What is ABM 2.0? A Strategy to Identify B2B Buying Committee Members at the Individual Level with AI and Optimize Your Approach

ABM 2.0 is a marketing strategy that leverages AI to identify individuals involved in purchasing decisions at a personal level, and increases B2B deal acquisition rates by delivering approaches optimized for each person.

Traditional ABM (Account-Based Marketing) has targeted companies—accounts—as the unit of focus. In reality, however, purchasing decisions are made through collective deliberation among multiple stakeholders with different titles and interests. What makes ABM 2.0 new is its use of AI to sharpen resolution down to the individual level—identifying "who is involved in the decision"—and delivering messages that resonate with each person. This article is intended for marketers and sales professionals who want to effectively reach multiple stakeholders involved in purchasing decisions, and walks through the process step by step: from preparation to a three-step implementation framework, to common pitfalls.

First, let's establish what sets ABM 2.0 apart from traditional ABM. It's not just a new label—the unit of targeting and the role AI plays have changed fundamentally. We'll break this down from three perspectives.

Traditional ABM was an "account-level" approach: select the target companies, then direct advertising and content at those companies. But even when a message is sent to a company, it's ultimately read by individuals within it. An executive and a frontline employee will care about entirely different things when evaluating the same product.

Contact-level targeting brings the resolution down to those individuals inside the company. Even within the same organization, the message changes: for the executive who controls the budget, the focus is return on investment; for the department head leading the implementation, it's how workflows will change; for the end users, it's the operational burden of day-to-day use.

This distinction has a direct impact on how well messages land. Account-level targeting tends to produce "averaged messages with an unclear intended recipient," whereas individual-level targeting allows the message to align with each recipient's specific interests. AI is used to support this personalized delivery at a scale that would be impossible to manage manually.

B2B purchasing—especially high-value decisions—is never completed by a single person. The person proposing the solution, the person approving the budget, the people who will use it on the ground, the IT and security teams conducting their reviews—multiple stakeholders with different roles make the decision collectively. In English, this group is called the buying committee.

A common failure here is focusing exclusively on the first person you make contact with. No matter how enthusiastic that individual is, the deal won't move forward if they can't secure internal approval. Conversely, targeting only the decision-maker won't work either—without buy-in from the people who will actually use the solution, it risks going unused after implementation.

When ABM 2.0 talks about "identifying individuals," the goal is to capture a complete picture of everyone involved in the purchase. If you can understand who is involved and what each person is concerned about, you can move away from depending on one person's enthusiasm and instead deliver the right materials to each stakeholder to bring the group as a whole toward consensus.

The logic of understanding each purchasing stakeholder individually and engaging them accordingly has existed for some time—but it was never practical to execute manually. With several people per company and hundreds of target companies, the number of individuals to track can easily reach into the thousands.

What changed this is AI technology, including generative AI. It is now possible to support processes that were previously done by hand: estimating from data who is likely to be involved in a purchase, analyzing each person's interests, and even drafting messages tailored to each individual.

That said, AI does not make the decisions itself. What AI handles is solving the "problem of scale" that arises when dealing with a large number of individuals. Ultimately, the judgment of which messages to deliver and where human involvement is needed remains with the marketer and the sales team. Treating AI not as "a magic automation tool" but as "an instrument that enables a level of resolution impossible by hand" is the starting point for avoiding inflated expectations.

ABM 2.0 is not something that starts working the moment you deploy a tool. The accuracy of AI and the effectiveness of internal collaboration both depend on the quality of the groundwork laid beforehand. Before moving into the practical steps, let's confirm three prerequisites that need to be in place.

The accuracy of ABM 2.0 is determined by the quality and quantity of the data fed into it. Start by verifying whether your company's customer data—deal histories, inquiries, website behavior, past transactions—is organized and accumulated in your CRM or MA tools.

A common problem is data scattered across departmental spreadsheets or locked inside individuals' heads. In this state, the data cannot be handed off to AI, and you cannot even reach the starting point for analysis. Centralizing customer information and cleaning up duplicates and inconsistent formatting is unglamorous work, but it is an indispensable preparation.

In addition to your own data (first-party data), consider leveraging external data such as intent data that captures signals of purchase intent. However, before adding more external data, prioritize getting your own data in order. If the foundation remains broken and you simply add more tools, AI will only produce large volumes of low-accuracy analysis. Use the ability to integrate and connect all of this data as your criterion for selecting tools to adopt.

Next, revisit your definition of who your ideal customer is. The Ideal Customer Profile (ICP) is an articulation of the conditions—industry, company size, challenges faced, and so on—under which your product delivers the most value. If this is vague, you cannot instruct AI on "who to look for."

In ABM 2.0, in addition to a company-level ICP, you need to flesh out individual personas for each stakeholder involved in the purchase. What is the position of the decision-maker, the champion, the end user, and the evaluator? What does each want to achieve, and what are their concerns? By mapping out personas by role, you create the foundation for message design later on.

