Knowing Your Customer Better Than They Know Themselves

Most companies have an ICP document somewhere. It usually says something like "mid-market B2B SaaS companies, 50-500 employees, US-based, $5M-$50M revenue." It sits in a Google Doc that was written during a planning offsite. It informs nothing. Nobody references it when writing ad copy or building landing pages or deciding which prospects to pursue. It's furniture.

The reason it's useless is that it describes a company profile, not a customer. It tells you the size of the building but nothing about the person inside it, what work they're doing, what's broken about how they do it, or what would need to change for them to care about what you're selling.

I built the ICP and Brand Strategy skills to fix this. They produce artifacts that are specific enough to actually use, grounded in the positioning decisions from the previous two skills in the chain. This is where strategy starts to feel tangible.

The five anti-patterns that keep ICPs useless

Before I explain how the skill works, here's what it's designed to prevent. Emily Kramer at MKT1 (she built marketing from scratch at Asana, Carta, and Ticketfly) defines ICP as "who's really going to be successful with my product?" Not "who matches the targeting criteria." That distinction matters, because most ICPs fail for the same five reasons. These are the patterns I've seen kill marketing programs across dozens of engagements, and they're specifically what the skill forces you to confront.

Too broad. "We serve all SMBs." If your ICP is everyone, you have no ICP. Your messaging will be generic, your targeting will be scattered, and your cost per acquisition will be high because you're competing for attention with every other company that also "serves all SMBs."

Too many segments. "We have twelve ICPs." No you don't. You have twelve ideas about who might buy. You don't have the resources, the messaging, or the campaign infrastructure to serve twelve distinct audiences well. Three is usually the right number. Two is better. One is ideal for the first six months.

Built from demographics, not behavior. "Companies with 200+ employees in healthcare." That's a firmographic filter, not an ICP. It tells you nothing about whether those companies do the work your product supports. A 200-person healthcare company that doesn't run digital patient acquisition will never buy a digital patient acquisition tool, no matter how perfectly they match your firmographic criteria.

No negative personas. Most companies can describe who they want. Almost none can describe who they don't want. Who are the customers that churn fastest? Who takes up the most support resources for the least revenue? Who do you lose to competitors every time? Negative personas are how you stop wasting money acquiring customers who will never succeed with your product.

No prioritization. All ICPs treated equally. This is the most common and most expensive mistake. If you have three segments, one of them is your best segment. It converts fastest, retains longest, has the highest lifetime value, and is most referenceable. That segment should get 70% of your attention. Instead, most companies spread their budget evenly and wonder why nothing is working.

These anti-patterns aren't theoretical. They show up in real budgets. I've walked into companies spending $500,000 a month on paid media distributed evenly across four segments, three of which had a cost per acquisition 3x higher than the fourth. Reallocating budget toward the segment that was actually working would have doubled their acquisition volume on the same spend. The fix wasn't better ads. It was knowing which customers to pursue.

The ICP skill is designed to surface these issues before you spend the money.

Workflow-first segmentation

The skill's core design principle is that the workflow matters more than the firmographics. This comes directly from Bob Moesta and Clayton Christensen's Jobs to Be Done framework, developed at Harvard Business School and codified in Moesta's "Demand-Side Sales 101." Their core insight is that every purchase is fundamentally a switch: a customer fires an old solution and hires a new one. What matters is understanding the work they're doing and why they'd switch, not what demographic bucket they fall into.

Companies don't buy software because they match a demographic profile. They buy software because they do a specific kind of work, and the way they currently do that work has problems. If they don't do the workflow your product supports, they won't buy. No matter how perfectly they match your firmographic criteria.

The skill asks five questions about the workflow.

1. What specific activities does your product support (not outcomes, activities)?
2. Who performs this workflow, what's their actual title and team?
3. How often do they do it?
4. What triggers it?
5. What happens if they do it poorly or don't do it at all?

After capturing the workflow, it layers on three additional elements.

