
A system that learns
Two capabilities I’m incorporating into these experiments make this particularly interesting.
The first is an evolving knowledge base. Agents use shared knowledge about the company, its products, and customers. But the knowledge isn’t static. New product capabilities can be introduced, and insights discovered through customer conversations can be validated and incorporated into what the agents know.
The second is hypothesis testing. Agents develop assumptions about customer problems and positioning, test different approaches, and use performance data to determine which hypotheses are supported by evidence.
Response rates are useful, but qualified conversations and conversions tell us much more. By comparing approaches and measuring meaningful outcomes, we can begin separating assumptions that sound compelling from those that actually resonate with customers.
These two capabilities create something powerful: individual learning and collective learning happening simultaneously.
An agent learns what matters to a particular prospect. Across many interactions, the broader system learns which problems, messages, and positioning are most effective. Validated discoveries from one relationship can improve future conversations with entirely different prospects.
This changes what marketers are designing.
Historically, we designed campaigns, audiences, content, schedules, and conversion paths. With AI agents, we increasingly design the systems that make those decisions, establishing objectives, providing knowledge, defining boundaries, and determining when human judgment is required.
Of course, making personalized outreach inexpensive could also produce enormous volumes of more convincing spam. The objective shouldn’t be more conversations. It should be better conversations, informed by genuine customer needs and supported by evidence.
I’m still early in these experiments, and there’s plenty to validate. But the potential shift is significant.
For decades, we’ve built marketing systems that execute campaigns designed in advance. Now we’re beginning to build systems that can learn what to do next, test whether it works, and adapt accordingly.
The campaign of one isn’t simply a campaign with one recipient. It’s a conversation that evolves with the relationship, supported by a system that learns from every interaction.