One important caution: do not build personas based on assumptions. Look back at the actual buying processes of your existing best customers, and observe who was involved, where the conversation moved forward, and where it stalled. Personas grounded in real data do more to improve the accuracy of AI-driven person identification and interest analysis than any other factor.

ABM 2.0 cannot be completed by the marketing department alone. It is often sales that ultimately reaches out to each individual buying stakeholder identified by AI. Without alignment between the two, even the most sophisticated analysis ends up unused in the field.

A common scenario is a disconnect where marketing believes it has "handed over leads," while sales feels it has received "a low-quality list." Because ABM 2.0 handles information down to the individual level, both departments need to align in advance on what criteria define someone as "a person worth approaching."

Concretely, this means deciding when a prospect's status warrants sales action, and how the results of outreach will be recorded and fed back to marketing. Creating a state where both departments are looking at the same people using AI analysis results as a shared foundation is a preparation that should come before any tool deployment.

Once the groundwork is in place, the first practical step is determining "who to approach." We will examine the process of identifying buying stakeholders from data through three distinct methods.

The starting point for identifying buying group members is combining two types of data. The first is your own first-party data—behavioral signals from individuals you already have touchpoints with, such as visits to your website, document downloads, and past inquiries. The second is intent data—external signals indicating what topics a given company is showing interest in.

First-party data alone will cause you to miss stakeholders you haven't yet engaged with. Intent data alone can tell you what a company is interested in, but not who specifically within it is active. Combining the two brings you closer to the resolution needed to determine: "This company appears to be evaluating a purchase, and this specific individual inside it is moving."

AI supports this combination. It connects related individuals within the same company from fragmented behavioral signals and estimates the contours of the buying group. However, since these estimates are not certain, it is safer to include a step where a human reviews the resulting candidates and removes any that are clearly off-target.

Once enough identifying signals have been gathered, the next step is translating them into an actionable contact list. Here, generative AI streamlines the work of organizing candidates from fragmented information and compiling them into a list with roles and affiliations filled in.

Traditionally, this process required staff to research and manually enter each entry one by one—a labor-intensive task. With AI, it becomes possible to identify individuals likely relevant to each target company and produce an initial draft list in a short amount of time. Human effort can then be focused on reviewing that draft and making prioritization decisions.

The key to improving accuracy is a feedback loop. By recording which contacts from the generated list actually led to sales opportunities and which did not, and feeding those outcomes back as input for the next generation cycle, list quality improves through ongoing use—rather than simply using the AI-generated list as-is. Since non-existent contacts may sometimes be included, verification before outreach is essential.

First-party data alone makes it difficult to see the full picture of buying group members. To supplement information on job titles, areas of responsibility, and organizational structure, it is useful to combine data obtained from business-oriented social networks and third-party data providers.

The key to integration is correctly linking the same individual across multiple sources. Variations in how names and company names are written, or changes in affiliation due to job changes, can cause the same person to be registered as different individuals, or different people to be mistakenly identified as the same person. Handling this carelessly can result in sending irrelevant messages to the wrong recipients.

When using external data, it is also important to verify that its acquisition and use comply with each service's terms of use and the personal data protection frameworks of the relevant jurisdictions. The more data you add, the richer your analysis becomes—but if the legitimacy of its sources and usage is not ensured, it can become a significant risk down the line. Before focusing on volume, prioritize the accuracy of data matching and the lawfulness of data handling.

Once you have a clear picture of who to approach, the next step is understanding each person individually. This section covers three AI-based methods for analyzing the role each buying group member plays in the purchase and what they care about.

Buying group members should not all be treated with equal weight. The person who makes the final decision, the influencer who champions adoption, the end user who will actually use the product, the person who evaluates technical and security requirements—depending on the role, the arguments that resonate and the timing that matters will differ.

AI helps estimate which role each individual is closest to, drawing on signals from job titles, past behavior, and patterns of internal involvement. For example, an individual who frequently views materials on return on investment is likely closer to the decision-making role, while someone who checks operational procedures is likely on the practitioner side—using behavioral tendencies to form an initial read on roles.

However, job titles and actual influence often do not align. Some people hold senior titles yet defer decisions to those on the ground, while others have modest titles but serve as the real driving force. AI-based classifications should be treated as initial hypotheses only, and it is essential to maintain the mindset of revisiting who the true key person is through actual interactions.

Once roles become clear, the next step is to understand what each member is currently interested in. Which pages they read, which materials they download, which email topics they respond to——these accumulated behaviors serve as clues that reflect a person's interests.

AI reads patterns from this behavioral history to infer topics of high interest. Someone who consistently reads content about cost reduction and someone who follows implementation case studies require different information to be delivered next. When content aligns with a person's interests, response rates tend to improve.

One important caution is to avoid drawing conclusions from a single action. An email opened by chance or browsing unrelated to their work can easily be mixed in. Only when multiple behaviors point in the same direction can an interest be considered reliable. The key to avoiding misdirected personalization is to treat interest as a trend over a given period, rather than as isolated data points.

When taking interest analysis further, matching on literal keywords alone has its limits. "Cost reduction" and "expense compression" are different words, but as interests they are nearly identical. To capture such paraphrasing, technologies such as semantic search and embeddings are used.