• The competitive alternative from positioning (what they use today, using April Dunford's "Obviously Awesome" methodology, which insists you start with alternatives, not features).
• The specific problems with that alternative (why they'd switch, through the demand-side lens Moesta developed).
• And only then, firmographic filters for targeting purposes (company size, industry, geography).

The result is an ICP statement that reads like this: "[Role] at [company type] who [workflow] using [current alternative] and struggles with [specific problem]." That's something a campaign manager can actually use. That's something an ad copywriter can write to. That's something a sales team can qualify against.

Synthetic customers: research before you spend

Here's where the system does something most companies never consider:

Before spending money on real customer research (surveys, interviews, focus groups), you can use the ICP artifacts to build synthetic customers and pressure-test your assumptions.

A synthetic customer is an AI-simulated conversation partner built from the ICP, persona cards, and positioning artifacts the system has already produced. You feed the model everything you know about your target customer, their workflow, their frustrations, their current alternatives, their buying triggers, and then you have a conversation with them. You test messaging. You test objections. You test value propositions. You ask questions you'd ask in a real customer interview. Kevin Indig has a great script to produce synthetic customers using these ICP artifacts.

This is not a replacement for talking to real humans. Real customer interviews are irreplaceable, and the research framework the skill produces is designed to guide those conversations. But synthetic customer research does two things that traditional research can't.

First, it lets you iterate on your positioning and messaging before you spend $10,000 on a research panel or waste a quarter on campaigns that don't resonate. You can test twenty different angles in an afternoon and go into real customer interviews with sharper hypotheses.

Second, it surfaces assumptions you didn't know you were making. When you simulate a conversation with your target customer and the synthetic persona pushes back on a claim you thought was obvious, that's a signal. It means real customers might push back too. It's not proof, but it's a flag worth investigating before you build your entire campaign strategy around an assumption that hasn't been validated.

The sequence is: build the ICP, simulate conversations, identify the riskiest assumptions, then design targeted real-world research to validate or invalidate them.

Brand Strategy: making the foundation usable

The ICP work feeds directly into Brand Strategy, the fourth skill in the chain. This is where personas, positioning, and competitive differentiation get turned into the message hierarchy that drives everything downstream. If the ICP skill is about understanding why customers switch (Moesta), and the positioning skill is about choosing which alternative to position against (Dunford), Brand Strategy is where Kramer's B2B content framework comes in: mapping messages to buyer stages, personas, and proof points so that every channel says the same thing in the right voice.

I built this skill to produce five outputs.

• A main message (the one-sentence anchor for all communications).
• Supporting arguments organized by buyer stage.
• Key messages mapped to specific personas.
• USP statements with proof points attached.
• And a voice framework that ensures consistency across channels.

The critical thing it does is force alignment. When your main message, your supporting arguments, your persona-specific messaging, and your proof points all derive from the same upstream positioning and ICP work, they're automatically consistent. You don't end up with a homepage that says one thing, ads that say another, and a sales deck that contradicts both.

I've seen companies spend six months and hundreds of thousands of dollars trying to "fix" their messaging by iterating on individual assets. The problem was never the assets. The problem was that there was no upstream foundation those assets could align to. They were all independently reasonable and collectively incoherent.

What to do?

Pull up your current ICP document and run it through the five anti-patterns. Is it too broad? Too many segments? Built from demographics instead of behavior? Missing negative personas? Treating all segments equally?

If it fails on even one of these, your downstream marketing is compromised. Not broken, necessarily, but operating at a fraction of its potential. Fixing the ICP is often the single highest-leverage change a marketing team can make, because it sharpens everything else: messaging, targeting, creative, channel selection, budget allocation.

I built the ICP and Brand Strategy skills into the Marketer in the Loop Skills MCP. They load your Discovery and Positioning artifacts automatically, enforce the workflow-first methodology, and produce artifacts that your Content Strategy and Design Systems can actually build from. The same frameworks that Dunford, Moesta, and Kramer spent decades developing, encoded into a repeatable process you can run on your own business.

Next week: how the system turns strategy into a content engine and design system, and why that's where most "AI marketing" falls apart.