5 AI Prompts for Better ICP Targeting and Account Lists

Many marketers have a bad ICP problem. The kind where the “ideal customer profile” reads: “B2B SaaS company with SMB, mid-market, and enterprise segments.”

That definition is useless. It doesn’t guide targeting, drive pipeline, or filter out bad fits.

In B2B, an ideal customer profile (ICP) is the domino that tips everything else that follows. Sales efficiency, pipeline quality, deal size, and retention all trace back to how sharply you’ve defined, and how consistently you’ve targeted, the right customer. Get it wrong and every marketing dollar you spend compounds waste.

AI can help fix that, but not by inventing a better-looking ICP from scratch.

AI isn’t just for cranking out LinkedIn posts. It can be leveraged for ABM account research, enriching data, and scoring accounts so you know who is worth your time (and who’s not).

If you’ve read our foundational article on AI, ABM, and ICP strategy, this is your hands-on follow-up to apply the model.

In CXL’s course on ICP Modeling and Targeting with AI, Eric Linssen, Growth Lead at Keyplay, showed how SaaS teams use customer data and AI to build and test more precise targeting models – which this article pulls from.

TL;DR

• AI-powered ICP targeting starts with actual customer outcomes. Compare high-value, retained customers with poor-fit leads to identify the attributes that separate them.
• Build your ICP using firmographic, technographic, and situational signals. Industry and company size provide a starting point; sales motion, hiring activity, and technology requirements help narrow the account list.
• Use five AI prompts to build the targeting model: compare customers, enrich account records, score fit, identify negative signals, and translate findings into targeting decisions.
• Validate account scores before using them. Check whether high-scoring accounts actually convert, retain, and generate more revenue. A 0–100 score is only useful if the underlying model predicts something meaningful.
• Measure ICP performance through pipeline quality. Prioritize account progression, deal size, retention, and cost per qualified opportunity rather than the number of accounts or leads generated.

Stop pretending your ICP slide means something

If your ICP deck still says “SMB and enterprise” in the same sentence, you’re not segmenting. You’re guessing.

PartnerStack had that same problem. Their ICP looked something like this:

“B2B SaaS business that sells to businesses, sells a subscription, and runs or wants to run partner programs.”

Sounds fine. But, it was so broad it covered thousands of companies that would never buy.

Eric Linssen calls this the “Standard SaaS ICP.” It’s what happens when you confuse potential customers with ideal ones. 

“If you pull every company that fits that definition, a huge percentage will be unqualified. You’re wasting marketing spend, sales time, and customer success resources.”

— Eric Linssen

That’s the root of bad go-to-market. Talking to the wrong people, with the wrong message, for the wrong reasons.

A good ICP is not about who you could serve, but who you serve best. In other words, the customers who get real business value from your product and stick around.

That distinction changes everything.

“Doing ICP right is solving for net revenue retention at the targeting level.” — Tyler, CMO at PartnerStack

The cost of missing the right accounts starts long before a sales conversation. In 6sense’s 2025 B2B Buyer Experience Report, based on nearly 4,000 buyers, 95% of purchases went to a vendor already on the buyer’s Day One shortlist.

If you’re targeting everyone who could theoretically use your product, you’re spreading your budget across accounts that may never seriously consider buying. The work is figuring out which accounts are worth winning, and getting in front of them early.

The only ICP framework that doesn’t suck

Forget personas and surface-level firmographics. A functional ICP answers one question: Who gets the most value from your product, and why?

A useful ICP defines the companies most likely to get value from your product, convert into customers, and stick around. It combines basic company characteristics with the specific situations that make a business worth targeting.

Here is how the best teams define it.

Start by writing a vivid description; 

“Our offering is best for [type of company] in [specific buying situation]. They focus on [desired outcomes] and are struggling with [specific problems].”

Then add what Linssen calls “better if” signals. These are secondary traits that make a good account great.

Here’s an example from Keyplay’s ICP:

• Best for: B2B SaaS companies with 100 to 2,000 employees in the U.S. and Canada;
• Selling motion: Sales led, 10+ AEs, selling into multiple verticals;
• Better if: SDR team is growing; RevOps is in-house; they use Salesforce or HubSpot; they are planning territory optimization.

