How to transform ICP from a stale document people ignore to an AI-driven system that continuously improves to drive more revenue quarter after quarter.

Every mature company has an ICP, but these definitions are often too broad, too shallow, and out of date. Even worse, they sit in a document somewhere instead of being built into the GTM engine. There’s no feedback loop as new leads, opportunities, deals, and customers are won and lost.

As a result, reps are hitting activities, leads are flowing in, and the pipeline looks full. But deals are slipping, cycles are running long, and customers are churning at increasing rates.

You know what the fix looks like: a more efficient, more profitable business, where focused efforts yield better results. But drafting yet another ICP definition doc, only to have it ignored and collecting dust, doesn’t feel like it will move the needle.

You need much more. You need a living ICP definition that actually drives execution and continuously improves as the market and product evolve.

You need an AI-driven system that sharpens itself every quarter.

Here’s how to create it!

Ready to Drive More Revenue for Your Org?

Apply for The First 90, an intensive GTM Ops program for newly appointed CROs and revenue leaders. See results tied to revenue in 3 months.

Learn More

1. Identify the “Best” Customers, Not the Biggest

• Identify the “best” customers, not just the biggest ones
• Who was easiest and cheapest to attract, win, retain, and grow?
• Who had the most pain and the best outcomes from your product?
• Who became raving fans and spread positive Word of Mouth everywhere?
• Which prospects and customers were hardest and most expensive to work?

Your best customers aren’t always the biggest. They’re the most profitable and the easiest to replicate. They’re cheaper to attract, retain, and grow, they have huge success on your platform, and they become raving evangelists.

Even better, there are thousands of other companies that look just like them. You can sell and market to them the same way, leverage your existing customer stories, and own the niche they’re in.

First identify who they are, and who they aren’t. This should give you a specific list of accounts to analyze, drawn from your customers, opportunities, leads, and prospects. For customers, look at revenues, costs to acquire and serve, and outcomes. Those outcomes include product usage, CSAT, NPS, renewals, churn, and expansion.

Then pick which accounts are the “best” customers. Compare them to your other customers, and to lost opportunities, leads, and prospects.

How to leverage AI: Feed AI your sales, marketing, CS, and financial data, and ask it to match the data and help you identify your best, most profitable customers. Do this once, then set up an ongoing feed via MCP to run the analysis going forward.

2. Go Beyond Standard Firmographics

• Revenue, headcount, geography, and industry aren’t enough.
• To identify your ideal customers, we need to be much more specific
• What criteria describes our best customers, those with the most pain?
• What data points differentiate our best customers from the worst ones?
• What about wins and losses? Renewals and churn? Expansion and stagnation?

Every ICP covers industry, revenue/headcount, and geography. Many also cover the tech they use. That might tell you who you could sell to, but it’s not enough to tell you who’s ideal. You need to go deeper and identify a few additional data points that truly differentiate the best from the rest.

One warning. A few is enough to matter. A dozen or two makes it too complicated, and no one will understand or trust it.

To help, here’s a list of +150 data points to help you find a few that are relevant for your market.

How to leverage AI: You haven’t enriched the data yet, so there’s not much to do here. But you can have AI run a first-pass analysis on the data you already have, to see which data points best differentiate the best from the rest. This is a one-time exercise, before you enrich the data.

3. Analyze Data, Enrich, Repeat

• Start with the data you have and a small set of accounts
• Run the initial analysis to look for data that differentiates
• Then expand by enriching the data with additional data points
• Enrich first w/ hard data in tools like ZoomInfo, then fuzzy data with AI
• Differentiate fit and timing signals, which describe the company vs. events
• Keep going until you’ve identified a few signals that really differentiate best vs rest

In a perfect world, we’d have all the data on all the accounts, dump it into AI, and out would come the perfect ICP. But we don’t have that data, and AI isn’t magic. We need to take this one step at a time.

Analyze with AI, Then Review Manually

Start by feeding AI the data you have and ask it to identify which data points best differentiate the best from the rest. Then spot check each data point to validate the conclusion.

Doing this manually sets the foundation. It ensures both the initial data and the conclusions are accurate. From there, you can bring in AI to analyze more data points. It’s important to maintain a human-in-the-loop to ensure that the conclusions are accurate before potentially diverting millions in sales and marketing spend to updated targets.

As you do this, you set the foundation for an always-on continuous improvement machine.

