Known Good Marketing
← Advisory

How I think about account-based marketing

ABM is the progression from an account to the six-to-ten humans on the buying committee. Everything else is targeted demand gen.

Most of what gets called ABM is targeted demand gen with an account list attached. That's not an insult. Targeted demand gen is useful. But the label matters, because it changes what you go build.

The whole point of real ABM is a progression. An account gets interested, you resolve that account down to the specific humans on the buying committee, and you hand sales a warm conversation with context attached. Going from a logo to the six-to-ten people who actually decide is the whole motion. Everything else is a delivery mechanism.

That progression is also where the economics live. What I'm building toward is one-to-one ABM with contact-level precision at the economics of one-to-many, and every design decision below comes out of that.

Fit first, then signals

Two ways to structure an ABM program, and only one of them scales.

The reactive version waits for an intent signal, then scrambles to figure out whether the account is even a good fit. Every trigger costs you evaluation time, and most triggers turn out to be noise.

Fit-first flips the order. You map fit across the whole addressable market before anything fires, so every account already has a tier and an assigned play. A signal from a Tier 1 account is an immediate call to action, and a signal from a non-ICP account is filtered noise instead of a meeting.

The practical difference is that your CRM stops being a pile of inbound leads and starts being a living map of the market.

Worth saying how you get to "fit," because most teams guess. Start from labeled outcomes instead of a hypothesized profile. Take your closed-won, closed-lost, healthy, and churned accounts, reconstruct what those companies looked like twelve to eighteen months before they bought, and hunt for the operational signals that separate good fit from bad.

Then run the backtest properly, because a closed-won-only pass will fool you. Read four cohorts as a Venn diagram. Closed won gives you the positives. Closed lost tells you which signals are necessary but not sufficient. Disqualified MQLs expose the false positives generating your noise, and sub-MQL leads show you what separates qualifying from stalling.

Backtests run this way have a habit of overturning live assumptions. Job posting signals hold up and turn out not to be tracked at all. Review-site spikes fail badly despite everyone treating them as strong intent, and content engagement that looks great at mid-market falls apart once you get to enterprise. Valid in a vendor's aggregate data and predictive in your cohort are two different claims.

Then hold back a portion of the data and re-test whatever survived on the sample it wasn't trained on. Signals that look clean on the training set and collapse on the holdout are coincidences, and they're extremely convincing coincidences.

Before any of that, verify the labels. A CRM calling a company a healthy customer when it has one employee and no website will invalidate the whole exercise.

Fit, relevance, engagement

The tiering model I use sorts every signal into one of three types. It's the most useful thing I've adopted in the last year.

Fit is static or slow-moving. Firmographics, tech stack, the shape of the company, whether they run the ad infrastructure your product assumes. It tells you whether an account belongs in your universe at all. Revisit quarterly.

Relevance decays. Job postings, funding, leadership changes, a new exec in your buyer's function, a competitor's pixel coming up on renewal. It tells you why now, and freshness is part of the value.

Engagement is real time. Pricing page visits, repeat sessions, ad clicks, content consumption. It says this specific person is aware and doing something about it.

The compound signal is what you actually build plays around, because individual signals are noisy by themselves. A single pricing page visit is barely information, but add fit plus two relevance signals plus engagement inside a sixty-day window and you're looking at a completely different animal.

So tier on signal density instead of a numeric score, and let the tier pick the motion:

Two mechanics that make this work in practice. Signals inside 60 days carry full weight, 60 to 90 days carry partial weight, and past 90 days they're background context and stop being a trigger. And a few combinations should skip the accumulation entirely and auto-promote to Tier 1, because waiting on a fifth signal there is just losing time. A new VP or Director of Demand Gen plus any engagement. A churn away from an incumbent ABM platform plus any engagement. Fresh funding plus an open demand gen role plus a website visit.

One structural argument I haven't settled. Most scoring models, mine included, add signals up, which means a disqualifier gets averaged away by a pile of positives. Jordan Crawford's case for multiplying instead of averaging is that a near-zero on any disqualifying axis should tank the whole score. My working answer is additive for accumulating fit and multiplicative gates for genuine disqualifiers. I'd call that a position I'm holding, not one I've proven.

Relevance signals expire, and most programs ignore the clock

This is the least discussed operational detail in ABM and it's where a lot of the value leaks out. Below are the windows I start from. They're reasoned priors out of my own signal taxonomy, not decay curves I've backtested, and I'd re-derive them against your data before I'd defend any specific number:

The point isn't the specific number. It's that these decay at wildly different rates, so treating them all as the same kind of alert means you're late on the ones that matter and early on the ones that don't.

Speed beats polish, by a lot

The finding that most changed my behavior is about timing.

Three or more meaningful touches inside 48 hours of an intent signal convert dramatically better than the same content a week later. In Edmondson's data, meeting acceptance ran 47% on the fast cadence against 11% on the delayed one. Identical content. Speed was the only variable.

The window exists because intent is contextual and competitive. When somebody reads your material they're in an active problem-solving frame, and that frame closes. Your competitors are watching similar signals, and whoever responds first frames the evaluation.