For a closer look at why broad definitions waste budget, see our breakdown of the ICP gap costing B2B marketers conversions. The problem isn’t always finding more companies. It’s being willing to rule some out.

Those “better if” conditions not only separate good fit accounts from dream accounts, but are also where ideal customer profile AI tools become indispensable.

Use AI for pattern recognition, not parroting

The lazy use of AI: “Write me 10 LinkedIn posts about SaaS marketing.”

The smart use: “Analyze my best and worst customers. Find the signals that predict fit.”

Most teams are already using AI somewhere in ABM. But using it and having a system are two different things.

In Demandbase and ForgeX’s 2025 research, 91% of surveyed B2B marketers said their companies had adopted AI to support ABM, yet only 19% had a formal AI roadmap. That’s a big gap between adding tools and knowing what they’re supposed to improve.

ICP modeling is a good place to close it. You’re using AI against real account data, with a measurable outcome: better targeting.

AI’s real power is pattern recognition at scale. It can research hundreds of companies, extract firmographic and behavioral data, and surface insights a human team would miss.

Linssen explained it clearly:

“AI is incredible at looking at an account and understanding signals that maybe your best AE would see. It can look at their website and say, ‘Yes, they sell to multiple markets,’ or, ‘They probably have an existing partner program.’”

That isn’t content automation. This is what using AI for account targeting and research looks like when applied by serious marketers.

5 prompts that turn AI into an ICP research assistant

You don’t need a massive data platform to start. But you do need customer data worth analyzing.

Export your closed-won customers, high-value retained accounts, closed-lost opportunities, and disqualified leads from your CRM. Include the business outcomes and account attributes you actually want to test. Remove unnecessary personal information and use an AI workspace approved for your customer data.

The five prompts below follow Linssen’s ICP modeling workflow, moving from customer analysis to account enrichment, scoring, and validation. They’re expanded templates based on that process, not verbatim prompts from the course.

Run them in sequence. Each output becomes the input for the next step.

And don’t let the model fill missing data with assumptions. That’s how you end up with a beautifully formatted version of the same bad ICP you started with.

Prompt 1: Identify contrast between best customers and disqualified leads

Ask AI to analyze patterns between your top customers and poor fit leads. Look for contrast, not averages. 

If both segments use Google Analytics, that tells you nothing. If disqualified leads skew to retail and winners skew to B2B SaaS, that’s a signal.

Copy-and-paste prompt:

You are a B2B revenue operations analyst helping me identify the characteristics that distinguish our best customers from poor-fit accounts.

I have uploaded a CRM dataset containing customer and opportunity records.

Business context:

• Product/service: [Describe your offering]
• Target market: [Industry, geography, segment]
• Successful customer definition: [For example, strong retention, high ACV, healthy margins]
• Poor-fit definition: [For example, disqualified leads, closed-lost opportunities, early churn]

Your task:

1. Compare successful customers with poor-fit accounts using the attributes available in the dataset.
2. Identify meaningful differences in industry, company size, business model, technology, sales motion, and other relevant characteristics.
3. Calculate the count and proportion of accounts with each characteristic in both groups.
4. Identify attributes that separate the groups, rather than characteristics common to everyone.
5. Flag missing data, small samples, and patterns that may be misleading.
6. Distinguish information available before targeting from information only known after a deal closes.

Output:

Create a comparison table showing each potential ICP signal, its frequency among successful and poor-fit accounts, the difference between groups, and any data limitations.

Then recommend the five strongest signals to investigate further.

Use only the supplied data. Do not invent missing information or present correlations as proof of causation.

Prompt 2: Enrich accounts with signal data

Have AI collect the kind of data that indicates fit: 

• Tech stack;
• Headcount growth;
• Number of AEs;
• SDR team size;
• Markets served; or 
• Product categories. 

Linssen’s team uses Keyplay to enrich CRM records with firmographic, technographic, and hiring signals. Similar workflows can use approved data-enrichment tools, CRM integrations, Apollo, LinkedIn Sales Navigator, or HubSpot’s data enrichment features, depending on what data and integrations you have access to.