How to leverage AI: Use AI for the analysis. Feed in the data and train it to identify the signals that best differentiate the best accounts from the rest. This can then become an ongoing exercise.

Enrich with Hard Data Sources, then LLMs

As you enrich your data, it’s best to start with deterministic databases (ZoomInfo, Apollo, Clearbit) for structured criteria with a right answer (employee count, revenue, tech stack), and, if needed, chain several providers in a “waterfall” to match more accounts with more data points.

How to leverage AI: For judgment questions no database has a field for: “What are the risk factors for the business in the 10-K?” These fuzzy attributes can sometimes be our most differentiating ICP criteria. 

Differentiate Fit with Timing

Fit tells you what companies buy from you. Timing tells you when they’re ready. Define fit first, then look among fit accounts for the intent signals that tell you who’s most ready.

First-party intent (marketing engagement, site behavior, product usage) is often the most valuable signal. Next are timely triggers like new hires, fundraising and M&A, and new product and market launches. Lastly, third-party intent signals (like Bombora, G2, etc) are noisier but can still be valuable.

Use this to qualify out poor-fit accounts, nurture best-fit accounts with no intent, and prioritize best fit accounts with high intent. This is how you make your entire sales and marketing engine more efficient.

How to leverage AI: Build agents to continuously search for some of this data, especially timely triggers, and update the team and/or the Account Score.

4. Build an Account Scoring Model

To prioritize accounts, especially for sales outreach, you need an Account Score. This is a somewhat separate (and big) topic, but it’s worth noting. Say your sales team can only reach out to 5,000 accounts this year, and 100,000 fit your ICP. How do you pick who they reach out to, and who they reach out to first?

You need an account score that takes the criteria identified above and scores each account against it.

This is another area where AI can help, on the first pass and with continuous improvement, as you feed it more wins, losses, and churns. Like a lead score, an account score starts with a simple hypothesis, then gets tested and iterated continuously and indefinitely.

How to leverage AI: An agent runs the account scoring model. It aggregates multiple data sources and analyzes many data points across a large set of accounts, on a regular schedule.

5. Make ICP Operationally Actionable

This is, by far, the most important part of ICP: making it operational. All of it is a waste of time if it doesn’t translate into how you define territories, targets, qualified leads and opportunities, at-risk customers, and expansion opportunities.

That means you need an ongoing system to enrich your data and qualify every account, at every stage of the journey, in one shared system. That way, sales, marketing, and CS stay aligned on the right targets.

How to leverage AI: Use a combination of automation and AI to continuously enrich data, qualify accounts across GTM, unify the data, flag issues, and update scoring and prioritization.

6. Build a Continuously Improving Engine

ICP has to evolve alongside your market(s) and your product(s).

For established products simply update the ICP as those products and markets evolve. But established companies still launch new products, markets, and segments.

The system you want to build is one where your customers, opportunities, leads, and prospects are continuously enriched with the right data. That data gets analyzed on a regular basis to validate and improve your ICP.

How to leverage AI: With AI and automation, you can build a system that enriches your data in real time and analyzes it as often as you want. It flags issues and opportunities as you win and lose. You review those, then use them to make data -driven decisions and update the model.

The AI acts as an always-on orchestration and analysis machine. You provide the judgment and make the final calls on where and how to invest our focus in GTM.

Ready to Drive More Revenue for Your Org?

Apply for The First 90, an intensive GTM Ops program for newly appointed CROs and revenue leaders. See results tied to revenue in 3 months.

Learn More

ICP Data Points

The following is a long list of potential data points for brainstorming ways to define ICP. It’s very important to narrow this to the few data points that really matter and that you can actually track.

The right data points are different for every company and often very unique. One company could be targeting restaurants that have Michelin stars. (Though Michelin has a solid list of these.) Another could be targeting companies with resources in Ukraine. It’s impossible to list every example in one place and no database will even come close to having the data to cover them all.

What’s important is to think through what’s likely most relevant for your ICP and if, where, and how you can get the data to test and validate it, and then to use it in GTM going forward. Now, you could just pop into ZoomInfo and look at the filters they have available, but it’s worth asking (especially if you haven’t bought ZoomInfo) if those are the best data points and if ZoomInfo is the best source for them.

With all that said, here’s a list to get you thinking.

Download the ICP Data Points graphic here.

Get the plain format ICP Data Points list here.