Embeddings work by converting text and interest themes into numerical vectors, making it possible to treat semantic similarity as a measurable distance. This allows interests that differ in expression but are close in meaning to be linked together. Multiple pieces of content read by a given member can be grouped at the semantic level, surfacing where the core of that person's interest lies.

In practice, this technology is used for purposes such as "grouping members with similar interests" or "finding content to recommend next to a given member." There is no need to understand the technical details, but knowing that there is a means to capture interests one level deeper than keyword matching will serve as useful input when selecting tools.

Once you understand who you are dealing with and what they are interested in, it is time to craft messages to deliver to each individual. Here we look at methods for producing contact-level personalized content at scale while maintaining quality.

At the heart of personalization is the individual optimization of content using LLMs. By providing a person's role, topics of interest, and past touchpoints as inputs, it becomes possible to produce tailored drafts of email copy, advertising messages, and proposal outlines——customized for each recipient.

Manually writing individualized copy for hundreds of people was never realistic. The limit was superficial personalization such as inserting a name into a template, and recipients could see right through it. LLMs make it feasible, within a practical workload, to prepare copy that reflects each recipient's context for as many recipients as needed.

That said, sending generated copy as-is should be avoided. Factual errors, tonal mismatches, and descriptions that do not fit the recipient's situation can all creep in. Letting AI handle the drafting while a human reviews the facts and refines the tone——this division of labor is the practical approach for achieving both volume and quality.

The quality of copy produced by an LLM varies greatly depending on how instructions are given. The practice of crafting instructions to elicit targeted quality and consistency is called prompt engineering. Even with the same model, vague instructions produce inconsistent output, while precise instructions produce stable results.

The key to maintaining consistency is to explicitly state the brand's voice and the rules to be followed within the instructions. Permitted expressions and prohibited words, sentence length, elements that must be included, assertions to avoid——embedding these guardrails into every instruction ensures that output of similar quality is produced regardless of who generates it. Sharing the instructions that produced good copy as a template helps stabilize output across the entire team.

It also helps to handle, within the instructions, the information that changes per recipient (such as role and interests) separately from the fixed guardrails (such as brand guidelines). Designing the instructions so that only the variable parts are swapped out preserves the flexibility needed for individual optimization while preventing consistency from breaking down.

There is more than one channel for reaching your audience. Email, web advertising, your own website, direct outreach from sales — you may be contacting the same person through multiple routes. When each channel sends a different message, recipients notice the inconsistency and feel a sense of disconnect.

The key to consistency is centrally managing "what stage of communication are we at" for each individual. Whether a member is in the early consideration phase or the comparison phase, the core of what you deliver should be aligned across every channel. The expression may vary by channel, but the central message should remain one.

One thing to watch carefully is frequency management. The more channels you add, the greater the risk of over-contacting the same person. Track the total number of touchpoints per individual across all channels, and keep it within a range that avoids coming across as pushy. The purpose of adding channels is not to accumulate more contact attempts — it is to reach people through the right channel at the right time, suited to their situation.

Finally, let us highlight two common pitfalls in ABM 2.0. The traps lie less in the technology itself and more in the quality of the underlying data and in the handling of over-personalization.

The most common stumbling block in ABM 2.0 is rushing to adopt AI techniques while underestimating the quality of the data being fed into them. Outdated contacts, duplicate entries, inconsistent formatting, missing fields — feed AI dirty data, and it will return low-accuracy analysis in kind.

Telling examples include continuing to craft messages for contacts who have already left a company, or tracking the same individual twice as if they were two different people. Because AI does not question the data it is given, errors in input become errors in output. In most cases where AI "feels inaccurate," the problem lies with the data, not the model.

The remedy is unglamorous but reliable. Build into your ABM 2.0 operations a regular practice of auditing outdated data, merging duplicates, and standardizing formatting. Rather than cleaning once and moving on, designate a responsible owner and set a recurring cadence to keep data fresh. Before thinking about what you want AI to learn, managing what you are feeding it is the foundation of accuracy.

Personalization becomes counterproductive when taken too far. A message that pinpoints a recipient's behavior in granular detail does not feel convenient — it feels unsettling, raising the question "how do they know all this?", and erodes trust. It is worth remembering that knowing someone deeply and making that knowledge conspicuous are two different things.

Legal considerations are also essential. For businesses operating in Thailand, the PDPA (Personal Data Protection Act) requires consent or a legitimate basis for the collection, use, and disclosure of personal data to third parties. Proceeding with personalization without first clarifying what data you may use, for what purpose, and to what extent creates compliance risk. It is necessary to review the applicable framework for each country and region you are targeting.

The practical guideline is straightforward. Define the scope of data you are permitted to collect and its intended purpose from the outset, and personalize within those boundaries. Limit yourself to providing information that is genuinely valuable to the recipient, and avoid making them feel monitored. Upholding both legal compliance and a level of familiarity that recipients find comfortable is what builds a relationship of lasting trust.