The important part isn’t the vendor. It’s whether you can verify the signals and keep them current.

Copy-and-paste prompt:

You are a B2B account researcher. I want to enrich a list of companies using the ICP signals identified in my previous analysis.

Inputs:

• Target account list: [Upload company names and domains]
• ICP signals: [Insert findings from Prompt 1]
• Approved data sources: [Specify connected CRM, enrichment tools, or public sources]

Your task:

For each company, research the following attributes where verifiable:

• Industry and business model
• Employee range and hiring trends
• Sales motion and approximate sales team structure
• Relevant technologies
• Markets or customer segments served
• Evidence of the business situation that makes the company a potential fit

For every signal, record:

1. The observed value.
2. The source URL or connected data source.
3. The date the information was collected or last updated.
4. Whether the signal is verified, inferred, or unknown.

Output:

Create an enrichment table with one row per company, including its domain, relevant ICP signals, supporting sources, and missing fields.

Do not guess a company’s tech stack, hiring trends, or team size. If you cannot access an approved source, mark the field as unknown and explain what data would be needed.

Finally, identify which signals are reliable enough to use in account scoring.

Before activating the enriched list, run it through your CRM. Our opt-in-first ABM strategy explains how to identify existing contacts, permissions, and opportunities so you don’t pay to acquire accounts you’re already working.

Prompt 3: Score every account 0 to 100 based on fit

Once your signals are defined, use AI to assign a fit score (0–100) to each company in your CRM. The score should represent how closely each account matches your validated ICP criteria. It isn’t a conversion probability unless you’ve built and calibrated a predictive model against actual outcomes.

Copy-and-paste prompt

You are a B2B account-scoring analyst. Build a transparent ICP fit model using my historical customer analysis and enriched account data.

Inputs:

• Successful versus poor-fit account analysis: [Prompt 1 output]
• Enriched account dataset: [Prompt 2 output]
• Commercial objective: [For example, higher ACV, stronger retention, lower acquisition cost]

Your task:

1. Select five to seven ICP signals supported by the historical data.
2. Recommend a weight for each signal based on its observed ability to distinguish good-fit accounts from poor-fit accounts.
3. Make the weights total 100 points.
4. Define exactly how points are awarded for each signal.
5. Identify any business-level exclusion criteria separately from the scoring model.
6. Calculate a provisional 0–100 fit score for every account.
7. Flag accounts where missing or unreliable data prevents a trustworthy score.

Output:

First, provide a scoring rubric showing the signals, weights, rules, and evidence behind each choice.

Then produce an account table containing:

• Company name and domain
• ICP fit score
• Provisional fit tier (A/B/C/D)
• Strongest positive signals
• Missing or negative signals
• Evidence or data-quality warnings

Do not treat the score as a statistical probability of conversion.

Do not invent evidence or silently penalize missing information. Flag uncertain scores for review.

Finally, identify which weights and tier boundaries need to be tested against historical outcomes before the model is used for targeting.

Prompt 4: Flag negative signals that predict bad fits

Linssen’s team found that not all “good-fit” signals translate into real deals.

Some categories, like industries or tech stacks that match your target profile, still underperform.

Use AI to spot those patterns and mark them as negative signals: attributes that look promising on paper but correlate with low close rates.

These are the red flags that disqualify accounts before they waste sales time.

Copy-and-paste prompt

You are reviewing an ICP scoring model for false positives and overlooked poor-fit signals.

I have supplied historical account records, commercial outcomes, and the provisional scores generated in the previous step.

Your task:

1. Identify accounts that received high fit scores but were disqualified, lost, or produced poor customer outcomes.
2. Compare them with high-scoring accounts that converted and retained successfully.
3. Look for characteristics associated with underperformance, such as incompatible technology, unsuitable sales motion, unsupported geography, or poor product requirements.
4. Calculate the number of accounts and outcome rates associated with each potential negative signal.
5. Identify low-scoring accounts that performed well, so the model doesn’t exclude good customers.
6. Flag patterns based on insufficient data or outcomes that cannot reasonably be predicted before targeting.

