Most target-account lists are political documents. A sales wishlist with a scoring model draped over it for dignity. A lookalike score from a vendor whose model nobody in the building can explain. Six logos the CEO would like on a slide. Everyone in the room knows it, nobody says it, and two quarters later the ABM program is "underperforming" when the truthful diagnosis is that the inputs were fiction from the first meeting.
An account list is a claim about the future: these companies, more than all the others, will buy from us, so we will spend disproportionately to reach them. A claim that expensive deserves evidence. Three layers of it.
Fit you can defend
Firmographics are the entry ticket, not the argument. Industry, size, geography, stack: necessary, nowhere near sufficient, and identical to the filter every competitor runs. The defensible version adds the job your product is hired for. For each account, name the job they would hire your product to do, and point to evidence the job exists there: the process that breaks at their scale, the regulation that just reached their sector, the system they have visibly outgrown. If the best available answer is "right size, right industry", you do not have a target account. You have a mailing list entry. Write the job sentence down next to the account name; if it takes a week to produce, that too is evidence.
A list you cannot defend line by line is a wishlist with a budget.
Motion you can observe
Fit says the job could exist. Motion says someone is working on it now. The bar for a motion signal is that it is observable, datable, and citable: you can say what happened, when, and where you saw it. Four kinds carry most of the weight:
- Hiring patterns: postings for the roles that precede your problem, in the function you sell to
- Technology changes: tools adopted or visibly ripped out, adjacent to what you do
- Community questions: their engineers or operators asking about the problem in public
- Leadership changes: a new owner of the function, with the mandate that usually follows
Weight recency while you are at it: one signal from last month outranks a stack of them from last year. A black-box "intent surge" from a data vendor meets none of those tests. It is a horoscope with an invoice. If a signal cannot be stated as a sentence with a date in it, it does not move an account up a tier.
The layer nobody checks
Now the overlooked one. Buying committees at those specific accounts are asking AI engines about your category right now, and the answers are not uniform. Ask for a shortlist as a mid-market fintech and as a global healthcare system and different names come back, with different caveats attached. Which means the answer layer is already tiering your list for you, whether you look or not.
So look. Run the questions their committee would run, framed the way their segment would frame them, and read what the engines say about the category, about you, and about the alternatives. If the answer layer already recommends you to a segment, tier those accounts up: your air cover exists before the first touch, and the committee's private research will keep confirming what your outbound claims. If it recommends competitors to precisely the accounts you care most about, that is not a mystery to be excavated in a loss review next year. It is a named deficit with a workplan: the comparison the engines cannot find, the community threads that have you wrong, the citations a competitor earned while you were polishing the deck. We wrote about the visibility problem in ABM is invisible to AI. Run it anyway; selection is where you make that invisible layer explicit, account by account, before the spend starts.
None of this requires exotic tooling. Sample per segment rather than per account: a dozen committee questions across the major engines is an afternoon of work, and it converts a tiering debate about feelings into a reading of the record.
The loop
Evidence decays, so selection cannot be an annual offsite exercise. Write the criteria down, on one page, plainly enough that a new hire could apply them without an interpreter. Revisit quarterly. And enforce the rule that makes the whole thing real: accounts leave the list when the evidence leaves. The hiring freeze landed, the champion moved on, the renewal went to the incumbent, the motion went quiet: off the list, resources re-tiered, no eulogy required.
That removal rule is the hard part, because in most organizations addition is a favor and removal is an argument, so lists only grow. But an ABM program is a concentration bet, and a concentration bet spread across every account anyone ever liked is not concentrated. It is just expensive. If your list has never shrunk, it is not an evidence-based list. It is a memorial to old enthusiasm.
The quarterly review is short when the criteria are written down: does the evidence still hold, has the answer layer shifted for this segment, who earned their way on, who lost their reason to stay. Four questions, one page, no politics surviving contact with any of them.
Pull up your current list, read the first ten names, and ask one question of each: what, exactly, is the evidence?