👋 Hi, it’s Kyle and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups.

As AI makes generic outreach cheap and ubiquitous, differentiation comes from precision, relevance, and deep context around a small set of high-value accounts. This sounds wonderful on paper until it’s time to set up a modern account-based GTM engine.

For help I turned to Dan Rosenthal, co-founder of Workflows.io, which builds go-to-market systems for B2B revenue growth. Dan and partner Fivos Aresti bootstrapped Workflows.io to $2.5M ARR in 10 months and have implemented dozens of modern ABM engines. Today Dan shares their extremely tactical ABM playbook. (This post may be cut-off by your email provider; read it in full on the Growth Unhinged website.)

Put simply, account-based marketing is an account-based GTM strategy that works backwards from a list of target accounts. While it has historically been expensive and highly manual, AI agents and data enrichment make it way more accessible.

We now recommend it to any B2B company with a ACV of $50k+ and a constrained addressable market (<20,000 qualified companies). For companies with a TAM of <20,000 companies, the relative low conversion rates of other GTM approaches (e.g. automated outbound, LinkedIn content) make them very hard to rely on in siloes.

I started Workflows.io to help companies build their GTM systems. We’ve personally worked with 250+ companies including dozens of ABM implementations. ABM is one of our most in-demand service lines, and I can confidently say that the technical implementation is still far harder than most expect. But the results are also far more impressive.

So I teamed up with Kyle to teach you how to build a modern, agent-compatible ABM system using HubSpot, Clay, and Claude. Every process will be rooted in real (but anonymized) examples from our clients. I’ll break it down into seven main components, which I recommend following step-by-step.

The primary goal here is to create a Target Account List (TAL) that captures 90%+ of a TAM. When we’re talking a TAM of <20k companies, incremental increases in coverage can have massive implications on revenue so don’t shortchange this.

First, we build a data-supported ICP model. It is surprising how many GTM teams we speak to that don’t have their ICP written down on an agreed-upon document. We’ll assume you have this in-place for the sake of this article, but here are some external resources that can help if you don’t:

Here’s an example of what our own ICP planning doc looks like:

We believe simpler is better when it comes to tiering models. Simple usually means more accuracy and clarity for the sales team. With agreed-upon ICP qualification and scoring parameters, you have everything you need to start pulling the actual lists.

There are four main types of data providers for creating the Target Account List. Generally, we find that smartly combining 3+ data sources massively increases our chances of reaching 90%+ TAM coverage.

The goal is to balance coverage and costs.

But the only way to maximize coverage beyond a certain point is to pull a higher volume of lists. And those lists naturally get broader, including more unqualified companies, but also more qualified companies that targeted searches won't capture. Which increases costs two-fold.

On one side, it’s the credits directly spent with the data providers to pull more data. But also more AI spend required to qualify the TAM list since we deploy AI research agents to qualify each company based on their website content. This is a necessary evil if you want to fully map out your TAM, because the filters / data quality of prospecting databases don't fully suffice (e.g. vertical SaaS on typical databases will be under their industry vertical instead of software development).

After pulling multiple lists, the next step is to merge + dedupe them. We used to do this in Clay, but we recently switched this workflow to Claude Code and find it far easier. We still end up in Clay afterwards, where we spend significant time back-testing and optimizing the centerpiece qualification prompt/research agents.

This powers the initial qualification we do on every company (usually industry and product fit), before enriching and qualifying further based on firmographics, technographics, and account fit signals.

Here is the framework we use to build our qualification prompts:

With a recent Series A SaaS company targeting B2B distributors, we scraped 70+ sources (mostly industry association directories) to maximize coverage. The Clay workbook got pretty hectic, but that's the price it takes.

The last step is to repeat the same process for contacts:

AI made creating a comprehensive account list much easier. It also made account research WAY easier to collect at scale.

Since we already cut down our initial list, our goal is to collect the additional data points that would be useful for (a) campaign personalization or (b) sales reps researching accounts without leaving the CRM.

For most of the companies we work with, that ends up being 30+ custom data points. Which usually means 30+ custom CRM properties that need to be configured.

This is why setting up ABM is resource-intensive on the RevOps side of things, but Claude Code and MCPs have also made it a lot easier.

We typically use a spreadsheet to map out every single data point with their official HubSpot names, as well as additional requirements e.g. overwrite rules. Here’s an example of what that looks like:

This work is all done before we build a single workflow, so we can accurately predict costs.

Doing a full TAM Map with account research can easily cost $2,000+ across data provider, Clay, and AI credits. But it’s a justifiable expense because you can now get extremely precise with the data points you collect.

These are real examples from clients, in order of simple to complex:

We deploy both web research agents (Claygent) and some of the 150+ data providers you can rent in Clay, since some data is not publicly available. In practice, building this requires a lot of prompt engineering and Clay configuration.

We also use the account research properties to set up account scoring directly in HubSpot. This way if a rep manually overwrites a value, the score can be automatically reassessed:

On top of the raw data, we typically apply some formulas or calculations data more actionable for HubSpot.

One important sub-component of account research are parent-child associations. When selling to PE-backed companies, you may have to sell through the parent company. We deploy web research agents to find this information and set up HubSpot associations to reflect those relationships.

All of this makes your CRM a more data-rich and pre-filtered prospecting database so your reps don’t ever need to login to Apollo or ZoomInfo themselves (you can just use their data through Clay).

This may seem like a lot of work for a one-time implementation, but this same workflow is converted to something that gives everlasting value: an automated CRM enrichment workflow.

Note: Everything up to this point can also be done in Claude Code with the same data providers, but the evergreen workflows coming after are more robust to build in Clay.

If we go through all the effort of setting up the target account infrastructure mentioned above, we want this same workflow to be applied to every record in the CRM. And every future CRM record.

Surprisingly, we find this workflow fairly straightforward to set up:

We typically set up an “Enriched by Clay” date field on both the company and contact objects, both for easier tracking and to power workflows that re-enroll records when the data freshness window passes (usually after one year).

Before we do any changes in the CRM, e.g. adding records or properties, we typically do a cleanup first. We recommend the following at the bare minimum: (a) hard domain and email duplicate removal, (b) basic property audit to prevent redundancies.

CRM hygiene can get deeper than we have space to explore in this article. In short, these three mediums should cover 99% of your use cases:

Once the CRM cleanup is done, the TAM/stakeholder map is uploaded, and custom enrichment data is automatically flowing, we typically recommend a few quality-of-life improvements to help activate this data:

Up to this point, I’ve just covered static data. For any data that needs to be monitored on an ongoing basis, we’ve found a dedicated signal tracking process to be the best approach.