Output:

Create a table containing:

• Potential negative signal
• Supporting account count
• Observed commercial outcome
• Evidence from the dataset
• Recommended action: exclude, reduce score, investigate, or retain
• Confidence and limitations

Then recommend which changes should be tested in the ICP model.

Do not automatically exclude an entire industry or segment based on a few failed deals. Separate observed patterns from explanations that still need validation.

Prompt 5: Identify and interpret the key signals

Have AI summarize the “why” behind those patterns, like why companies with certain tech stacks, headcount trends, or sales motions close faster or retain longer.

Copy-and-paste prompt

You are a B2B go-to-market strategist. Use the ICP research, enrichment, scoring, and negative-signal analysis completed in the previous four steps.

Your job is to explain which signals are useful for targeting decisions and how marketing and sales should act on them.

Your task:

1. Identify the strongest positive and negative ICP signals supported by the available evidence.
2. Explain the observed relationship between those signals and outcomes such as win rate, ACV, retention, or sales-cycle length.
3. Separate what the data demonstrates from possible explanations that need further research.
4. Recommend which accounts should receive high-touch ABM, lighter-touch campaigns, or lower-priority treatment.
5. Suggest how messaging should change for the most important account segments.
6. Identify what additional CRM data or customer interviews would strengthen the model.

Output:

Provide:

• A concise ICP definition.
• Five priority signals and their supporting evidence.
• The main negative signals.
• Suggested account tiers and recommended marketing actions.
• Three hypotheses to validate with sales or customer research.
• A list of assumptions, missing evidence, and model limitations.

Finish with a short summary explaining which accounts deserve more marketing investment and why.

Do not claim a characteristic causes higher revenue simply because it correlates with previous wins.

Use these insights to refine your ICP model and sharpen your targeting and messaging around what truly drives revenue.

(Image Source: Keyplay)

This is what prompting AI for ICP research should look like: an analytical process that finds which companies deserve attention.

The goal is not to outsource your ICP to AI. The goal is to make your thinking sharper and faster.

How do you validate an AI-generated ICP scoring model?

An AI-generated ICP score is a hypothesis until you’ve tested it against real outcomes. To validate the model, check whether high-scoring accounts actually convert, generate stronger deal values, and retain better than low-scoring accounts.

And don’t test it only against the same customer records you used to build it. That’s a good way to convince yourself the model works when all you’ve really done is teach it to recognize the training data.

Linssen’s approach starts by checking whether your strongest customers receive high scores and your poor-fit accounts receive low ones. That’s useful as an initial backtest. To make the model more reliable, take it further:

1. Check historical fit. Run successful and unsuccessful accounts through the model. Do the strongest customers consistently score higher?
2. Test accounts the model hasn’t seen. Hold out a separate set of historical accounts, ideally from a later period, and score them without using their outcomes to adjust the model.
3. Compare real business outcomes. Look at win rate, ACV, retention, and opportunity creation across the score bands. An A-tier account should perform meaningfully better than a C-tier account on the outcomes you’re optimizing for.
4. Investigate the misses. Look at high-scoring accounts that failed and low-scoring accounts that converted. Are important signals missing, incorrectly weighted, or unavailable when the targeting decision is made?
5. Keep validating. Once the model is live, monitor whether account tiers continue to predict useful outcomes. Update it when your product, market, or sales motion changes.

If the model can’t consistently separate valuable accounts from poor-fit ones, don’t use it to allocate budget yet.

(Image Source: Keyplay)

You’re not chasing theoretical perfection. You’re chasing predictive accuracy.

That kind of validation earns buy-in from skeptical sales teams because it proves the model correlates with real pipeline quality. 

Without testing, using AI for customer segmentation or to build a targeting model is just algorithmic guesswork dressed up as strategy.

What happens when you take ICP seriously

The results Linssen shared show what can happen when teams get more precise about who deserves their time and budget. They’re individual company examples, not a guarantee that changing your ICP will produce the same outcomes.

• Hone: Kept the marketing machine the same and changed who they targeted; ACV increased 2.3x;
• Mutiny: Doubled deal size and grew overall pipeline 35%;
• PartnerStack: Dropped their cost per dollar of pipeline created by 34% while increasing overall pipeline value by 58%;
• Airbase: Drove a 90% lift in outbound conversion by focusing outreach on the right accounts.

