ABM With GTM Engineering: Account-Based Marketing Without a Marketing Team

Most ABM advice is written for someone who doesn't exist on your team. A marketer with a demand-gen budget, a seat in 6sense, and a colleague whose whole job is orchestration. You read it, you hit "align sales and marketing," and it stops being actionable because you're one person with a Clay tab open.

This is the other version. ABM run by the operator building the pipeline, not buying a platform to run it for them.

The classic playbook and the engineered one aim at the same outcome: a small set of high-fit accounts, worked deeply. They diverge on how you get there. Four things change. You select accounts by signal instead of by static fit. You get data through an enrichment waterfall instead of one tool. You do real per-account research at scale instead of capping it at fifteen minutes. And you engineer plays that fire off signals instead of scheduling a campaign calendar. Work through them in order.

Step 1: Select accounts by signal, not just firmographic fit

The classic list starts from ICP fit. Right industry, right size, right geo. That gives you a list of companies that match on paper and tells you nothing about whether any of them will buy this quarter.

Signal-first flips the order. You still care about fit, but you rank on what's happening right now. Who's hiring for a role your product makes unnecessary. Who just changed their tech stack, took funding, moved a leader into the seat that owns this problem. Those are timing signals, and timing is most of what separates a reply from silence.

This is not the same as buying an intent feed and calling it done. An intent tool tells you a company looked at a topic. A real signal tells you why now, specifically, for this account. One is a probability. The other is a reason to reach out today. Build your account selection on the reasons.

Step 2: Build the enrichment waterfall

One data provider always leaves gaps. Bad emails, stale titles, a missing firmographic field right where you needed it. You feel it most at scale, when the misses stack up into a list that's half unusable.

A waterfall fixes that by cascading sources. You try the first provider, and when it comes back empty on a field, you fall through to the second, then the third. Clay is the orchestration layer where this lives; a contact-level waterfall like FullEnrich handles the email and phone fallbacks underneath it. Each source covers the last one's blind spots.

1. Account-level enrichment first: firmographics, tech stack, the signal data from Step 1.
2. Contact-level waterfall next: find the right people on the buying committee, then cascade providers for verified email and direct dial.
3. Fallback logic on every field: never let one empty response kill a row that a second source could have filled.

The point isn't the specific tools. It's that you stop accepting one vendor's coverage as the ceiling.

Step 3: Run per-account research at scale

Here's where the classic lean-team advice gives up. The honest ones tell you to cap research at fifteen minutes an account. The rest just call it "time-intensive" and move on. Both are admitting the same thing: real research doesn't scale by hand, so they ration it.

It scales when you put an intelligence layer on it. Claude Code, in our case, doing the depth a strategist would do by hand, on every account in the program instead of the top five. And depth means more than firmographics. It means the account's business model, who actually sits on the buying committee, the competitive context they're operating in, and real evidence of the pain your product addresses. The research a good rep would kill for, run across the whole list instead of rationed across it.

That's the substitution. Not fifteen minutes per account. Not nothing. Strategist-depth research on all of them, because the research is engineered, not manual.

Step 4: Engineer the plays, don't schedule campaigns

A campaign calendar says "touch these accounts on Tuesday." A play says "when this account does X, and the research shows Y, do Z." The difference is the trigger. One fires on a date. The other fires on a reason.

That's what "engineered" means concretely. A play watches for a specific signal, checks it against a specific research finding, and only then acts. No fixed cadence, no spraying the same sequence at everyone on the same day. And the logic stays inspectable. You can open any play and see exactly why it fired for this account and not that one. That matters more than how many channels you're running or how many agents you've stacked. Inspectable logic over agent count, every time.

Step 5: Measure it like you own the pipeline

If you built the system, the borrowed metrics don't fit. MQLs and a platform's "account engagement score" were designed for a program you bought, not one you engineered. They measure activity against someone else's model.

Measure the things your own system actually controls. Signal-to-play conversion: of the accounts that tripped a signal, how many did a play fire on, and how many turned into a real conversation. Research-to-meeting yield: is the depth actually earning replies, or just costing compute. Cost and time per engaged account, so you know what the program costs to run at the account level. If you want an external benchmark to compare against, find a sourced one before you trust it. I'm not going to hand you a percentage I can't stand behind.

This ties back to how we sell, too. Prove-It-First means delivering a piece of the research before asking for anything, and that same account deep-dive is both the play and the proof.

The common ways operators break this

Doing it yourself is the point, and it's also where the predictable mistakes live. Treating the waterfall as one tool, so you're back to single-source coverage with extra steps. Skipping signal validation and trusting an intent feed as if it were a reason. Engineering elaborate plays before the research process underneath them actually works. Measuring an engineered program with a marketer's borrowed KPIs. Each one quietly turns the system back into the thing you were trying to replace.

Where to start

Pick ten accounts. Run them through the whole loop by hand once: signal, waterfall, deep research, one engineered play, honest measurement. You'll feel exactly which step is weakest for you, and that's the one to build first. The system is worth engineering only after the loop is worth repeating.

If you'd rather see it run before you build it, that's how we work anyway. We'll deliver a live account deep-dive first, then talk.