Most ABM programs optimize creative quality, content depth, and channel sophistication while underinvesting in response speed. That's backwards. A fast decent response beats a slow polished one most weeks.

The operational consequence is that the infrastructure has to exist before the signal fires. You can't improvise a 48-hour multi-touch response by hand for every Tier 1 account, and that's the actual reason to build any of this.

Orchestration, not campaigns

Campaigns have a start date, an end date, and a performance window. Accounts that don't engage inside that window fall through the cracks, and between campaigns your best accounts go dark.

Orchestration runs continuously. You define stages and tiers, pre-build plays for each combination, and accounts move between plays as behavior changes. The program doesn't end. Accounts progress or regress, and the plays follow them.

The other thing orchestration fixes is over-investment in volume. Running a dozen simultaneous campaigns across four hundred accounts and nine hundred contacts is impossible to do well. Cutting to three plays and fifty accounts doubled conversion inside eight weeks in Edmondson's example, and the mechanism isn't mysterious. Fewer things, done at a depth that registers.

Spend tiers the same way. One-to-many, one-to-few, one-to-one is a budget decision before it's a creative decision, and the mistake I see most is one-to-one effort at one-to-many list sizes, which gets you neither.

The alert isn't the deliverable

The most common way a good signal system fails is that it ends in a Slack message telling a rep something happened.

A Tier 1 trigger should produce a brief with three parts. What they did, with specific signals and dates. Why it matters, which is the context layer marketing is uniquely positioned to supply. And which play to run, pulled from a pre-built menu organized by signal combination and persona.

Marketing builds and maintains that menu. Sales runs it. If the rep has to invent the motion every time, you built a notification system with an ABM budget.

This is also where the automation boundary sits, and I hold it firmly. Automate the research, the enrichment, and the synthesis, so a rep opens a brief instead of a blank page, and then stop, because the narrative is where humans are still clearly better. AI can route a precise signal to the right person, and it can't weave the account narrative, because that takes relationship history it doesn't have. Hand it the writing and the sending and you've automated the one step humans are still clearly better at, which is the argument I made here.

People cite volume outbound as the counter-evidence, and the numbers there make my case instead of theirs. Reply rates on automated sequences sit around 3% now, the lowest they've ever been, at the exact moment we have more signal data than ever. That only works at supreme volume, which runs directly counter to the ABM model. One operator I talked to was getting three to four meetings a week from signal-based manual outbound after eight months of automated sequences produced nothing.

Multi-thread or the deal stalls

Average closed-won deals involve around six stakeholders. Gartner puts a complex purchase at six to ten decision-makers, most of whom do their research before they ever speak to a rep. Most ABM programs target one or two.

Single-threading is the most common reason deals stall in committee. If only your champion has context, they carry the internal sell alone. Legal raises a security question, the technical evaluator asks about integration, finance wants an ROI case, and your champion improvises. Improvised internal selling loses deals.

Expanding from two to six stakeholders per account moved deal velocity materially in the same dataset. The mechanism is boring and reliable. Each stakeholder shows up to the evaluation already holding relevant context, so fewer questions are new questions.

This is the argument for resolving to the named person on both sides. Contact-level targeting on the buyer side, one named rep owning the account through to close on the seller side. Stop at the account and you lose sales trust immediately, because you've handed them a logo and asked them to go find the humans.

What the ads can and can't do

Ads carry the account through the early stages and then hand off. Being honest about where that handoff happens saves a lot of arguing.

The metric moves with the stage. Awareness earns its keep on whether you cleared an impression threshold against the right people, direct response on click-through, and brand on dwell time. Grade all three on cost per demo and you'll kill the two that were working.

Further down the progression, ads stop driving the relationship and the rep takes over. At that point the job of the ad changes to buying committee awareness and social proof, warming the people around your champion so the internal sell gets easier. Contact-level impressions before an outbound touch aren't trying to earn a click at all, and measuring them like they are is a category error.

Build stage pages, not hundreds of personalized variants

The landing page side of ABM is where budgets go to die, usually by way of token personalization. Dropping the account name, logo, and industry onto a page is low-value, faintly creepy, and it forces dozens of hand-built variants that help nobody decide anything.

Build for where the account is instead of who they are. Five to twelve reusable stage pages: what is this, how does it compare including against the status quo, can I trust you, what does it cost, let me talk to someone. Map accounts to stages and route them. That collapses an unscalable variant problem into a small system you can actually maintain.

Then optimize those pages for consumption instead of conversion, using heatmaps and session recordings, because most of the buying journey happens before the conversion event and last-touch optimization shows up too late to act on. Put the buyer's internal business case on the page, so your champion can carry it into a room you're not in.

Per-account ad creative runs on the same principle. A fixed set of concepts, with the variance between accounts coming from that account's own signal evidence rather than from asking a model to be creative fresh each time. That's what makes one-to-one economics work at one-to-few effort.

What I'd look at first

If the answer to the timing question is measured in days instead of hours, start there. It's the cheapest large improvement on the list, and it doesn't require buying anything.