These are ICP clarity wins.

Linssen put it simply:

“Everything in your company is downstream of ICP. It’s a small investment, but it drives every major metric.”

Recap: The ICP to AI flywheel

If your pipeline metrics are weak, start here.

1. Define your ICP clearly

Write a vivid description of the companies you serve best. Then add the “better if” qualifiers and make it specific enough that you can score it.

(Image Source: Keyplay)

2. Translate the definition into a model

Once you’ve defined who you serve best, use AI to enrich your serviceable market with the signals you’ve identified and score accounts against those criteria. The goal is a prioritized list based on evidence, not a set of made-up probabilities about who will close.

(Image Source: Keyplay)

3. Test and improve continuously

A scoring model isn’t static. Run it on a cadence that matches your market; quarterly if you’re adding new segments, yearly if your ICP is stable.

Each cycle, validate it against real outcomes: do your best-fit customers still score A or B? Are disqualified leads still ranking low?

Refine signals and weights until your model mirrors reality instead of assumptions.

(Image Source: Keyplay)

4. Feed the insights into targeting

Use your fit scores to prioritize paid campaigns, outbound lists, and ABM account research.

An A-tier account might justify a personalized campaign and direct sales involvement. A B-tier account could receive more scalable, segment-based outreach. Lower-fit accounts may be better suited to lighter-touch nurture rather than expensive one-to-one campaigns.

The investment should match the opportunity.

For the next step, see how to turn ICP data into scalable AI-powered ABM campaigns, including one-to-few personalization and account-level activation.

5. Report impact by account quality, not just quantity

Don’t measure success by how many leads you generate. Instead, measure how effectively you’re moving target accounts through the funnel.

Track account penetration, progression, and conversion quality.

Replace vanity metrics like impressions or MQLs with indicators that show business impact: cost per qualified opportunity, deal size, and pipeline efficiency.

That’s the data sales and leadership will actually trust.

Frequently asked questions

How can AI help build an ideal customer profile?

AI can compare successful customers with poor-fit accounts, identify patterns in company attributes, enrich records, and apply scoring rules. The results need to be validated against actual customer and revenue data.

What’s the difference between ICP fit scoring and intent scoring?

ICP fit scoring measures how closely a company matches your ideal customer characteristics. Intent scoring estimates whether an account may be actively researching or considering a purchase. A strong-fit company isn’t necessarily ready to buy.

What data do you need for AI-powered ICP targeting?

Start with CRM records containing closed-won and lost opportunities, customer value, retention, firmographics, and relevant business attributes. Enrichment tools can add verified technology, hiring, and situational signals.

Can ChatGPT or Claude score accounts without an ABM platform?

Yes. With structured data and an explicit scoring rubric, an LLM can help build and apply a basic ICP fit model. However, automated enrichment, ongoing signal tracking, and reliable CRM integration require suitable data access and tools.

How do you know whether an AI-generated ICP model works?

Test the model against accounts it wasn’t built from. High-fit accounts should demonstrate stronger commercial outcomes than low-fit accounts. Compare opportunity creation, win rates, deal value, and retention before using the model to allocate substantial marketing spend.

What to do next

AI won’t save a bad targeting strategy. But once you know what makes a good customer, it can help you turn that knowledge into a system your team can actually use.

CXL’s ICP Modeling and Targeting with AI course, taught by Eric Linssen, walks through building an ICP model, enriching and scoring accounts, and connecting those scores to ABM campaigns and pipeline reporting. For the next step, AI for Account-Based Marketing covers building lookalike account lists, identifying buying signals, and turning account research into coordinated outreach.

Join CXL’s AI Native Marketing program

A good ICP gives you a clearer answer to who you’re targeting. The next challenge is making sure your AI workflows use that same context instead of starting from scratch every time.

CXL’s AI Native Marketing program helps marketers build working systems around their own data. The current program covers creating a shared marketing brain, a campaign engine with quality checks, and a reporting agent. The goal isn’t more prompts sitting in a document. It’s a system that turns customer information into work your team can use.